Horizon Quantum: The Compelling Quantum Software Layer & Expanding Integration and Opportunity

The Horizon Quantum Platform Thesis, the Technology, and the Case for Substantial Upside — 

August 23, 2026  

quantumtechintegration.blogspot.com

THE THESIS

Horizon Quantum is attempting something no other company in quantum computing is attempting: to own the layer that developers will write against as these machines continue to become useful and as they are integrated into the very fabric of the global technological landscape — and it has assembled a combination of capabilities that no competitor holds.

On the evidence assembled in this report, we believe they will strongly succeed — and succeed on a scale that is nowhere near what is currently reflected in their stock price. Section 24.7 sets out precisely why, and states plainly what would have to be true for that judgment to be wrong.

That layer, in quantum computing, does not yet have an owner. This is the rare fact this report is built on, because in every prior computing paradigm the same thing happened: the machines commoditized, and the durable economics migrated — permanently — to whoever owned the programming layer. IBM built the personal computer in 1981 holding every advantage, and licensed the operating system from a company of about forty people; within a decade the hardware was a commodity assembled by interchangeable vendors competing on price, and the economics of personal computing belonged to Microsoft. Mainframes, PCs, mobile, and now accelerated computing, where NVIDIA's position rests less on transistors than on two decades of software written against CUDA that nobody can practically rewrite.

By the time most paradigms are visible enough to invest in, that layer is already claimed and the window has closed. Quantum computing is in the unusual interval where the machines are advancing fast, enormous capital is flowing into hardware, and the position that historically captures the most durable value is still unoccupied — because the hardware vendors best placed to build it have the least reason to (a layer that works everywhere makes customers indifferent to their machines), and the software companies that want it cannot get at the problem (developing it requires owning quantum computers).

What Horizon holds that no competitor holds:

• Turing-complete programming languages — conditionals, indefinite loops, mid-circuit measurement, concurrent classical-quantum execution — where nearly every competitor produces static circuits fixed before execution. This matters beyond expressiveness: quantum error correction requires mid-circuit measurement with classical feedforward, so a static-circuit framework structurally cannot express the fault-tolerant computation the entire industry is building toward.
• A compiler that ingests existing classical code. c2q compiles C and C++ functions directly into partial quantum circuits, callable in one line — replacing hand-built reversible circuits that can require dozens to millions of operations. No competitor offers any path from a validated classical codebase into quantum execution.
• Owned, operating hardware in two physical modalities. Ember-1 (Rigetti superconducting, live in Singapore) and the contracted AQ256 (IonQ trapped-ion, Dublin, $35M). Pulse-level access and real-time execution without post-selection. No other software pure-play owns hardware in even one modality, let alone two.
• An issued US patent over the core mechanism — claims covering compilation from "a unified language, that is effectively a classical language, as opposed to a quantum language" through to hardware-specific gate code. The principal comparators are open-source, with no exclusionary position at all.
• Founders who originated the underlying science. Fitzsimons created the field of blind and verifiable quantum computation (FOCS 2009; Science 2012 with Nobel laureate Anton Zeilinger). Si-Hui Tan, MIT PhD under Seth Lloyd, is co-inventor on the patent. They have worked together since 2013.

Fifty-seven people are doing this while no incumbent contests the ground.

The investment case does not require Horizon to win a hardware race, pick the winning qubit, or out-engineer IBM on circuit optimization. It requires the industry to eventually need a programming layer above circuits — which is the explicit reason Fitzsimons says useful quantum computing means reinventing eighty years of computer science — and for Horizon to have built it first while the ground was open. That is a far wider set of futures in which this works than any single-modality hardware bet offers.

What is being priced at roughly $1.08 billion is not a software business with revenue. There is no revenue, no named customer, and no published pricing, and this report says so repeatedly and in detail. What is being priced is a claim on the abstraction layer of a computing paradigm, held by the team with the strongest technical provenance in the field, protected by an issued patent, before that layer has an owner.

Microsoft needed IBM to hand it the opportunity. Horizon bought the machines.

Update note

This report is built directly on primary documents: Horizon's Form 20-F, the Q1 and Q2 2026 Form 6-K financial statement exhibits, the Quantum Systems Agreement disclosure, Schedule 13D filings, Section 16 insider filings, the granted patent text, and published transcripts of the Q2 2026 earnings call and the Canaccord Genuity Growth Conference. Two figures deserve early emphasis because they are frequently misreported: the IonQ AQ256 system's $35 million cost is a disclosed related-party transaction rather than a press estimate, and IonQ's equity stake stood at 7.8% as of June 30, 2026, reported as 7.8% of total shares outstanding in the Q2 6-K and as 13.3% of Class A shares in an April 2026 Schedule 13D — the same holding measured against two different denominators, not a decline. As always, this research was conducted to challenge my own long position in the company. This is not investment advice.

Author's Note

I've now taken this report apart four times looking for reasons my thesis is wrong, and the company keeps clearing the very high bar I set for it. In five months as a public company, Horizon shipped Beryllium from preview to early access, brought Ember-1 online for outside users, locked in a $35 million hardware commitment with the field's leading trapped-ion vendor, and added a Fortune-500-caliber CFO to its board. None of that required optimistic interpretation to find — it's dated and disclosed in the filings.

The part of this story I find genuinely compelling is the one that's hardest to fake: Horizon is the only software-layer company in quantum computing that also operates its own hardware, across two fundamentally different modalities, by design rather than accident. That's a testbed in Singapore running today and a second one under contract for Dublin. Everything else in this report — the widening losses, the dilution, the zero revenue — is the real cost of building that position before the market rewards it. I'd rather back a team already operating hardware from two vendors than one still waiting to see which modality wins. 

What ensues in this report is a genuinely compelling story of Horizon because the narrative is continually challenged and despite how we come at Horizon, the opportunities and expected global impact the company will have on the global quantum ecomsystem are unbowed. 

That last word 'unbowed' - I think that sums up Horizon Quantum right now - even though it is early - the market is maturing quickly - and under Joe Fitzsimons' leadership no word fits Horizon better than 'unbowed'.

Table of Contents

1. Report-at-a-Glance 5

2. Evidence-Tier Legend 6

3. Twelve Questions, Answered 7

4. Executive Summary 8

5. The Problem: Why Quantum Software Is the Binding Constraint 8

6. Technology Platform: The Triple Alpha Stack 9

7. Hardware Co-Design: Ember-1 and the Testbed Strategy 12

8. Intellectual Property Position 14

9. Revenue Model and the Path to Monetization 14

10. The IonQ Relationship and the AQ256 / Dublin Path 15

11. Dublin: What Is Actually Unique About It 16

12. Multi-Modal Hardware Ecosystem 18

13. The Founder: Joe Fitzsimons and Why It Matters Here 20

14. Leadership and Recent Organizational Changes 22

15. Business Development and Ecosystem Engagement 24

16. Financial Statements: Balance Sheet 25

17. Financial Statements: Operations and Cash Flow 26

18. Capital Structure and Dilution 27

19. Ownership and Governance 28

20. Five Targeted Commercial Sectors 29

21. The Future Market: Where Value Accrues and What Horizon Captures 29

22. Chemistry and the Health Sciences: The Deepest Opportunity 32

23. The Competitive Field 35

24. The Platform Thesis: Why This Position Is Rare 40

25. The Bear Case, Answered 49

26. Valuation Context 52

27. Risks and Challenges 53

28. Honest Concessions 54

29. What Would Change This View 54

30. Signal Quality: What Counts as Evidence and What Doesn't 56

31. Conclusion 57

32. Source Appendix 59

1. Report-at-a-Glance

Item	Detail
Ticker / Exchange	Nasdaq: HQ (Class A ordinary shares); warrants Nasdaq: HQWWW
Shares outstanding (Jun 30, 2026)	34,227,495 Class A + 19,744,585 Class B = 53,972,080 total (per Q2 6-K balance sheet)
Implied market cap (Aug 21, 2026)	≈$682M using publicly traded Class A shares only (34,227,495 × $19.93), or ≈$1.08B using total shares incl. Class B (53,972,080 × $19.93) — both calculated from the Q2 6-K share count; the latter is the convention data vendors apply. Now at or above both published analyst targets
Revenue status	$0 revenue in Q2 2026 and in H1 2026; FY2025 revenue was $38,873. Pre-revenue by stated strategic choice — management declines to book services revenue ahead of demonstrated quantum advantage (Section 9)
Inbound demand (20-F, Mar 2026)	Early-access interest from 40+ major corporations, 80 universities, 10 quantum software companies, and 15 national labs / government agencies — interest the company has deliberately not yet converted
Cash position	$113.3M at June 30, 2026 (up from $96.6M at Q1); H1 2026 operating cash used was $9.3M, investing cash used $5.5M (mostly testbed/facility capex)
Q2 2026 operating loss / Adj. EBITDA loss	$7.2M / $5.5M (vs. $2.7M / $2.1M in Q2 2025)
Q2 2026 GAAP net loss	$115.2M ($2.20/sh), including a $108.3M non-cash warrant-liability remeasurement
Total stockholders' equity (Jun 30, 2026)	$43.7M (vs. $83.9M at Mar 31, 2026 — the sequential decline is the warrant remeasurement flowing through accumulated deficit)
AQ256 system	$35M Quantum Systems Agreement with IonQ, dated March 31, 2026 (confirmed related-party transaction, Q2 6-K Note 13); installation site confirmed as Dublin, Ireland per Horizon's own June 11, 2026 press release; targeted 2027
IonQ equity stake	4,230,118 Class A shares — 7.8% of total shares (Class A + Class B) per the Q2 6-K, or 13.3% of Class A alone per the April 2026 Schedule 13D. Same position, two denominators; SEC filings report percentage of class. IonQ participated in the PIPE and was not diluted by it; only warrant exercises have diluted it, by under 5%
Largest disclosed holders	Joseph Fitzsimons (CEO): 65.0% of voting power via Class B shares (Mar 19, 2026 basis); Peak XV–affiliated funds: 20.3% of Class A per a July 9, 2026 filing on a stale share basis — recalculates to ≈18.9% (current Class A only) or ≈12.0% (current combined Class A+B), depending on denominator, if their share count is unchanged; no fresher filing confirms which
Warrant overhang	765,554 public warrants remain outstanding (exercise price $11.50) after ~79% of the original public pool was exercised; separately, 2,884,660 private placement warrants from the SPAC remain outstanding, none exercised. Also 6,341,712 employee options (weighted-avg exercise $3.36) and 474,784 RSUs
Analyst coverage	Needham: Buy, $20 PT (Jun 3, 2026). Craig-Hallum: Buy, $21 PT (Aug 17, 2026). Wall Street Zen: Strong Sell (Aug 8). Weiss Ratings: sell (e+) (Jun 17). At the Aug 21 close of $19.93 the stock has essentially reached both Buy-side targets
Industry engagement	Anchor sponsor of QIP 2027 (Singapore, Feb 2027); technical presentation at Q2B Tokyo with Quantum Machines; Q2 S&M expense growth explicitly tied by the company to trade-show/industry activity

2. Evidence-Tier Legend

Every substantive claim in this report is tagged by evidentiary basis, consistent with the methodology used across the Quantum Technology Integration Series.

• FACT — Directly sourced to a primary document: SEC/EDGAR filing (10-Q/20-F/6-K exhibits, Schedule 13D), company press release, or a recorded management statement.
• INFER — A reasonable inference or calculation from disclosed facts (e.g., a computed market cap, a dilution comparison), not itself a company-stated figure.
• ARG — An analytical judgment or interpretive argument — my view of what the facts mean, clearly separated from the facts themselves.
• UNDISC — A claim the company has not disclosed and that remains speculative; flagged explicitly rather than presented as established.

3. Twelve Questions, Answered

Before the analysis begins, the twelve questions a reader is most likely to bring to this company — answered directly, each pointing to the section carrying the underlying evidence and sourcing.

Question	Short answer	See
What does Horizon Quantum actually build?	A quantum software stack (Triple Alpha IDE/compiler; Beryllium, Helium, Hydrogen languages) plus its own in-house operated hardware testbed — it is not a hardware seller.	Section 6
Which quantum hardware companies does Horizon work with?	Deepest tier: Rigetti (Ember-1, in-house) and IonQ (AQ256, under contract). Integration tier: AQT and Alice & Bob. 30+ backends total, incl. named IBM, IonQ, IQM, OQC, Rigetti, AWS systems.	Section 12
Does Horizon have any revenue?	No commercial-scale revenue. $0 in Q1 and Q2 2026; FY2025 revenue was $38,873.	Section 17 / Report-at-a-Glance
What is the IonQ relationship, and what is the AQ256?	A $35M related-party contract (Mar 31, 2026) for a 256-qubit trapped-ion system, to be installed at Horizon's Dublin, Ireland facility, targeted for 2027.	Sections 10–11
Who owns Horizon Quantum?	Fitzsimons holds 65.0% of voting power via Class B shares. IonQ's stake has fallen to 7.8%. Peak XV discloses 20.3% on a stale basis (recalculates to ≈12–19% today).	Section 19
How much cash does Horizon have, and how long does it last?	$113.3M as of June 30, 2026; roughly six years of runway at the current six-month operating burn rate, before AQ256/Dublin capex likely accelerates that burn.	Section 17 / Report-at-a-Glance
Who leads the company, and who's new?	Founder/CEO Dr. Joe Fitzsimons; CFO Gregory Gould (since Aug 2025, pre-dating the de-SPAC); CLO Catherine Fitzsimons (May 2026); Peter Oey (Grab CFO) joined the board Apr 2026. Jill Turner, Harry You and Danielle Lambert were on the public-company board from day one. Amanda Chew promoted internally to CPO, effective Aug 17, 2026.	Sections 13–14
What are the biggest risks?	Widening losses, zero revenue, a dilution overhang, a 2027 (not near-term) AQ256 timeline, and open-source competitors (Qiskit, PennyLane) that are free.	Sections 25 and 27
How does Horizon compare to Qiskit or PennyLane?	Multi-vendor hardware access and in-house hardware differentiate it; Fitzsimons also named Classiq as a comparator. Qiskit/PennyLane are free, a structural risk.	Section 23
What is Horizon Quantum worth?	≈$682M (Class A only) to ≈$1.08B (all shares) at the $19.93 Aug 21 close — now at both published analyst targets (Needham $20, Craig-Hallum $21).	Section 26
What is Horizon doing to build industry visibility?	Anchor sponsor of QIP 2027 (quantum theory's flagship conference); technical presentations at Q2B Tokyo; this activity shows up directly in the company's own disclosed sales & marketing expense growth.	Section 15
What would change this thesis, for better or worse?	A named commercial contract, an updated Peak XV filing, or Beryllium reaching general availability would strengthen it; widening losses or a slipped Dublin timeline would weaken it.	Sections 28–30

4. Executive Summary

The one-sentence version of this report: Horizon Quantum is the only publicly traded pure-play quantum software company that operates its own hardware across multiple modalities by design, which positions it to matter regardless of which physical architecture eventually wins — and it is deliberately declining revenue today in order to hold that position. Everything else here is an attempt to test that claim against the primary record and to be precise about what it does and does not yet prove. [ARG] 

This report is grounded in Horizon's Q1 and Q2 2026 Form 6-K financial statement exhibits and the Form 20-F rather than in press-release summaries or data aggregators. Two findings deserve emphasis at the outset. First, the AQ256 system's $35 million price is not a trade-press estimate — it is disclosed as a related-party transaction in Horizon's own financial statement notes, dated March 31, 2026. Second, and more consequentially for the ownership narrative, the widely repeated claim that IonQ's stake "fell from 13.3% to 7.8%" is not correct. Those are the same position expressed against different denominators: SEC filings report percentage of class, so 13.3% measures IonQ's Class A shares against Class A only, while the 7.8% in Horizon's Q2 6-K measures the same shares against total Class A plus Class B. IonQ participated in the PIPE and was therefore not diluted by it. The only genuine dilution since Q1 comes from warrant exercises — approximately 2.5 million warrants, under 5% of outstanding shares.

On the operating side, Horizon remains a pre-revenue, cash-funded software company whose reported losses are widening in real terms (operating loss, adjusted EBITDA loss) even after backing out the noisy non-cash warrant-remeasurement swings. The balance sheet is solid for the near term — $113.3M in cash against roughly $9.3M in six-month operating cash burn — but the AQ256/Dublin buildout and continued headcount growth will likely accelerate that burn rate before any offsetting revenue appears. [ARG] 

What the primary filings also show is a company executing a coherent and unusually disciplined plan. The $0 revenue line is a stated strategic choice rather than absent demand: the 20-F documents inbound early-access interest from 40-plus major corporations, 80 universities, and 15 national labs and government agencies, and management has deliberately declined to convert that interest into consulting-style revenue that would misrepresent the platform's eventual value. Meanwhile Beryllium reached early access, Ember-1 opened to outside users with a real technical differentiator, the AQ256 converted from announcement to signed contract with a named Dublin site, a Chief Legal Officer was appointed and a Chief Product Officer promoted from within, and two fundamental research houses initiated at Buy. That is a substantial amount of verifiable execution for five months as a public company. [ARG] 

The sections below include full balance sheet, income statement, and cash flow detail drawn directly from the underlying financial statements, a complete ownership and dilution analysis, and a valuation section with peer market-cap context. Charts are built from the company's own reported quarterly figures.

The report also draws on the full transcript of Canaccord Genuity's 46th Annual Growth Conference (August 11, 2026) — the most recent primary management commentary available as of this writing — which supplies the rationale for remaining pre-revenue, the internal applications team, hardware-modality trade-offs, and management's hedged claim about the AQ256's chemistry performance. [FACT] 

5. The Problem: Why Quantum Software Is the Binding Constraint

Horizon's framing of the market problem, in the company's own words: hardware is only half the picture, classical code does not run on quantum machines, and only a few hundred specialists worldwide currently have the skills to formulate quantum algorithms. The question the company poses is not whether useful quantum computers will exist, but who will be able to program them when they do. [FACT] 

Fitzsimons put the scale of the task directly at the Canaccord Genuity conference: most quantum programs today are "little more than circuits — a sequence of operations followed by measurement," and making quantum computing broadly useful will require reinventing roughly eighty years of computer science and software engineering, including operating system kernels, dynamic memory allocation, input/output, and network communications. [FACT] 

This is the single most important framing in the report. If quantum hardware continues to improve on anything like its current trajectory, the binding constraint on commercial value shifts from qubit counts to the supply of people who can write programs for the machines. A hardware vendor's addressable market is capped by the number of teams able to use its product. That constraint is exactly what Horizon is attempting to remove — and it is a software problem, not a physics problem.

"The question facing the industry is no longer whether useful quantum computers will exist — but who will program them when they do."  — Horizon Quantum, company mission statement, horizonquantum.com [FACT]

The specific technical limitation Horizon targets: most quantum programming frameworks restrict developers to static circuits — a fixed sequence of gates fixed before execution begins. That constraint limits expressivity and makes algorithm design difficult, because a static circuit cannot branch on a result computed halfway through, loop an indeterminate number of times, or call a subroutine whose runtime is not known in advance. [FACT] 

6. Technology Platform: The Triple Alpha Stack

Triple Alpha is Horizon's web-based IDE and compiler suite. Its architecture is a four-level abstraction ladder, each level a complete programming language in its own right, with an optimising compiler translating downward through the stack to hardware-specific executables.

Level	What it is	What it buys the developer	Status
Beryllium	Object-oriented quantum programming language	Classes, inheritance, data structures, reusable libraries; operate on data rather than on its quantum representation. Enables code reuse and library ecosystems — the substrate for network effects	Early access since end of Q2 2026
Helium	BASIC-like imperative language; Turing-complete	Conditionals, definite and indefinite loops, mid-circuit measurement, concurrent classical-quantum computation, C/C++ subroutines, external I/O	Shipping in Triple Alpha
Hydrogen	Architecture-agnostic assembly-level language; Turing-complete	User-defined instruction sets; control flow as flow-chart blocks with conditional jumps; execution infrastructure supplies branch information at runtime so programs respond dynamically	Shipping in Triple Alpha
Algorithm synthesis (4th layer)	Automatic construction of quantum-accelerated algorithms from ordinary classical programs	The end state: write C or Python, receive a quantum-accelerated application. Goes beyond translation — generates implementations that exploit quantum speedups where they exist	Under development; company states it is 'exploring techniques'

Turing-completeness is the load-bearing technical claim. Helium and Hydrogen are both Turing-complete, meaning they can express algorithms of arbitrary complexity — including algorithms whose runtime cannot be determined before execution. This is precisely what static-circuit frameworks cannot do. [FACT] 

In plain terms: a static-circuit framework is closer to a fixed pipeline than a programming language. Horizon is arguing that quantum computing cannot become a general-purpose industry on that foundation any more than classical computing could have on punch cards, and it has built actual languages — with loops, branches, subroutines, and I/O — instead of a circuit-description library. Whether the industry needs that yet is debatable; that it will eventually need it is a much stronger claim, and it is the one the platform is built on. [ARG] 

"We believe that enabling conventional software developers to harness quantum computers will be key to unlocking new applications."  — Dr. Joe Fitzsimons, on the debut of Beryllium, December 2025 [FACT]

6.1 User-Definable Instruction Sets: The Feature That Explains the Rest

Horizon's own developer documentation series reveals a capability that is rarely mentioned in coverage of the company and which, on examination, underpins the hardware-agnosticism claim more concretely than any marketing language does.

"Our languages give developers full flexibility in defining the instruction sets they use to design quantum operations. Our compiler can switch between instruction sets to output code suited to run on any targeted quantum processor."  — Horizon Quantum, Triple Alpha developer documentation, "Instruction set: Defining commands for quantum operations" [FACT]

In Triple Alpha, the instruction set is not fixed by the vendor — it is a user-definable object. A developer defines a command by specifying its input dimensionality, its output dimensionality, and the Kraus operators that mathematically describe the operation. The compiler then switches between instruction sets to emit code for whichever processor is targeted. [FACT] 

Three consequences follow that are individually significant and jointly hard to replicate. [ARG] 

First, operations are defined by Kraus operators rather than by unitary gates. This means the instruction set can express non-unitary and noisy processes — Horizon's documentation gives amplitude damping at a specified parameter value as a worked example — and measurements are defined by associating Kraus operators with outcomes such that they become POVM elements. A framework that only knows about unitary gates cannot describe a noise channel or a generalised measurement in its own native language. [FACT] 

Second, the system is not restricted to qubits. Commands can act on qudits — subsystems of arbitrary dimension — with the documentation giving an operation acting on a qutrit and a qubit as an example. Subsystem creation is likewise expressed as an operation: a command takes a vacuum state of dimension one and produces a qubit of dimension two. [FACT] 

Third, parameterised gates are backed by user-supplied functions written in Python or MATLAB and uploaded to the platform, with the function computing the Kraus operators for given parameter values. The classical toolchain the developer already uses is wired directly into the definition of the quantum instruction set. [FACT] 

Why this is the load-bearing feature: hardware-agnosticism is usually implemented as a translation table between a fixed canonical gate set and each vendor's native gates, which works until a device does something the canonical set cannot express. Horizon has instead made the instruction set itself a first-class programmable object defined in the general mathematical language of quantum operations. A new processor with an unusual native operation, a qudit-based architecture, or a device requiring explicit noise modelling does not require Horizon to extend a fixed abstraction — it requires a developer to define a command. This is the difference between supporting many backends and being architecturally indifferent to what a backend is. [ARG] 

6.2 Hydrogen at the Hardware Boundary

Horizon's documentation describes Hydrogen as the gate-level language using commands defined in the developer's instruction set, with a restricted form used when targeting hardware. Each line has two parts: a command acting on the quantum system — name, optional parameters, and the addresses of the subsystems it acts on — and an optional assignment of the command's output to a variable, where variables are either subsystem addresses or numeric values. [FACT] 

Two details deserve emphasis. Subsystems are dynamically allocated: a qubit command creates a new subsystem and binds its address to a variable, exactly as a classical program allocates memory and binds a pointer. And developers can bypass allocation entirely through qubit pinning — assigning a variable that names a specific physical qubit from the processor's connectivity graph — with pinned and dynamically allocated qubits mixed freely in the same program. [FACT] 

Dynamic subsystem allocation is the concrete instance of a claim made in Section 5 that might otherwise read as rhetoric. Fitzsimons argues that useful quantum computing requires reinventing decades of computer science including dynamic memory allocation; Triple Alpha implements it. Qubit pinning is its complement: when a developer needs to control exactly which physical qubits are used — because connectivity, calibration quality, or crosstalk on a specific device matters — the abstraction can be set aside for that qubit while remaining in force for the rest of the program. Providing both, and allowing them to coexist in one program, is the behaviour of a systems language rather than a circuit description format. [ARG] 

6.3 Classical-to-Quantum Subroutines (c2q)

Triple Alpha compiles classical C and C++ functions directly into partial quantum circuits, callable as subroutines inside Helium programs via a single line of code and a c2q specification file that sets the entry point and optimisation trade-offs. Many quantum algorithms require classical processing on data held in superposition; implementing that classical computation by hand using reversible quantum gates is, in Horizon's description, a task that may require anywhere from dozens to many millions of operations. [FACT] 

This deserves emphasis because it is the most concrete productivity claim in the entire platform and it is checkable rather than aspirational. Hand-building a reversible circuit for a non-trivial classical function is specialist work measured in weeks. Compiling an existing, tested C function into that circuit automatically — and calling it like any other subroutine — collapses that to a build step. If the compiler output is efficient enough to run on real hardware, this alone accounts for a large share of the order-of-magnitude development-time reduction Horizon claims.

6.4 Portability, Pulse Control, and Deployment

The stack spans an unusually wide range of abstraction. At the top, developers can work in Beryllium without handling quantum mechanics directly — the word "can" is load-bearing. Beryllium does expose quantum operations directly, and many developers will use them; what the language enables is the construction of libraries that abstract the quantum mechanics away, so that one developer can build algorithmic primitives and data structures on quantum processing and the next can import those objects and program with them without engaging the underlying physics. At the bottom, Triple Alpha exposes pulse-level control — direct manipulation of the analogue signals sent from control electronics to the processor. Developers can also write in hardware-vendor frameworks such as OpenQASM or Quil and still use Horizon's deployment infrastructure. [FACT] 

Compiled programs deploy as an API endpoint, letting quantum backends be called from Python scripts, web pages, or even Excel macros. For systems that lack support for complex control flow, Triple Alpha executes programs as hybrid computations, with a classical program dynamically generating circuits and stitching results together to emulate general control flow. [FACT] 

Two things follow. First, the hybrid-execution fallback means Turing-complete programs can run on today's limited hardware rather than waiting for machines that natively support branching — the capability degrades gracefully instead of being unusable until hardware catches up. Second, API deployment is the mechanism by which quantum reaches developers who will never learn quantum computing: the calling application does not need to know what is behind the endpoint. That is a distribution strategy, not just an engineering convenience.

Horizon's stated design order is unusual and worth noting: target the ideal quantum computer first, then translate down to the limited hardware that exists today. Most tooling is built the other way around, starting from present hardware constraints. [FACT] 

The bullish reading is that this front-loads the hard architectural work and leaves Horizon positioned to exploit each hardware generation immediately as it arrives, rather than needing to re-architect. The bearish reading is that it risks building for machines that do not yet exist while competitors ship against machines that do. Both are live; the AQ256 installation in 2027 is the first real test of which is right.

7. Hardware Co-Design: Ember-1 and the Testbed Strategy

Triple Alpha remains Horizon's core IDE and compiler stack, letting developers express algorithms at multiple levels of abstraction while the compiler handles hardware-aware qubit mapping and scheduling. It compiles down through three layers: Beryllium (the highest-level, object-oriented language), Helium (a BASIC-like intermediate language), and Hydrogen (Triple Alpha's assembly language), before targeting actual quantum hardware.

Beryllium, Triple Alpha's object-oriented quantum programming language, was first previewed in December 2025 at the Q2B (Quantum 2 Business) trade show, was extended and stabilized through H1 2026, and reached early access for outside users at the end of Q2 2026. A detailed company technical write-up (published August 11, 2026) documents specific language capabilities: classes and inheritance, structs that bundle classical and quantum data types together, function pointers for dynamic hybrid quantum-classical dispatch, and standard control-flow constructs (if/elif/else, switch, for/while loops, break/continue) applied to quantum programs. [FACT] 

This is a more substantive technical claim than "object-oriented quantum language" alone conveys. Concretely, Beryllium lets a developer define a reusable class (say, an error-correction routine or an algorithmic primitive) once, then instantiate and reuse it the way a classical Python or C++ developer would — rather than hand-writing gate-level circuits for every variant of a task. These are the specific constructs a working developer uses to build and maintain large systems, and their presence is concrete evidence of intent to serve production software engineering rather than circuit-level research. Whether that translates into faster real-world development for paying customers is not yet demonstrable from public disclosures, but the language design itself is documented in specific, checkable detail rather than asserted only in marketing language.

Ember-1, Horizon's in-house testbed, combines a Rigetti Novera 9-qubit superconducting processor with Quantum Machines OPX1000 control electronics and a dilution refrigerator sourced from Maybell Quantum Industries Inc. (the 20-F discloses a material purchase commitment with Maybell, unpaid contract value approximately $581,000 as of March 2026; the refrigerator was already part of the Ember-1 testbed, with the outstanding balance relating to additional components still undergoing acceptance testing); it was inaugurated in Singapore in January 2026 and opened to first external users during Q2 2026. Per Horizon's own Q2 2026 earnings release, Triple Alpha on Ember-1 supports real-time execution of complete programs — eliminating the need for post-selected execution — with pulse- and gate-level access supporting a broad range of quantum operations and workflows. [FACT] 

"Real-time execution eliminating post-selection" is a meaningful technical distinction, not filler language: many cloud-accessed quantum systems require post-selecting results after the fact because the classical control system can't react fast enough during a program's execution to make decisions mid-circuit. Direct, in-house control of the hardware stack is what lets Horizon claim this capability — it's the practical payoff of operating Ember-1 itself rather than renting time on someone else's cloud-accessible system. That said, Ember-1's strategic value here is as a controlled environment for that kind of software-hardware co-design — not as a source of commercially significant compute capacity. A 9-qubit chip should not be read as a production system, and this report states that plainly rather than letting "testbed" carry more weight than it should.

At Canaccord Genuity's Growth Conference (Aug 11, 2026), Fitzsimons quantified why physical co-location matters: roughly one nanosecond of added latency per foot of distance between classical control systems and the quantum processor, which he said makes cloud-only control too slow for the concurrent classical processing Triple Alpha is built to support. This is the speed-of-light limit and therefore a fundamental bound on latency rather than an engineering shortfall; the propagation speed of electrical signals in copper is somewhat slower still. He also laid out explicit hardware-modality trade-offs — trapped-ion systems are typically the most accurate with the longest coherence times but have slower gate times; superconducting systems are faster and more cost-effective for high-shot-count problems; neutral-atom systems are slow when atoms must be physically moved but are fully reconfigurable. [FACT] 

On which modality will ultimately win, Fitzsimons was explicit: "I cannot pick a winner... I do not know which hardware platform is going to win," despite 22 years in the field. That candor is worth noting in a report that stress-tests this company's own multi-modal bet — the CEO's own stated uncertainty is consistent with, not contradicted by, Horizon's hardware-agnostic strategy. [FACT] 

A strategic collaboration with Quantum Machines, announced July 29, 2026, targets an embedded calibration framework for Ember-1 — lightweight calibration routines that run as part of normal system operation rather than requiring lengthy full-system calibration cycles, intended to reduce downtime and increase the Triple Alpha access time available to users. Horizon's VP of Commercial Operations Philip Tan and Staff Scientist Kyle Chu are named participants alongside QM's CEO Dr. Itamar Sivan in the companies' own announcement. [FACT] 

"We are taking the step of creating this testbed because we believe that tight integration between hardware and software is the shortest path to truly useful quantum computing."  — Dr. Joe Fitzsimons, Ember-1 testbed announcement [FACT]

The testbed strategy compounds rather than merely existing. Operating real hardware surfaces control-system behaviour, calibration drift, and latency constraints that cloud APIs abstract away; that information improves the compiler; a better compiler makes Horizon a more valuable partner to hardware vendors; and that partnership yields deeper access to the next generation of hardware — Rigetti's Novera in Ember-1, then IonQ's sixth-generation AQ256 in Dublin. Each turn widens the gap against a competitor working only against cloud endpoints. [ARG] 

Fitzsimons gave two additional operational reasons for on-premises systems at the Canaccord conference: deeper integration with control hardware, and avoiding intermediate software layers that obscure what a machine can actually do. He also noted a practical benefit — the ability to test capabilities safely without risking damage to scarce machines, which a vendor renting cloud time cannot do. The scale progression is 9 qubits superconducting today to 256 qubits trapped-ion in 2027, across two modalities, in two jurisdictions.

8. Intellectual Property Position

For a company whose thesis rests on proprietary compiler technology, the patent position is central rather than incidental.

Horizon holds granted US Patent 11,842,177 B2, "Systems and methods for unified computing on digital and quantum computers," filed June 3, 2021 and granted December 12, 2023, assigned to Horizon Quantum Computing Pte. Ltd., inventors Joseph Francis Fitzsimons and Si-Hui Tan. A continuation application (US 2024/0111506 A1) was filed November 30, 2023, and a European application (EP 4012552 A1) was filed June 9, 2021. [FACT] 

The granted claims describe a compiler that obtains a program in a unified language that is "effectively a classical language, as opposed to a quantum language," performs code refactoring, converts the refactored code through intermediate representations into quantum data structures, and then emits gate-level code conforming to the instruction set and gate-locality constraints of a specific target processor. [FACT] 

That claim language maps directly onto the commercial differentiator described in Sections 5 and 6 — compile classical code, target arbitrary hardware. The core mechanism of the platform is not merely a trade secret or a first-mover head start; it is the subject of an issued US patent with a continuation pending and European coverage sought. The company also reports three patents and 60+ publications associated with Fitzsimons, and both named inventors remain in executive roles (CEO and Chief Science Officer respectively). [ARG] 

Two honest limits. First, a granted patent is not the same as an enforceable moat: claim scope gets tested only in litigation, and none has occurred. Second, industry practice in quantum software leans heavily toward open source and trade secrecy rather than patent enforcement, so the practical value of the filing may lie more in defensive positioning and investor signalling than in the ability to exclude competitors. I have not reviewed the file wrapper or any prior-art challenges, and this section should not be read as a patentability opinion.

9. Revenue Model and the Path to Monetization

Revenue is zero today. How revenue is intended to work is the more useful question, and it is answerable from disclosure.

Per quarterly-update analysis of the company's disclosed strategy, applications built with Triple Alpha run through Horizon's proprietary stack, enabling a usage-based pricing model on cloud deployment. This is a consumption-metered software model layered on top of quantum hardware access, rather than a per-seat licence or a professional-services engagement. [FACT] 

Combined with the API-deployment mechanism described in Section 6.2, the intended economics become legible: a customer's application calls a Triple Alpha endpoint, the workload executes on whichever backend is appropriate, and Horizon meters the usage. That model scales with customer workload rather than with Horizon's headcount — which is precisely the property professional-services revenue lacks, and the reason management's refusal to book services revenue is consistent with the design of the business rather than merely a narrative preference. [ARG] 

Beryllium's object-oriented design is a second-order monetization asset: code reuse and reusable libraries are what create switching costs and network effects in developer platforms. A developer with a library of Beryllium classes has an accumulating reason to stay. That mechanism does not exist yet at any meaningful scale — early access began at the end of Q2 2026 — but it is the intended shape of the moat, and it is why Beryllium's move to general availability is the single most important product milestone to watch. [ARG] 

The gap that remains, stated plainly: a designed revenue model is not a demonstrated one. No pricing has been published, no customer has been named, and no revenue has been recognised. Section 31 treats this as the central open question rather than a solved one.

10. The IonQ Relationship and the AQ256 / Dublin Path

On March 31, 2026, Horizon and IonQ entered into a Quantum Systems Agreement under which Horizon purchased a dedicated trapped-ion quantum computing system from IonQ for aggregate consideration of $35 million. IonQ is responsible for procuring, constructing, installing, and verifying the system against specified performance benchmarks, and installing it in a data center designated and operated by Horizon. This is disclosed as a related-party transaction because IonQ is also an equity holder in Horizon. [FACT] 

The $35 million figure is disclosed directly in Horizon's Q2 2026 financial statement notes (Note 13, Related Party Transactions), not estimated by trade press — a distinction worth making, since the number is frequently cited without its source.

Horizon's own June 11, 2026 press release confirms Dublin, Ireland as the site for its second quantum computer testbed — the IonQ 256-qubit system — describing it as anchoring the company's European headquarters. This is a primary company disclosure rather than trade-press corroboration alone. [FACT] 

Installation remains targeted for 2027, contingent on facility completion. To state it unambiguously: as of this report the AQ256 has not been delivered and the IonQ system is not yet operational. This is a multi-year dependency, not a near-term catalyst, and the report should continue to frame it that way.

At the Canaccord Genuity conference, Fitzsimons offered a specific, heavily hedged technical claim: based on the AQ256's expected qubit count and gate/measurement fidelities, the system should land "right on the cusp of an advantage for certain chemistry problems" if it performs to specification — while explicitly stating "we will not know until we have the system" whether a real advantage materializes. Treat this as management's own stated expectation, not a demonstrated result. [FACT] 

Note: a separate secondary-source transcript summary of this same conference stated the trapped-ion system would be "installed in Dublin in 2025" — that appears to be an error in that source, since it contradicts both Horizon's own 6-K disclosures and Fitzsimons' own remarks elsewhere in the same conference referring to the system as "coming next year." This report retains the 2027 installation target from the primary financial statements.

"I could not be more delighted to be working with IonQ to bring trapped ion and world-leading gate fidelities to our testbed."  — Dr. Joe Fitzsimons, Founder & CEO, April 9, 2026 [FACT]

IonQ holds 4,230,118 Class A shares. The April 2026 Schedule 13D reports this as 13.3%; Horizon's Q2 6-K related-party note reports 7.8%. These are not a before-and-after — they are the same holding measured against different denominators. SEC filings report percentage of class, so 13.3% is computed against Class A shares alone (31,833,549 at the time), while 7.8% is computed against total shares outstanding including the 19,744,585 Class B shares. Comparing the two directly and describing a fall from 13.3% to 7.8% is an error, and one that appears widely in coverage of this company. [FACT] 

On actual dilution: IonQ participated in the PIPE and was therefore not diluted by it. The only dilution IonQ has experienced since Q1 comes from warrant exercises — approximately 2.5 million warrants, corresponding to just under 5% of outstanding shares. Class A shares outstanding grew from 31,833,549 in April to 34,227,495 at June 30, 2026; the 53,972,080 total includes Class B. Note also that Class B shares are convertible into Class A on a one-to-one basis. [FACT] 

Framing IonQ as a stable, 13%-plus strategic anchor investor is no longer accurate. Per the IonQ Side Letter disclosed in the 20-F, IonQ's board-nomination right is explicitly conditioned on holding "not less than 5% of the Company's outstanding voting securities" — at 7.8%, IonQ sits comfortably above that floor today with roughly 2.8 percentage points of headroom, and the commercial relationship — a signed $35M system contract with delivery obligations running into 2027 — is contractually independent of the equity percentage. The threshold is worth monitoring rather than alarming: it would take substantial further issuance, absent any IonQ purchase, to reach it. IonQ separately waived its right to select a company-approved independent director in March 2026, retaining only the ongoing nomination right.

11. Dublin: What Is Actually Unique About It

Dublin has been treated across this report as a line item in the IonQ relationship. It deserves standalone treatment: it is the single largest capital commitment in the company's history, the central catalyst on the 2027 horizon, and — on the evidence assembled here — a configuration no other company in quantum computing currently has.

11.1 A Software Company Buying a Frontier Machine Outright

Under the Quantum Systems Agreement dated March 31, 2026, Horizon purchased a dedicated trapped-ion system from IonQ for $35 million. IonQ is responsible for procuring, constructing, installing, and verifying the system against specified performance benchmarks, then installing it in a data centre designated and operated by Horizon. Horizon's June 11, 2026 press release confirms Dublin, Ireland as the site, describing it as anchoring the company's European headquarters. [FACT] 

The ownership structure inverts the industry's normal relationship. Software vendors rent time on hardware vendors' machines through cloud allocations, and what they can learn is bounded by what the vendor's API exposes. Horizon is buying the machine. That makes it a customer that owns rather than a partner that borrows — with the access, the freedom to instrument, and the operational control that follows. [ARG] 

It is also early-generation frontier hardware. The AQ256 is among IonQ's first sixth-generation, chip-based 256-qubit trapped-ion systems. A 57-person pre-revenue software company is receiving one of the first units of a flagship system ahead of most national laboratories and enterprise buyers. [FACT] 

11.2 Multi-Modality by Ownership, Not Integration

With Dublin operational, Horizon will own and operate hardware in two fundamentally different physical architectures: superconducting (Rigetti Novera, 9 qubits, Singapore) and trapped-ion (IonQ AQ256, 256 qubits, Dublin). These are not variants of one approach — they differ in how qubits are physically realised, how gates are executed, what connectivity is native, how errors accumulate, and what timescales govern operation. [FACT] 

This is the difference between claiming hardware-agnosticism and demonstrating it. A compiler that produces good output for superconducting devices and good output for trapped-ion devices, developed against both at pulse level by the same team, is a materially stronger claim than a compiler that targets many backends through vendor APIs. Dublin converts Horizon's central differentiator from an architectural argument into an operating fact.

The scale progression matters as much as the modality change: 9 qubits to 256 qubits is not an incremental step. Compiler behaviours that are invisible at 9 qubits — routing pressure, scheduling under crosstalk, error-aware placement across a large register, the cost of mid-circuit measurement at depth — only become observable and optimisable at scale. Ember-1 lets Horizon develop the mechanisms; Dublin is where they get stress-tested. [ARG] 

11.3 Jurisdiction as Strategy

Dublin gives Horizon an EU operating footprint alongside its Singapore headquarters and its US listing. Irish government officials publicly welcomed the investment, and the company frames the facility as its European headquarters. [FACT] 

Three consequences follow. First, sovereign and defence programmes in Europe frequently carry data-residency and jurisdictional requirements that cannot be met from Singapore or the United States; an EU facility is a precondition for that class of customer, not a convenience. Second, Ireland has existing semiconductor and data-centre infrastructure, a dense pharmaceutical manufacturing base, and an established regime for attracting technology investment. Third, the three-jurisdiction structure — Singapore, Ireland, United States — gives Horizon standing with allied-nation programmes in each bloc, which is unusual for a company of this size and directly relevant to the national-security vertical.

A detail connecting the leadership bench to the Dublin decision: Chief Legal and Compliance Officer Catherine Fitzsimons is an Irish-qualified lawyer. Per the 20-F, she holds a Bachelor of Civil Law and a Post-Graduate Diploma in International Financial Services Law from University College Dublin plus a Diploma in Applied Finance Law, is a member of the Law Society of Ireland, practised financial services law in Ireland from 2009 to 2015, and served as a non-executive director on Fidelity International's Irish fund ranges, ultimately becoming Fidelity's Director of Strategic Initiatives and Head of Global Product Legal. [FACT] 

Her appointment was announced in the pre-close materials alongside the incoming board — that is, before the Dublin site was publicly disclosed in June 2026. A company establishing a European headquarters in Ireland, navigating Irish corporate structuring, IDA Ireland engagement, property, and cross-border regulatory work, hired a senior Irish-qualified financial-services lawyer with Irish fund-directorship experience as its Chief Legal Officer. Read together, the Dublin decision looks less like an opportunistic site selection and more like a planned European strategy with the legal capability put in place ahead of it.

Horizon is also building the Irish talent pipeline directly rather than assuming it. On August 20, 2026 the company hosted a panel in Dublin on quantum computing careers in Ireland, featuring Horizon's Dr Ray Lloyd alongside Dr Felix Binder of Trinity College Dublin and Dr Mike Dascal of Fidelity — pairing an academic institution with a financial-services employer already active in quantum. [FACT] 

A facility requires staff. Recruiting quantum engineers into a new jurisdiction is a real constraint on any hardware deployment, and convening the local ecosystem eleven months before an installation is the behaviour of a company that intends to operate the site rather than merely site it. It also reinforces the reading in the preceding paragraph: legal capability seated first, talent pipeline built second, hardware last.

The same footprint carries the corresponding risk, already noted in Section 27: three jurisdictions means three export-control regimes, and quantum hardware is precisely the category most exposed to shifting national-security trade rules. [ARG] 

11.4 Why Dublin Is the Report's Central Catalyst

At the Canaccord Genuity conference, Fitzsimons said that based on the AQ256's expected qubit count and gate and measurement fidelities, the system should land "right on the cusp of an advantage for certain chemistry problems" if it performs to specification — while stating explicitly that "we will not know until we have the system." [FACT] 

That is the sharpest testable claim management has made about any of its hardware. It converts Dublin from a capacity expansion into an experiment with a stated hypothesis, and it is why Section 22 examines the chemistry pathway in depth: if the claim holds even partially, Horizon will have demonstrated its software layer producing a real computational result on production-grade hardware, in the one application domain where quantum advantage has the cleanest theoretical foundation.

What to watch, in order: site selection and lease completion, construction and fit-out milestones, IonQ's manufacturing and delivery schedule, installation start, benchmark verification against the contract's performance specifications, and finally published results. Each is observable and dated. Section 29 tracks them.

Two honest qualifications carried forward. The $35 million is the system cost under the Quantum Systems Agreement; total facility capex — building, cryogenics, power, shielding, staffing — is not separately disclosed, and Dublin is the most likely source of acceleration in investing cash outflow beyond the $5.5 million recorded in H1 2026. And the Dublin location itself appears in Horizon's press release and trade coverage rather than in the SEC filing text, which refers only to a data centre designated and operated by the Company. [FACT] 

12. Multi-Modal Hardware Ecosystem

Horizon's 20-F names its "key collaborations" as Rigetti Computing, Oxford Quantum Circuits (OQC), Alice & Bob, and QuEra Computing Inc. Horizon collaborates with QuEra via cloud systems and is not currently pursuing an on-prem neutral-atom deployment. Fitzsimons' Aug 11 conference remark about a neutral-atom testbed ("if anyone wants to send me one in the post") was a joke about accepting a free system to operate on-prem, and was not a comment on the QuEra relationship or any other company's. [FACT] 

The tiering used in this report is our own classification of Horizon's relationships by integration depth, and should not be read as the company's own framework. Horizon does describe partnership levels in its public materials, but those levels differ: they do not treat the in-house testbed or operating hardware as a distinct level, and the highest level in Horizon's own formulation is hardware vendors adopting Triple Alpha as their default software layer — a category in which the company states it is only very early in discussions. [FACT] 

That distinction matters and is worth stating before the table below. Our classification is a reasonable way to sort these relationships, but it is ours. It is also worth noting what Horizon's own top tier implies: a hardware vendor shipping Triple Alpha as the default software layer with its machines would be a materially stronger commercial outcome than anything in our Tier 1, and no such arrangement exists today.

By integration depth: Rigetti (Ember-1, in-house operated) and IonQ (AQ256, contracted for Dublin) are the deepest. AQT — Alpine Quantum Technologies, trapped-ion, cloud access, announced April 2026 — and Alice & Bob — cat-qubit emulators for fault-tolerant computing, announced January 2026 — sit at an integration and joint-development level. [FACT] 

Horizon's own technology pages list specific, named processors accessible as backends: IBM's Brisbane and Sherbrooke, IonQ's Forte, Aria 1, and Harmony, IQM's Garnet, OQC's Toshiko, Rigetti's Ankaa-2, Ankaa-3, and Ankaa-9Q-3, plus AQT's simulator and AWS's TN1 simulator — illustrative examples out of the 30+ backends the company states are supported in total. [FACT] 

IQM's inclusion (Garnet) is worth flagging on its own: it is a named hardware vendor rarely cited in coverage of Horizon. Whether Horizon's relationship with IQM, IBM, AWS, or OQC goes beyond basic backend/API-level integration — the lightest tier in the company's own stated model — isn't disclosed; those four are named as processor/simulator sources, not as collaboration or joint-development partners the way Rigetti, IonQ, AQT, and Alice & Bob are.

This is materially broader than a simple IonQ-Rigetti framing. The differentiated claim is genuine multi-modal breadth across trapped-ion, superconducting, and (via Alice & Bob) fault-tolerant/cat-qubit architectures — though it is worth noting that "backend support" in a compiler/SDK sense (targeting a vendor's API) is a lower bar than the tightly integrated, in-house-operated relationships Horizon has with Rigetti and (eventually) IonQ.

Stated plainly, this is the structural claim the whole thesis rests on: on the evidence assembled here, Horizon is the only publicly traded pure-play quantum software company that operates its own hardware, across two different modalities, by deliberate design — Ember-1 running today in Singapore and the contracted AQ256 for Dublin. Qiskit is coupled to IBM's roadmap; PennyLane and Classiq are software-only. Neither configuration produces the same position. [ARG] 

Why that position may be genuinely hard to replicate rather than merely unusual: the nanosecond-per-foot latency constraint Fitzsimons described (Section 7) is a physical limit, not an engineering preference. Concurrent classical processing inside a running quantum program requires the classical control system to sit physically adjacent to the processor. A software vendor working only against cloud APIs cannot develop or validate that capability at all, and a single-vendor stack can only develop it for one modality. Horizon's in-house testbeds are the mechanism by which a software company gets access to a problem that is otherwise closed to it — and the reason the multi-modal bet is a hedge rather than a hedge-shaped indecision.

The partnership set is best understood as four tiers of decreasing integration depth, not as a vendor list. Each tier buys Horizon something different. [ARG] 

The table below is our own classification by integration depth, not Horizon's stated partnership framework.

Tier (our classification)	Partners	Nature of the relationship	What it gives Horizon
Tier 1 — Owned and operated hardware	Rigetti (Novera QPU, Ember-1); IonQ (AQ256, Dublin)	Horizon purchases and operates the machines. Rigetti's 9-qubit Novera runs in Singapore today; IonQ's $35M 256-qubit system is contracted for 2027 with IonQ responsible for procurement, construction, installation, and benchmark verification.	Pulse-level access, real-time execution without post-selection, freedom to test destructively — the co-design flywheel (Section 7). Two different modalities.
Tier 2 — Control-layer co-development	Quantum Machines (OPX1000; embedded calibration collaboration, Jul 29, 2026)	Joint engineering on an embedded calibration framework running as part of normal system operation rather than in long dedicated cycles. Named participants include Horizon's VP of Commercial Operations Philip Tan and Staff Scientist Kyle Chu alongside QM CEO Dr. Itamar Sivan.	Uptime and utilisation on Ember-1 — the metric that determines how much Triple Alpha development time the testbed actually yields. Also joint external presence (Q2B Tokyo).
Tier 3 — Architecture-specific integration	AQT (trapped-ion, Apr 2026); Alice & Bob (cat-qubit fault-tolerant emulators, Jan 2026); QuEra (neutral-atom, named in the 20-F)	Integration and joint-development relationships without Horizon owning the hardware. Alice & Bob provides exposure to error-corrected architectures; AQT to an independent European trapped-ion stack.	Coverage of the modalities Horizon does not operate — including, via Alice & Bob, the fault-tolerant era the whole industry is building toward.
Tier 4 — Backend compatibility	IBM (Brisbane, Sherbrooke); IQM (Garnet); OQC (Toshiko); AWS (TN1 simulator); Rigetti Ankaa series; IonQ Forte/Aria/Harmony	API-level targeting through the compiler. 30+ backends in total.	Portability proof: the claim that a Triple Alpha program runs anywhere is demonstrated across vendors, not asserted.

Reading the tiers together reveals the strategy: Horizon owns hardware in the two modalities most likely to reach commercial advantage first, co-develops the control layer that determines whether owned hardware is actually usable, integrates with the architectures it does not own including the fault-tolerant frontier, and maintains compiler compatibility across everything else. That is coverage of the full architectural space by a company of 57 people — achieved by varying integration depth according to strategic importance rather than by attempting equal depth everywhere. [ARG] 

Two supporting relationships deserve mention. Maybell Quantum Industries supplies the dilution refrigerator for Ember-1 under a purchase commitment disclosed in the 20-F (approximately $581,000 unpaid as of March 2026), which is what makes an in-house superconducting testbed physically possible for a software company. And the QIP 2027 anchor sponsorship (Section 15) is a partnership with the research community itself — the venue where the theoretical work Horizon depends on is presented, and where it recruits.

The honest qualification carried forward: Tier 4 breadth is the least meaningful of the four. Supporting forty backends instead of thirty is an engineering achievement given that each has a distinct gate set, connectivity graph, and error profile — but it is not evidence of commercial traction, and Section 30 is explicit that it should not be scored as such.

The asymmetry against hardware vendors is worth stating explicitly. A superconducting or trapped-ion manufacturer carries binary architectural risk: if its modality loses, decades of capital and engineering are stranded. Horizon's exposure is structurally different — its compiler targets 30+ backends across four architectures, so the question it faces is whether quantum computing works at all, not which physical approach wins. That is a materially wider set of futures in which the company survives, and it is the single strongest structural argument in the bull case. [ARG] 

The honest limit on that argument: surviving is not the same as capturing value. A software layer that works everywhere still needs someone to pay for it, and no disclosure to date establishes that enterprises will license a third-party compiler rather than use whatever their hardware vendor ships for free. Modality-neutrality removes one risk; it does not by itself create a business.

13. The Founder: Joe Fitzsimons and Why It Matters Here

Prior versions of this report treated Fitzsimons' background as a credential list. That undersells something material to the investment case: he is not a technologist who moved into quantum computing, nor a businessman with a physics degree. He is one of the people who created a subfield of quantum information science, and the specific subfield he created is directly adjacent to the problem Horizon is now solving commercially.

13.1 The Research Record

Fitzsimons holds a BSc in theoretical physics from University College Dublin (first-class honours) and a doctorate from Oxford's Department of Materials (2007), with a thesis titled "Architectures for Quantum Computation under Restricted Controls." He held Junior and Senior Research Fellowships at Merton College and Oxford's Department of Materials (2007–2010), then built a research group in Singapore, rising to Principal Investigator at the Centre for Quantum Technologies and a tenured associate professorship at the Singapore University of Technology and Design. [FACT] 

His Google Scholar profile records 7,781 citations across quantum computation, quantum information, quantum cryptography, blind quantum computing, and quantum verification — a figure materially higher than the 6,900+ cited in the 20-F, reflecting continued accrual. He is credited with 60+ publications and three patents. [FACT] 

The foundational paper is Broadbent, Fitzsimons and Kashefi, "Universal Blind Quantum Computation," published at the 50th IEEE Symposium on Foundations of Computer Science (FOCS 2009) — FOCS being one of the two top-tier venues in theoretical computer science globally. The protocol it introduced allows a user to delegate a computation to a remote quantum computer while keeping the computation private even from the machine executing it. [FACT] 

The follow-on record is unusually strong for any researcher: an experimental demonstration published in Science in 2012 (Barz, Kashefi, Broadbent, Fitzsimons, Zeilinger and Walther); experimental verification of quantum computation in Nature Physics in 2013; "Optimal blind quantum computation" in Physical Review Letters in 2013; "Unconditionally verifiable blind quantum computation" in Physical Review A in 2017; a review article in npj Quantum Information in 2017; and "Post hoc verification of quantum computation" in Physical Review Letters in 2018. [FACT] 

The 2012 Science demonstration was co-authored with Anton Zeilinger, who received the 2022 Nobel Prize in Physics for experimental work on entanglement. Independent literature in the field describes verification via encryption through blind quantum computing as a scheme "initiated by Fitzsimons and Kashefi" — that is, the field itself credits him with originating the line of work, not merely contributing to it. [FACT] 

His research was funded by Singapore's National Research Foundation Fellowship (NRF-NRFF2013-01) and by the US Air Force Office of Scientific Research (grant FA2386-15-1-4082) — the latter worth noting given the national-security dimension of Horizon's targeted sectors. [FACT] 

13.2 Why This Specific Background Is the Right One

The relevance is not prestige by association. Blind and verifiable quantum computation is the study of how to run a computation on a quantum machine you do not control and cannot fully observe, and still trust the answer. That is, structurally, the same problem as building a compiler and runtime that must reason about what a quantum program will do on hardware whose noise, connectivity, and error behaviour vary by vendor and drift over time. Fitzsimons spent roughly fifteen years on the theoretical version of the problem before starting a company to solve the engineering version. [ARG] 

He is also, by his own account, the first tenured quantum-computing professor to leave academia for a startup — he gave up a tenured position in 2018 to found Horizon. Tenure is the single most secure position in academic life; the decision to abandon it is a costly signal about conviction that a salaried hire could not send. [FACT] 

Two further points that a purely financial reading of this company misses. First, the deep-research background explains the company's unusual design order — target the ideal quantum computer first, then translate down to real hardware (Section 6.4). That is how a complexity theorist approaches a problem and it is not how most startups build; whether it proves right or wrong, it is a coherent strategy rather than a drifting one. Second, it explains the recruiting: Chief Science Officer Dr. Si-Hui Tan is co-inventor on the granted patent (Section 8) and a CQT alumna, and the QIP 2027 anchor sponsorship (Section 15) is the company placing itself at the centre of the research community it draws from. A founder with standing in that community can hire from it in a way an outsider cannot.

Governance note in the founder's favour: Fitzsimons holds 19,744,585 Class B shares carrying three votes each — 65.0% of voting power — under a two-year lock-up from March 19, 2026. He controls the company's direction through the AQ256 delivery window and cannot sell into it. Founder control is a risk in the abstract; here it means the person who originated the underlying science sets the technical roadmap rather than a quarterly-earnings cycle. [ARG] 

13.3 The Co-Founder of the Science: Dr. Si-Hui Tan

Prior versions of this report gave the Chief Science Officer a single line. That is a serious omission: she is co-inventor on the granted patent that covers the company's core mechanism, and her own research record is independently strong.

Si-Hui Tan holds a BSc with honours in Physics from Caltech and a PhD in Physics from MIT, where she was a recipient of the MIT Presidential Fellowship and wrote her thesis, "Quantum state discrimination with Gaussian states," under Seth Lloyd — one of the founding figures of quantum computing. Her Google Scholar profile records 1,981 citations across quantum cryptography, quantum information, and quantum optics. [FACT] 

Her career path: A*STAR's Data Storage Institute in Singapore from 2010 on industry-applied quantum optics; then in 2013 she joined Joseph Fitzsimons' group to work on quantum cryptographic protocols for secure delegated quantum computing; subsequently postdoctoral researcher at the Singapore University of Technology and Design and Visiting Professor at the Niels Bohr International Academy, University of Copenhagen. She was named to the SG100 Women in Tech list in 2021 and sits on the College Advisory Board of the College of Science at Nanyang Technological University. [FACT] 

Her research contributions include quantum homomorphic encryption — schemes that allow computation on encrypted quantum data, so a server can process a client's problem without learning what the problem is. That is the same intellectual territory as Fitzsimons' blind and verifiable computation work, and the two have collaborated since 2013. [FACT] 

The significance for the investment case: Horizon's two most senior technical people spent over a decade jointly building the theory of how to compute reliably and privately on machines you do not control, and are named together as co-inventors on the patent that turns that background into a compiler. A CEO with a strong research record and a Chief Science Officer hired for name recognition would be a weaker configuration than two people with a shared, decade-long research programme who then commercialised it together. This is a genuine founding technical partnership, not a founder plus a credential.

14. Leadership and Recent Organizational Changes

Gregory Gould has served as CFO since August 2025 and appears throughout the Q1 and Q2 earnings calls. He holds a BS in Finance from MIT's Sloan School of Management (1990), began his career at Goldman Sachs (ultimately Managing Director and Co-Deputy Director of its Global Technology Investment Research Group), and has since held CFO roles at Groundspeed Analytics (InsurTech, 2022–23) and FitMatch Inc. (2023–24), among earlier finance and advisory roles. Per the 20-F, he also serves as an unpaid strategic advisor to Penchant Holdings, Inc. and a venture partner at 14 Peaks Capital Advisors — disclosed by the company as not a conflict of interest. [FACT] 

The Board per the 20-F (filed March 25, 2026) consisted of four people from day one of the public company: Fitzsimons (Chairman/CEO), Harry You, Danielle Lambert, and Jill Turner — all but Fitzsimons designated independent under Nasdaq rules. Turner, You and Lambert were founding public-company directors rather than later additions; Peter Oey is the only subsequent board addition. Jill Turner, Chief Human Resources Officer of Broadcom Inc. since April 2021 (previously SVP of HR at Lumen Technologies and various HR leadership roles at Honeywell). Danielle Lambert is CEO and Founder of Penchant Holdings, Inc. (since December 2022), was an early investor in and advisor to Nest Labs through its acquisition by Alphabet, and previously spent eight years at Apple, ultimately as VP of Human Resources. [FACT] 

A governance detail worth flagging on its own: an entity called Penchant Family Holdings LLC — controlled by Danielle Lambert's Penchant Holdings, Inc., of which she serves as President — invested $1,000,000 in the PIPE that helped fund the SPAC close, per the 20-F. Lambert is a sitting Company director. This is disclosed rather than hidden, but it is a related-party PIPE participation by a sitting director and is rarely noted in coverage of the company. [FACT] 

Dr. Joe Fitzsimons serves as Founder & CEO, Dr. Si-Hui Tan as Chief Science Officer, Catherine Fitzsimons as Chief Legal and Compliance Officer (effective May 11, 2026), and Harry You as a Director.

The 20-F's filed version of Fitzsimons' biography is considerably more specific than his own self-reported conference remarks: BS in theoretical physics from University College Dublin (first-class honors), doctorate from Oxford's Department of Materials (2007, thesis "Architectures for Quantum Computation under Restricted Controls"), Junior/Senior Research Fellow at Merton College and Oxford's Department of Materials (2007–2010), roles at Singapore's Centre for Quantum Technologies rising to Principal Investigator (2017), and tenured associate professor at the Singapore University of Technology and Design (2018) before founding Horizon that same year. He is credited with over 6,900 citations, 3 patents, and 60+ publications, and serves as president of the Southeast Asia Quantum Industry Association. [FACT] 

The Board approved Qian Yi Amanda Chew's appointment as Chief Product Officer on July 28, 2026, effective August 17, 2026, per company announcement. Chew joined Horizon in 2020 as Product Manager, was promoted to Director of Product and then Vice President of Product — a title she continues to hold until the CPO appointment takes effect — and previously held product roles at Microsoft, including Senior Program Manager for Visual Studio App Center. She holds a Bachelor of Science in mathematics and computer science from Brown University, plus executive certifications from Stanford, Wharton, and INSEAD. [FACT] 

As of this report's preparation, Chew's operative title is still VP of Product; any reference to her as CPO should carry the August 17, 2026 effective date.

Peter Oey — CFO of Grab Holdings since April 2020 (previously CFO of LegalZoom.com and Mylife.com, and earlier Corporate Controller at Activision Blizzard) — was appointed to Horizon's board on April 29, 2026, per a 6-K filed May 4, 2026. He serves as chairman of the Audit Committee and a member of the Compensation Committee, succeeding Harry You as chair; You remains an Audit Committee member. Danielle Lambert stepped down from the Audit and Compensation Committees as a result — she had chaired the nominating & corporate governance committee at the time of the 20-F. [FACT] 

A CFO-caliber board member with SEA-market operating experience (Grab) chairing the Audit Committee is a reasonable governance-strengthening signal for a Singapore-headquartered issuer, though I have not independently assessed his committee work beyond what the 6-K discloses.

Dr. Denis Chevallier holds the title of Director of Hardware at Horizon and is a named external spokesperson for the company's technical program — for example, presenting alongside Quantum Machines' VP of Business Development Dr. Gilad Ben-Shach at Q2B 2026 Tokyo on real-time execution using Triple Alpha and the OPX1000 control system. [FACT] 

Taken together, the bench Horizon has assembled in twelve months is disproportionate to its size and is one of the more underrated facts in this report. A pre-revenue company with 57 employees has recruited a CFO who was a Goldman Sachs Managing Director and Co-Deputy Director of Global Technology Investment Research (Gould, in post since August 2025, ahead of the de-SPAC); an Audit Committee chair who is the sitting CFO of Grab Holdings, a company several orders of magnitude larger (Oey); a founding public-company director who is CHRO of Broadcom (Turner); a director who spent eight years at Apple ending as VP of HR and was an early Nest investor (Lambert); and a Chief Product Officer promoted from within after six years at the company (Chew), who previously held product roles at Microsoft. The founder holds an Oxford doctorate with 6,900+ citations and presides over the Southeast Asia Quantum Industry Association. [FACT] 

Why this is evidence rather than decoration: executives at that level have optionality, and accepting a board seat or an operating role at a pre-revenue Nasdaq micro-cap carries reputational cost if the company fails. Their willingness to attach their names is a form of diligence signal that does not appear anywhere in the financial statements — though it is a signal about people's judgment, not about product-market fit, and Section 30 is explicit that it should not be scored as commercial validation.

15. Business Development and Ecosystem Engagement

The items below are visible, dated, and — in the case of conference activity — tied directly to a disclosed expense line rather than resting on a narrative claim.

Horizon Quantum has been named anchor sponsor of QIP 2027 (the 30th International Conference on Quantum Information Processing), to be held February 20–26, 2027 in Singapore, hosted by the Centre for Quantum Technologies at the National University of Singapore. QIP is the flagship annual venue for theoretical quantum information research, historically drawing roughly 1,000 attendees and 500+ submissions. Founder & CEO Dr. Joe Fitzsimons is a CQT alumnus. Sponsorship was announced by both QIP 2027's official account and Dr. Fitzsimons personally in early August 2026. [FACT] 

An anchor sponsorship of the field's leading theoretical conference is a credibility and recruiting signal aimed at the academic and research community Horizon draws its technical talent from — it is not a commercial pipeline indicator, and this report doesn't treat it as one. But given that Horizon's core claim is technical leadership in quantum software, standing as anchor sponsor of the venue where the field's peer-reviewed theoretical work gets presented is a relevant, checkable data point that a purely financial reading of the company would miss entirely.

At Q2B 2026 Tokyo, Horizon's Dr. Denis Chevallier (Director of Hardware) and Quantum Machines' Dr. Gilad Ben-Shach (VP of Business Development) jointly presented "Real-Time Execution Beyond Circuit Programs with Triple Alpha and the OPX1000" — a technical session specifically on the real-time execution capability described in Section 6, not a generic marketing appearance. [FACT] 

"During the second quarter of 2026 we reached important milestones with Beryllium... we also announced a strategic collaboration with Quantum Machines, which aims to further our technical capabilities in calibration."  — Dr. Joe Fitzsimons, Q2 FY2026 shareholder communication, August 4, 2026 [FACT]

Horizon's own Q2 2026 earnings release attributes a 63% period-over-period increase in sales and marketing expense (ex-compensation items) primarily to "increased trade show activity and industry engagement." Total S&M expense was $0.4 million in Q2 2026, up 51% year-over-year on a reported basis — a small absolute number, but one that corroborates the conference presence documented above rather than leaving it as an unverified claim. [FACT] 

This is a useful example of triangulating a qualitative claim against a financial disclosure: the company says it's showing up at industry events, and the expense line moved in a way that's consistent with that claim, on a dated, filed document. It doesn't prove the events are generating commercial pipeline — no disclosure available says that — but it does confirm the underlying activity actually happened at the scale implied.

Horizon has an internal applications team, discussed on the Q2 2026 earnings call and again at the Canaccord Genuity conference, focused on three unnamed high-value problems across three industries. He framed this explicitly as an alternative to near-term enterprise sales: in his view, proof-of-concept engagements with outside enterprises are unlikely to convert to production deployments before the company can demonstrate real quantum advantage, so Horizon is prioritizing internal, self-directed application development over broad customer acquisition for now. [FACT] 

Read constructively, this is proactive product development rather than a holding pattern. Building the applications in-house means that when quantum advantage does arrive, Horizon owns working demonstrations in three high-value domains and can convert the inbound interest documented in Section 20 against a finished artifact rather than a pitch. The obvious counter is that internal work generates no external validation and no revenue in the meantime; both readings are available on the evidence, and the report does not claim the favorable one is proven.

Fitzsimons also explained the company's pre-revenue positioning as a deliberate choice rather than a simple absence of demand: he said any revenue earned before demonstrated quantum advantage would likely come from professional services bundled with system access, which he considers misleading because it wouldn't reflect the step-change value the company expects once real advantage arrives — an analogy he drew to the gap between GPT-1/GPT-2 and GPT-5. [FACT] 

This is a genuine addition to the investment case, not just color: it reframes the $0 revenue figures in Section 17 from an ambiguous "no traction yet" reading into an explicit, stated strategic choice — and, read on its merits, an act of capital and product discipline. A pre-advantage quantum vendor can almost always manufacture a revenue line by wrapping consulting hours around hardware access; declining to do so forgoes an easy optics win in exchange for keeping the eventual step-change legible. The GPT-1/GPT-2 versus GPT-5 analogy Fitzsimons used is the substance of the argument, not decoration: early-generation output can be technically real and still commercially meaningless, and booking revenue against it would misprice what the platform is actually for.

The honest counterweight is that this choice removes a data point investors could otherwise use to gauge market pull, and it asks shareholders to fund several more years of losses on management's judgment about timing. It is not the same claim as "the company has tried to sell and failed to find buyers" — but it is also not costless, and Section 30 is explicit that it should not be scored as evidence of demand.

On developer adoption strategy, Fitzsimons was explicit that Horizon has taken a narrower approach than competitors: rather than pushing broad developer mindshare the way Qiskit has, the company's near-term go-to-market focus is on hardware manufacturers, with broader access opening over time. [FACT] 

16. Financial Statements: Balance Sheet

The following is drawn directly from the unaudited condensed consolidated balance sheets in Horizon's Q1 and Q2 2026 Form 6-K exhibits.

(US$)	Dec 31, 2025	Mar 31, 2026	Jun 30, 2026
Cash and cash equivalents	$222,939	$96,602,279	$113,254,440
Total current assets	$969,311	$98,835,566	$120,122,813
Total assets	$4,831,803	$102,301,702	$124,045,763
Derivative liabilities – warrants	$0	$15,273,680	$76,778,574
Total liabilities	$9,494,428	$18,429,117	$80,370,229
Total stockholders' equity	$(4,662,625)	$83,872,585	$43,675,534

The single largest balance-sheet swing between Q1 and Q2 2026 is the warrant derivative liability, which grew from $15.3M to $76.8M as Horizon's share price rose sharply during the quarter (the Level 3 valuation model used a $27.76 stock-price input as of June 30, versus $19.93 at the August 21, 2026 close). That liability growth — not operating deterioration — is what drove total stockholders' equity down from $83.9M to $43.7M sequentially, even as cash grew. [FACT] 

This is a useful illustration of why adjusted EBITDA, not GAAP net loss or the equity roll-forward, is the right lens for assessing operating trajectory here — the warrant mechanics move faster and larger than the underlying business.

Stated for readers who see only the headline: the $115.2M Q2 net loss is not an operating result. Roughly $108.3M of it is a non-cash mark-to-market charge that exists because the share price rose during the quarter, and it consumed no cash, funded no expense, and reflects no deterioration in the business. Anyone comparing HQ's GAAP net loss to peers without adjusting for warrant accounting will reach a badly wrong conclusion. [FACT] 

The corollary — which the bull case must own rather than skip — is that the genuinely operational figures are still moving the wrong way. Operating loss and adjusted EBITDA loss are both running at roughly double their year-earlier levels, and adjusted EBITDA loss has risen in each of the last three sequential quarters, driven by real headcount and testbed spending. Those numbers are not accounting artifacts, and the case for them is that they are investment ahead of an inflection, not that they are small.

17. Financial Statements: Operations and Cash Flow

 

Operating loss by quarter: $4.87M (Q3'25) → $4.55M (Q4'25) → $6.50M (Q1'26) → $7.18M (Q2'26). The trend is not monotonic — operating loss fell from Q3 2025 to Q4 2025 before rising in the two quarters since. Adjusted EBITDA loss has risen in each of the last three sequential quarters: $3.01M → $3.03M → $4.10M → $5.46M. Both measures are now roughly double where they stood a year earlier, driven primarily by headcount growth (52 to 57 employees quarter-over-quarter, ~68% YoY) and the costs of operating as a public company. [FACT] 

By expense line, Q2 2026 vs. Q2 2025: R&D $2.6M, up 117% (100% ex-comp adjustments), attributed to headcount growth and hardware testbed setup costs. Sales & marketing $0.4M, up 51% reported (63% ex-comp), attributed to increased trade show activity and industry engagement — see Section 15. G&A $3.8M, up 236% reported (191% ex-comp), attributed to headcount and the costs of operating as a newly public company. [FACT] 

The S&M line is the smallest of the three in absolute terms, but it's the one with a direct, named tie to a specific activity (conference sponsorship and presentations) rather than generic "public company cost" — which is why Section 15 exists as more than a marketing aside.

 

Cash flow for the six months ended June 30, 2026: $9.28M used in operating activities, $5.52M used in investing activities (primarily property/equipment and construction-in-progress — consistent with testbed and facility buildout), and $127.85M provided by financing activities ($98.17M net from the merger/PIPE, $27.19M from warrant exercises, $2.5M from SAFE note proceeds). [FACT] 

At the current six-month operating burn rate (~$9.3M), the $113.3M cash balance implies roughly six years of operating-only runway — but this is a simple arithmetic extrapolation, not a company guidance figure, and it excludes AQ256/Dublin facility capex, which is highly likely to increase from H1 2026's $5.5M investing outflow as the Dublin buildout proceeds toward a 2027 installation target. Treat this runway figure as a floor, not a forecast. [INFER] 

Revenue was $0 in both Q1 and Q2 2026, and $0 for the six months ended June 30, 2026 (versus $38,462 in H1 2025). FY2025 full-year revenue was $38,873. The trend is toward less revenue, not more, ahead of any inflection tied to quantum advantage. [FACT] 

The Q2 2026 6-K states that management has concluded that going-concern doubt has been alleviated for at least the next twelve months given the ~$120M in gross proceeds received at the business combination close, but also notes that future capital requirements will depend on revenue growth and R&D/commercialization spending, and that the company may need to raise additional equity or debt financing if required. [FACT] 

18. Capital Structure and Dilution

Dilution is frequently discussed qualitatively; the overhang is quantifiable, and quantified here.

Instrument	Outstanding (Jun 30, 2026)	Detail
Class A ordinary shares	34,227,495	Publicly traded, Nasdaq: HQ
Class B ordinary shares	19,744,585	Held by Fitzsimons; 3 votes/share; 2-yr lock-up from Mar 19, 2026
Public warrants (HQWWW)	765,554	Exercise price $11.50; ~79% of the original 3,159,500 already exercised
Private placement warrants	2,884,660	Exercise price $11.50; none exercised as of Jun 30, 2026
Employee stock options	6,341,712	Weighted-avg exercise price $3.36; 3,793,506 vested/exercisable
RSUs	474,784	Weighted-avg grant-date fair value $11.11

Fully diluted, the option pool alone (6.34M shares at a $3.36 weighted-average strike, well below the $19.93 August 21, 2026 close) represents a meaningful in-the-money overhang relative to the 53.97M shares currently outstanding — roughly 12% of outstanding shares before counting warrants or RSUs. [INFER] 

None of this is unusual for a recent de-SPAC technology company, but it should be modeled explicitly in any per-share valuation work rather than left as a qualitative risk bullet.

19. Ownership and Governance

Dr. Fitzsimons holds 19,744,585 Class B ordinary shares (three votes per share), representing approximately 38.3% of ordinary shares on an as-converted basis but 65.0% of total voting power as of the March 19, 2026 closing; these shares are subject to a two-year lock-up. [FACT] 

Peak XV–affiliated funds disclosed beneficial ownership of 6,468,999 Class A shares — 20.3% "of the class" — in a Schedule 13D filed July 9, 2026, based on 31,833,549 Class A shares outstanding as of April 24, 2026 (a Class A-only denominator, per the 13D's own "Note to Row 13" language). [FACT] 

That Class A-only basis is now stale on two counts: Horizon's Class A count alone had grown to 34,227,495 by June 30, 2026, and total shares outstanding (Class A + Class B) reached 53,972,080. Recalculating Peak XV's disclosed share count against the current Class A-only figure — the basis comparable to their original 20.3% — gives approximately 18.9%. Recalculating against the combined Class A + Class B total instead gives approximately 12.0%. Neither is a disclosed current figure; both are arithmetic, assuming Peak XV's own share count hasn't changed since April, which no filing confirms. [INFER] 

IonQ's stake, per Horizon's June 30, 2026 related-party note, is 7.8% — computed against the combined Class A + Class B total (4,230,118 / 53,972,080). The April 2026 Schedule 13D reported 13.3% against Class A alone. Both describe the identical 4,230,118-share position. The 13.3% figure was never an accurate reflection of IonQ's ownership of the company, because it omits the 19,744,585 Class B shares; the source of the confusion is that SEC filings report percentage of class rather than overall percentage. IonQ participated in the PIPE and so was not diluted by it, and has been diluted only slightly by warrant exercises — under 5%. [FACT] 

Net effect: this is a more diffuse ownership picture than the headline percentages suggest, and the exact current figures depend on a denominator convention that isn't consistently disclosed across filings. Neither IonQ nor Peak XV should be described as holding a fixed, current stake without a fresher filing and a stated denominator.

Two things are worth holding alongside the dilution arithmetic. First, the governance fact that actually controls outcomes is unchanged: Fitzsimons retains 65.0% of voting power through Class B shares carrying fixed 3x weight, so no amount of Class A issuance shifts control of the company. Second, the same PIPE, warrant exercises, and share issuances that mechanically diluted every percentage stake are what delivered the $113.3M cash balance, funded Ember-1, and made the $35M AQ256 commitment affordable without debt. The dilution and the capability were the same transaction. [ARG] 

That does not make dilution free — Section 18 sizes the remaining overhang precisely, and future raises would compound it. But describing the percentage decline purely as erosion misses that the capital bought the assets this thesis rests on.

20. Five Targeted Commercial Sectors

Five verticals recur throughout Horizon's materials: chemistry/materials science, healthcare/pharmaceuticals, financial services, optimization/logistics, and national security/defense. An important framing correction: per management, these are the main sectors from which Horizon sees demand in inbound access requests, rather than a focus the company imposes. They describe where interest is arriving, not a chosen go-to-market target list.

Horizon's 20-F discloses a concrete inbound-demand figure: as of the filing (March 2026), the company had received early-access interest from more than 40 major corporations, 80 universities, 10 quantum software companies, and 15 national labs, government agencies, and research organizations. [FACT] 

This is a genuinely useful correction to how this report has framed commercial demand. "No evidence of demand" would be too strong a claim — there is disclosed inbound interest at a real, named scale. What remains true is that inbound interest is not the same as revenue, a signed contract, or even a completed pilot; on the Q2 earnings call, Fitzsimons said the company is deliberately not converting this interest into proof-of-concept engagements, calling that approach a distraction from the technical roadmap. So there is real evidence of demand-side interest, but it is interest the company has chosen not to monetize yet — a more precise and more favorable picture than "no demand exists," but still short of "the company has commercial traction."

No 6-K, press release, or earnings-call transcript I've reviewed through Q2 2026 identifies a signed commercial contract, named enterprise customer, or recognized revenue in any of the five sectors. The sector narrative remains a strategic thesis, not a demonstrated pipeline — and the $0 revenue figures in Section 17 are the clearest evidence of that gap. [FACT] 

I was not able to locate additional primary-source detail (pilot programs, named design partners, grant awards) specific to any of the five sectors beyond what the original report already described. If sector-specific evidence exists, it has not surfaced in company disclosures to date; this section should be revisited if and when it does, rather than expanded on the basis of general market narratives about quantum's applicability to these industries.

21. The Future Market: Where Value Accrues and What Horizon Is Positioned to Capture

This is the most important forward-looking section in the report, and it is deliberately not built on a market-size estimate. Total-addressable-market figures for quantum computing are unfalsifiable, vary by an order of magnitude between publishers, and prove nothing. What follows instead is a structural argument: who the customer becomes, why the bottleneck moves, where value has historically accrued in comparable computing stacks, and what specifically Horizon is positioned to collect at each stage.

21.1 The Bottleneck Shift: From Qubits to Programmers

Horizon's stated market premise: only a few hundred specialists worldwide currently have the skills to formulate quantum algorithms. Set that against the global professional software developer population, conventionally estimated in the tens of millions. [FACT] 

That ratio is the entire thesis in one line. Today the binding constraint on quantum computing's commercial value is hardware — qubit counts, fidelities, coherence. Every credible hardware roadmap, including IonQ's move to 256 chip-based trapped-ion qubits, is engineered to relax that constraint. When it relaxes far enough, the constraint does not disappear; it moves. It moves to the number of people who can express a valuable problem in a form a quantum machine can execute. At that point the industry's growth rate is set by the size of the programmer population, not the size of the machines — and a five-order-of-magnitude gap between hundreds of specialists and tens of millions of developers is the largest single unexploited leverage point in the sector. [ARG] 

This is why the abstraction ladder in Section 6 is a market strategy rather than a developer convenience. Every layer added — Hydrogen to Helium to Beryllium to automatic synthesis from classical code — moves the required skill level down and the addressable population up by orders of magnitude. A company that succeeds in compiling ordinary C or Python into quantum-accelerated applications has not built a better tool for quantum specialists. It has converted every competent classical programmer into a potential quantum user without their needing to learn quantum mechanics.

21.2 Where Value Accrues in a Computing Stack

The relevant historical pattern is not that software beats hardware — it is that in a new computing paradigm, durable margin concentrates in whichever layer becomes the thing developers build against and cannot easily leave. Machines get commoditised; the programming interface, once an ecosystem accumulates on top of it, does not. [ARG] 

The sharpest modern instance is NVIDIA's CUDA. NVIDIA's competitive position in accelerated computing is not primarily explained by transistor-level superiority over every rival; it is explained by the fact that roughly two decades of libraries, frameworks, teaching material, and trained engineers were built against CUDA specifically. Competing silicon has repeatedly been benchmarked as competitive on raw specifications and has repeatedly failed to displace the incumbent, because displacing it would require the ecosystem to rewrite itself. The lesson relevant here is the timeline as much as the outcome: CUDA launched in 2006, and the position it created did not become obviously decisive until the deep-learning era roughly a decade later. Platform layers look unmonetised for a long time and then do not.

Applied to quantum: the analogous position is the layer that developers write against when quantum computing becomes broadly useful. Today that layer is genuinely unsettled. Qiskit has the largest community but is coupled to IBM's roadmap and its hardware assumptions. PennyLane is strong in quantum machine learning and hardware-agnostic, but is a Python framework for expressing quantum circuits rather than a general-purpose language stack. Classiq addresses synthesis but not the full compilation-to-deployment chain. No incumbent yet owns the position that CUDA occupies in accelerated computing, and Horizon is building directly at it with Turing-complete languages, classical-code ingestion, and hardware-agnostic output. [ARG] 

The bull case does not require Horizon to be better at circuit optimisation than IBM. It requires the industry to eventually need a general-purpose programming layer above circuits — which is Fitzsimons' explicit thesis that quantum computing must reinvent roughly eighty years of computer science (Section 5) — and for Horizon to have built it first while nobody else was contesting that ground.

21.3 Three Phases of Market Development

Phase	Characteristics	Who the customer is	What Horizon captures
Phase 1 — Research and enablement (now to first advantage)	No commercial quantum advantage. Buyers are hardware vendors, national labs, and universities. Value is in enabling other people's machines to be usable.	Hardware manufacturers first — Horizon's own stated near-term go-to-market focus — plus research institutions.	Deep integration into vendor stacks, testbed co-design, standing in the research community. Little revenue by design (Section 9).
Phase 2 — First advantage in narrow domains	Specific problems become faster on quantum hardware; chemistry is the most cited candidate and the one Fitzsimons named for the AQ256. Buyers are sophisticated enterprises with in-house scientific computing.	Pharma, materials, and specialised finance groups with existing HPC budgets and hard problems.	Usage-based metering on production workloads (Section 9); the internal applications team's three high-value problems become reference implementations.
Phase 3 — Broad deployment	Quantum acceleration becomes a routine option in the developer toolchain, invoked through APIs by developers who never learn quantum mechanics.	The general enterprise software market — every organisation with a computationally hard problem.	Platform position: metered usage at scale, plus accumulated switching cost from Beryllium libraries and code reuse.

The critical observation about this sequence is that Horizon's current strategy is legible only if you read it as Phase 1 positioning for Phase 3 economics. Refusing services revenue, focusing go-to-market on hardware manufacturers rather than enterprises, building an in-house applications team instead of a sales team, anchoring the field's leading theoretical conference, targeting the ideal machine before the real one — each of these is a poor Phase 1 revenue decision and a strong Phase 3 positioning decision. Management is explicitly optimising for a market that does not exist yet. [ARG] 

21.4 Sector Demand and the Compilation Capabilities That Serve It

The five verticals Horizon targets are conventional in quantum marketing. What is not conventional is connecting each one to a specific capability in the Triple Alpha stack rather than to a generic claim that quantum is good at hard problems.

Sector	Why the demand is real and durable	The specific Triple Alpha capability that serves it
Chemistry and materials	Simulating quantum-mechanical systems is the one application where quantum computers have a theoretically clean advantage, because the problem and the machine share a substrate. Fitzsimons named chemistry as the domain where the AQ256 should sit "on the cusp of an advantage."	Deep circuit depth on high-fidelity trapped-ion hardware (AQ256), plus c2q compilation (Section 6.3) of existing classical chemistry routines into quantum subroutines rather than rewriting them by hand.
Pharmaceuticals and life sciences	Binding-affinity and molecular-dynamics problems are chemistry problems with a large commercial multiplier; the industry already funds enormous classical compute and would adopt acceleration without needing to be convinced quantum is interesting.	Same chemistry pathway, plus API deployment (Section 6.4) so quantum acceleration can be called from existing computational-chemistry pipelines without replacing them.
Financial services	Derivative pricing, risk, and portfolio optimisation are Monte Carlo–heavy workloads with direct, measurable P&L consequences and buyers who already pay for marginal speed.	Turing-complete control flow and indefinite loops — Monte Carlo and iterative optimisation cannot be expressed as static circuits, which is precisely the constraint Horizon's languages remove.
Optimisation and logistics	Combinatorial routing and scheduling at global scale remain intractable classically; the buyers are large industrials with continuous, quantifiable cost exposure.	Mid-circuit measurement with classical feedforward and hybrid execution — adaptive algorithms that branch on intermediate results, unavailable in static-circuit frameworks.
National security and defence	Sovereign programmes fund capability ahead of commercial proof and value portability across approved hardware. Horizon's Singapore/Ireland/US footprint and the founder's prior AFOSR funding are relevant here.	Hardware-agnostic compilation across 30+ backends: algorithms remain portable across whichever approved machines a programme is permitted to use, without rewriting.

21.5 What Would Have to Be True

Stating the load-bearing assumptions explicitly, because a market thesis that cannot be attacked is not a thesis:

• Quantum hardware must continue improving to the point where some commercially valuable problem runs materially better than on classical machines. If this never happens, no software layer matters. This is the sector-wide risk Horizon shares with every quantum company, and it is unhedgeable.
• The industry must need a general-purpose programming layer above circuits. If quantum computing turns out to be useful only for a handful of narrow, hand-optimised algorithms that specialists write once and reuse, the developer-population argument collapses and the market is small regardless of who serves it.
• Enterprises must be willing to license a third-party layer rather than use whatever their hardware vendor ships free. This is the open-source risk named in Section 27, and it is the assumption with the least supporting evidence today.
• Horizon must convert a first-mover technical position into a commercial one. History offers many examples of the company that built the right layer first not being the company that monetised it.

Weighing those honestly: the first assumption is the sector's, not Horizon's. The second is the one Fitzsimons has been arguing for two decades and has a granted patent on. The third and fourth are the real ones, and they are execution and business-model risks rather than technology risks — which is the more tractable category of the two, and the reason I hold the position I do. [ARG] 

22. Chemistry and the Health Sciences: The Deepest Opportunity

Fitzsimons named chemistry as the domain where the AQ256 should approach advantage. This section examines that claim seriously — what the underlying computational problem is, why classical methods fail at it, what the resource requirements actually are, where Horizon sits in the value chain, and what the commercial opportunity in health sciences looks like if the pathway works. It is the longest forward-looking analysis in the report because it is, in my assessment, the highest-value one.

22.1 The Problem Classical Chemistry Cannot Solve

The class of chemistry where quantum computers have a theoretically clean advantage is strongly correlated electronic structure — molecules whose electrons interact so strongly that their behaviour cannot be approximated as independent particles. Standard workhorse methods (density functional theory, coupled-cluster) rely on approximations that break down precisely in these systems, which are described as having multi-reference character. [FACT] 

The reason quantum computers help is structural rather than incidental: the difficulty of simulating these systems classically comes from the exponential growth of the quantum state space with the number of correlated electrons. A quantum computer represents that state space natively. This is the one major application area where the problem and the machine share a substrate, and it is why chemistry is consistently the first credible commercial target rather than a marketing preference.

22.2 Two Molecules That Define the Frontier

Cytochrome P450 (CYP) is a family of heme enzymes responsible for metabolising roughly 70–80% of all marketed drugs, with the CYP3A4 isoform alone handling approximately half. Predicting how a drug candidate is metabolised — and whether metabolism produces toxic intermediates — is among the most expensive determinations in pharmaceutical development. P450 is poorly handled by current computational methods because the heme active site involves extreme spin-state fluctuations requiring multi-reference treatment. In computer-aided drug design, P450s are frequently treated as anti-targets: interactions with them cause drug-drug interactions and accelerated clearance of the active compound. [FACT] 

FeMoco, the iron-molybdenum cofactor of nitrogenase, is the biological catalyst that fixes atmospheric nitrogen at ambient conditions — the process the industrial Haber-Bosch route replicates at enormous energy cost. It has resisted decades of biochemical study and is the canonical benchmark problem for quantum chemistry resource estimation. [FACT] 

Published resource estimates set the scale honestly. Goings et al. (PNAS, 2022) estimated P450 electronic structure at approximately 4,900 logical qubits, on the order of 10⁹ Toffoli gates, and roughly 73 hours of runtime. A 2021 Google benchmark placed FeMoco at approximately 2.7 million physical qubits; Alice & Bob's 2025 cat-qubit study reported a 27-fold reduction to approximately 99,000 physical qubits under equivalent error and runtime assumptions. Independent analysis places first industrially relevant FeMoco simulation in the 2033–2036 window. [FACT] 

The unavoidable arithmetic: the AQ256 is a 256-physical-qubit machine. P450 at industrial relevance requires roughly 4,900 logical qubits, each of which requires many physical qubits under error correction. The Dublin system is three to four orders of magnitude short of solving P450 outright, and no honest reading of Fitzsimons' "cusp of an advantage for certain chemistry problems" should be stretched to imply otherwise. The qualifier "certain" is doing essential work — the target is small, carefully chosen model systems, not industrial enzymes. [ARG] 

This is the most important calibration in the section, and stating it plainly is what makes the rest of the argument credible. Anyone who reads a 256-qubit trapped-ion installation as the machine that will simulate drug metabolism has misunderstood the timeline by a decade.

22.3 Why Horizon Still Matters — And Why It May Matter Most

Reframe the question from "can Horizon's hardware solve P450" to "who builds the software layer that will run these calculations when hardware arrives." Every published resource estimate above assumes a compiled quantum program: qubitized quantum walks, phase estimation, tensor factorisations, Toffoli-gate budgets, and active-volume compilation. Those are compiler outputs. The efficiency of the compilation is not a detail — Alice & Bob's 27-fold improvement came substantially from better algorithmic and architectural choices, not from more qubits. [ARG] 

That is precisely Horizon's layer. Improvements in compilation efficiency translate directly into reduced qubit counts and runtimes, which translates into pulling the commercial timeline forward. A software company that reduces the resource requirement for P450 by a further order of magnitude has done something economically equivalent to a decade of hardware progress — without building a machine. [ARG] 

The partnership structure is unusually well-matched to this. Alice & Bob — the company that produced the cat-qubit resource estimates for P450 and FeMoco — is a Tier 3 integration partner of Horizon's, providing fault-tolerant emulators (Section 12). Horizon therefore has a working relationship with the error-corrected architecture on which these calculations will eventually run, while operating trapped-ion hardware with the highest current gate fidelities in Dublin, while compiling across both. [FACT] 

The c2q capability (Section 6.3) has a specific and underappreciated role here. Computational chemistry groups in pharma have decades of validated classical code — geometry optimisation, integral evaluation, post-processing pipelines — that is scientifically load-bearing and expensive to reproduce. Horizon's toolchain compiles existing C and C++ functions into partial quantum circuits callable as subroutines. That means a pharmaceutical company adopting quantum acceleration does not have to rewrite its computational chemistry stack; it can compile the classical components it already trusts and call quantum routines where they help. For a regulated industry with validated pipelines, the ability to preserve existing code is not a convenience — it is often the difference between adoption and refusal.

22.4 The Commercial Pathway in Health Sciences

Application	The computational problem	Commercial consequence if solved	Horizon's role
Drug metabolism and toxicity prediction (P450/ADMET)	Multi-reference electronic structure at heme active sites; spin-state energetics that classical DFT handles poorly	Metabolic liability and toxic-intermediate formation are among the leading causes of late-stage clinical failure. Moving that determination earlier reduces the cost of failure by orders of magnitude	Compiler efficiency determines when this becomes feasible; c2q preserves existing ADMET pipelines
Structure-based drug design against difficult targets	Binding affinity where metal centres, radical intermediates, or strong correlation defeat classical approximation	Access to target classes currently considered undruggable by computational methods	Hardware-agnostic compilation lets a pharma group run the same algorithm across whichever machine is best or available
Enzyme and biocatalyst engineering	Reaction mechanism and transition-state energetics in metalloenzymes	Greener manufacturing routes for existing drugs; novel biocatalysts for synthesis steps that are currently expensive or low-yield	Same chemistry pathway; benefits directly from trapped-ion fidelity in Dublin
Photodynamic and radiopharmaceutical agents	Excited-state electronic structure, which is harder classically than ground state	New therapeutic modalities in oncology where selectivity is determined by excited-state behaviour	Excited-state algorithms are compilation-intensive; the layer that compiles them efficiently captures the value
Materials for delivery and devices	Polymer, lipid, and surface chemistry with correlated electronic effects	Improved delivery vehicles, biomaterials, and diagnostic sensing surfaces	Shared chemistry stack; the same compiled primitives serve materials and pharma

The economic logic that makes this the highest-value vertical: pharmaceutical R&D already spends heavily on computation and already accepts long timelines and high failure rates. Buyers do not need to be persuaded that computational chemistry is valuable — they are already paying for it. The adoption question is therefore narrower and more tractable than in most quantum verticals: not whether to fund computational chemistry, but whether quantum-accelerated methods produce answers that classical methods cannot, at a price justified by the value of the answer. That is a question quantum advantage in chemistry, if demonstrated, answers directly. [ARG] 

Independent analysis is appropriately restrained about scope: the quantum edge applies to roughly the hardest 5–10% of chemistry problems, with most calculations continuing to run classically. That fraction, however, includes some of the highest-value calculations in pharmaceutical, chemical, and energy R&D. [FACT] 

A narrow slice of the highest-value problems is exactly the right shape of market for a metered software layer. Horizon does not need to serve all of computational chemistry. It needs to be the layer through which the hardest and most valuable calculations are expressed and executed — and to meter that usage (Section 9).

22.5 Honest Assessment of This Pathway

• The timeline is long. Industrially relevant FeMoco simulation is independently placed in the 2033–2036 window; P450 at full scale requires fault-tolerant machines that do not exist. Nothing in this section is a 2027 revenue thesis.
• Horizon has published no chemistry benchmarks of its own. Every resource estimate cited here is third-party work by Google, Alice & Bob, and academic groups. Horizon's contribution to reducing these requirements is potential, not demonstrated.
• Competing approaches exist. Quantum-centric supercomputing hybrids, improved classical methods such as DMRG, and better tensor factorisations all move the advantage boundary — and some of that movement favours classical incumbents rather than quantum challengers.
• No pharmaceutical customer has been named, no pilot disclosed, and no revenue recognised in this vertical or any other. The internal applications team's three unnamed high-value problems across three industries may or may not include chemistry.

Weighing it: this is the most credible large commercial opportunity in the report and simultaneously the furthest from realisation. What makes it a genuine asset rather than a hope is that Horizon does not need to win a hardware race to participate — it needs the compilation layer for these calculations to be valuable and to be theirs. On the evidence, that layer is exactly what they are building, they hold a granted patent covering its core mechanism, and they have partnerships spanning both the trapped-ion hardware nearest to early advantage and the cat-qubit architecture nearest to the fault-tolerant era. The pathway is long, but the positioning along it is unusually good. [ARG] 

23. The Competitive Field

This section maps every credible participant in or adjacent to Horizon's layer, states what each does well, and assesses honestly where Horizon is behind. Section 24 then argues why the platform position nonetheless remains open. Readers should be able to check that argument against a complete field rather than a selected one.

23.1 Mapping the Stack

Quantum software is not one market. It divides into layers that are frequently conflated, and most companies described as Horizon's competitors occupy a different layer entirely. [ARG] 

Layer	What it does	Principal participants	Relationship to Horizon
Cloud access platforms	Route jobs to hardware; billing, queuing, identity	IBM Quantum, Microsoft Azure Quantum, AWS Braket, Strangeworks	Distribution channels, not competitors. Horizon compiles to backends these platforms also expose
SDKs / circuit frameworks	Construct and submit quantum circuits from Python	Qiskit (IBM), Cirq (Google), PennyLane (Xanadu), tket (Quantinuum)	The incumbents in developer mindshare — and the layer Horizon is arguing must be superseded
Synthesis and high-level design	Generate optimised circuits from higher-level functional descriptions	Classiq	The closest genuine competitor
Control and error suppression	Pulse-level calibration, noise suppression, hardware tuning	Q-CTRL, Quantum Machines	Adjacent. Quantum Machines is Horizon's partner, not rival
QEC decoding middleware	Real-time decoding of error syndromes for fault tolerance	Riverlane	Adjacent and complementary — a decoder needs a language to be invoked from
Application libraries	Domain algorithms in chemistry, finance, optimisation	Algorithmiq, Phasecraft, QunaSys, Multiverse, QC Ware, 1QBit	Potential customers or partners; they build on layers like Horizon's
General-purpose programming layer	Turing-complete languages, classical-code ingestion, compilation to any backend, metered deployment	Horizon Quantum	The layer this report argues is unclaimed

23.2 The Closest Competitor: Classiq

Classiq Technologies, founded 2020 in Tel Aviv by Nir Minerbi and Amir Naveh, is a Series C company that has raised approximately $180–200 million. Its platform lets developers work at a higher level of abstraction than gate-level programming, automating the creation of optimised quantum circuits from functional models. It has announced commercial engagements including work with Rolls-Royce on quantum computational fluid dynamics and a multi-million-euro quantum hub with TEA TEK Group in Naples. Third-party competitive databases list Horizon Quantum as Classiq's single closest competitor. [FACT] 

Classiq is a real competitor and the report should say so plainly. It is better capitalised than Horizon, has been more commercially visible, and attacks the same fundamental problem: raising the abstraction level above hand-built circuits. [ARG] 

Where the approaches diverge: Classiq synthesises circuits from functional models — the developer describes what the circuit should accomplish and the platform generates an optimised circuit. Horizon's languages are Turing-complete programs that compile to executables, with control flow resolved at runtime rather than circuit structure fixed at synthesis. Classiq's output is a better circuit; Horizon's output is a running program. The distinction matters most where a computation must branch on a mid-circuit result, loop an indeterminate number of times, or call a classical subroutine coherently — capabilities described in Section 24.3.1 that a synthesis-to-circuit model does not natively provide. Horizon additionally ingests existing C and C++ and operates its own hardware; Classiq does neither. [ARG] 

An honest reader should weigh the possibility that Classiq's approach proves sufficient. If quantum computing's commercially valuable workloads turn out to be expressible as optimised static circuits — which is true of much variational and near-term work — then synthesis is the right abstraction and Turing-completeness is an elegant answer to a question the market did not ask. Our judgment, set out in Section 24, is that fault tolerance makes that outcome unlikely, because error correction is inherently adaptive. But it is a judgment, not a fact, and it is the crux on which Horizon-versus-Classiq turns.

23.3 Adjacent Layers Frequently Mistaken for Competitors

Riverlane has raised approximately $195 million and is, in the assessment of independent observers, the most focused company in the world on quantum error correction decoding. Its middleware processes error-syndrome data between hardware and applications, with partnerships across Rigetti, IQM, and the UK National Quantum Computing Centre; with Rigetti it demonstrated real-time decoding under one microsecond on Ankaa-2 in October 2024. Founder Steve Brierley is a former Cambridge quantum information lecturer. [FACT] 

Riverlane is not competing with Horizon — it is building a component that a fault-tolerant program must invoke. A decoder must be called from somewhere, conditioned on somewhere, and integrated into a program's control flow. Riverlane's success arguably increases the value of a Turing-complete layer capable of expressing the adaptive computation its decoder enables. [ARG] 

Q-CTRL has raised approximately $190 million, including a $113 million Series B in October 2024, and specialises in error suppression and hardware control — improving what a given machine can do before any program is written. It has partnered with Classiq on end-to-end development environments. Quantum Machines, which has raised over $280 million in control electronics, is Horizon's collaboration partner on Ember-1 calibration rather than a competitor. [FACT] 

Strangeworks provides a multi-vendor access platform letting enterprises reach hardware across IonQ, IBM, Rigetti and others without committing to one stack. It is an access and orchestration layer — solving procurement and routing rather than programming. Notably it is an IBM Ventures portfolio company, alongside QEDMA and QunaSys, which illustrates a strategic pattern: IBM invests in companies that complement Qiskit and drive traffic to IBM Quantum. [FACT] 

23.4 Where Horizon Is Genuinely Behind

A competitive section that finds no disadvantages is not a competitive section. Four, stated plainly:

Capital. Horizon raised roughly $120 million gross at the de-SPAC, plus a further $28.7 million from warrant exercises — proceeds which are reflected in the $113.3 million held at June 30, 2026, and which have continued to accrue since, strengthening the balance sheet further. Total capital raised is therefore closer to $149 million than the de-SPAC figure alone suggests. Classiq, Riverlane, and Q-CTRL have each raised more — approximately $180–200 million, $195 million, and $190 million respectively — and Quantum Machines over $280 million. Horizon is not the best-funded company in its own neighbourhood, and it is simultaneously the one carrying a $35 million hardware commitment. Its capital efficiency is a strength of the story only for as long as it holds. [FACT] 

Commercial visibility. Classiq has announced named engagements with Rolls-Royce and a regional quantum hub in Naples. Riverlane has hardware partnerships with named vendors and a demonstrated technical result with Rigetti. Horizon has announced no named commercial customer. Whatever the strategic reasoning in Section 9, competitors are accumulating public reference points and Horizon is not. [ARG] 

Developer mindshare. Qiskit has the deepest tutorial library, the largest community, and the most trained users; PennyLane dominates quantum machine learning. Horizon's Beryllium entered early access at the end of Q2 2026 with a small group. On the metric that historically decides platform outcomes — how many developers have already learned your tool — Horizon is far behind, and that is precisely the metric the CUDA analogy says matters most. [ARG] 

Price. Qiskit, Cirq, PennyLane, and tket are free. Horizon must demonstrate that a proprietary layer delivers enough value to displace tools that cost nothing. This remains the single least-evidenced assumption in this report. [ARG] 

23.5 The Scenario That Would Break the Thesis

The most serious competitive risk is not Classiq. It is IBM, Google, or Microsoft deciding to build a genuinely general-purpose, hardware-agnostic quantum programming layer and giving it away. [ARG] 

Each has the engineering capability and the capital. IBM has already added dynamic-circuit capability to Qiskit, which is a step in this direction. Microsoft has decades of compiler and language expertise and no meaningful quantum hardware position of its own to protect. Any of them could commit resources far beyond Horizon's.

Three reasons we assess this as less likely than it first appears. First, incentive: IBM's and Google's tooling exists to sell their own machines and cloud time; a truly hardware-agnostic layer would make their customers indifferent to their hardware, which is the incumbent's dilemma that historically transfers platform positions to specialist outsiders. Microsoft is the genuine exception here and the one to watch. Second, architecture: retrofitting Turing-complete semantics onto a circuit-centric framework with a large installed base is harder than building from that premise, as IBM's incremental dynamic-circuit additions illustrate. Third, hardware access: developing real-time execution requires owning and instrumenting machines, which IBM can do for its own architecture but not across modalities without buying competitors' hardware. [ARG] 

If a major incumbent shipped a credible free general-purpose layer across multiple modalities, the central argument of this report would be substantially weakened. Section 29 tracks this as a formal trigger, and we would revise rather than defend.

23.6 How Management Frames the Competition

The preceding sections assess the field from the outside. It is worth setting alongside them how Horizon's own management characterises the competitive picture, in their words rather than ours — both because it is a materially more optimistic frame than a disadvantages table conveys, and because readers can judge for themselves whether it is insight or convenience.

Asked directly about competitive positioning on the Q2 2026 earnings call, Fitzsimons reframed the question rather than answering it on its own terms. His formulation: "the real competition is classical compute" — not other quantum software companies. His reasoning was that the addressable market for any quantum software vendor remains tiny until quantum systems broadly outperform classical alternatives, so the contest that matters is quantum versus classical, not vendor versus vendor. [FACT] 

"The real competition is classical compute."  — Dr. Joe Fitzsimons, Q2 2026 earnings call, August 4, 2026 [FACT]

He was direct in the same exchange that Qiskit, Classiq, and others represent real alternative approaches — not dismissing them — while arguing that Horizon's attempt to automatically accelerate ordinary code, rather than requiring circuit-level programming or supplying point-solution libraries, is a materially different approach rather than a better version of the same one. [FACT] 

On hardware modality, his stated position is equally non-adversarial: "I cannot pick a winner... I do not know which hardware platform is going to win," after twenty-two years in the field. A company whose commercial thesis depends on hardware-agnosticism benefits from every modality advancing, and its CEO's stated uncertainty is consistent with that rather than a hedge against it. [FACT] 

Read at face value, this is a positive-sum framing of the sector, and it is defensible on the structure of the market rather than merely as optimism. If the total commercial market for quantum software today is close to zero — which the revenue figures across the entire category suggest — then competitors are not dividing a fixed pool. Every advance by IBM in hardware, by Riverlane in decoding, by Q-CTRL in error suppression, and by Classiq in developer accessibility expands the set of problems any quantum software layer can address. In a market this early, a rival's success is closer to market creation than to share capture. [ARG] 

The layer map in Section 23.1 supports this reading concretely. Riverlane's decoder must be invoked from a program and conditioned within its control flow. Q-CTRL's error suppression improves the fidelity of hardware that Horizon's compiler targets. Quantum Machines' control electronics are the substrate for Ember-1's real-time execution — an outright partnership. The application-layer companies in chemistry and finance need a programming layer beneath them. Of the participants commonly listed as Horizon's competitors, the majority are building components that a general-purpose layer would call, integrate, or serve. [ARG] 

The honest counterweight, stated so the frame is not simply accepted: "the real competition is classical compute" is also a convenient thing to say when a company has no revenue and several named competitors have disclosed commercial engagements. It relocates the scoreboard to a contest nobody is currently winning. And a positive-sum market eventually becomes a zero-sum one — the moment commercial quantum workloads exist at scale, the question of whose layer they run on becomes precisely a share question, and Horizon's mindshare disadvantage in Section 23.4 becomes acutely relevant at exactly that moment.

23.7 Our Own Read: Why the Competitive Picture Is Better Than the Deficits Suggest

Weighing management's framing against the disadvantages catalogued above, our judgment is that the competitive position is stronger than a straight capital-and-customers comparison implies — for four reasons. [ARG] 

First, most of the field is complementary rather than competitive. Once the stack is mapped by layer, the list of companies genuinely contesting Horizon's position reduces to essentially one: Classiq. Riverlane, Q-CTRL, Quantum Machines, Strangeworks, and the application-library companies occupy adjacent layers, and several of them succeeding makes Horizon's layer more valuable rather than less. A competitive set of one, in a category this large, is not a crowded field. [ARG] 

Second, the capital gap is smaller than it appears in context. Classiq, Riverlane, and Q-CTRL have raised more than Horizon — but Horizon holds $113.3 million against roughly $9.3 million of half-year operating burn, with its decisive hardware milestone already contracted at a fixed price. The relevant question is not who raised the most but who is funded through their proving event, and Horizon is. Capital advantages matter when the constraint is money; Horizon's constraints are time and physics, which additional capital does not obviously relieve. [ARG] 

Third, the mindshare deficit is measured against the wrong end state. Qiskit's community is large because Qiskit is the default way to write circuits — and the entire premise of Horizon's architecture is that circuits are the wrong abstraction for the fault-tolerant era. Being behind on adoption of an abstraction you believe will be superseded is a different situation from being behind on the abstraction that wins. This may of course be wrong; if circuits remain the durable interface, the deficit is real and probably decisive. But it is a bet on which abstraction endures, not a simple gap. [ARG] 

Fourth, no competitor is assembling the same combination. Classiq raises abstraction but does not ingest classical code, own hardware, or hold a patent on the compilation mechanism. Riverlane owns error correction but no language. Q-CTRL owns control but no programming layer. Strangeworks owns access but builds nothing. Each is building a strong single capability; Horizon is building the layer into which those capabilities resolve. Whether that integration proves to be the winning strategy or an overextension by a 57-person company is the genuine question — but it is not a question anyone else in the field is currently forcing. [ARG] 

Our conclusion on the competitive field is therefore narrower and more confident than the deficits table alone would support: Horizon faces one real competitor in its layer, is funded past its decisive milestone, and is behind principally on an abstraction it is arguing against. That is a substantially better competitive position than a company with less capital and fewer customers usually occupies — and it is why Section 24 argues the platform position remains open despite everything catalogued here.

23.8 Summary Judgment

Horizon is behind on capital, commercial references, and developer mindshare, and it charges for something competitors give away. Those are real disadvantages and we do not discount them. What it holds that no competitor holds is the combination detailed in Section 24.3 — Turing-complete languages, classical-code ingestion, owned multi-modal hardware, a granted patent, and founders who originated the underlying science. The competitive question is whether that combination is worth more over a decade than a funding lead and a head start on community. Our judgment is that it is, because the disadvantages are the kind that capital and time can close, and the advantages are the kind that they cannot. [ARG] 

24. The Platform Thesis: Why This Position Is Rare

24.1 The Pattern That Repeats

Every computing paradigm shift of the last fifty years has resolved the same way. A new class of machine arrives. Early value accrues to whoever builds the machines. Then the machines commoditize — and the durable economics migrate, permanently, to whoever owns the layer that developers write against.

The canonical case is the personal computer. In 1981 IBM held every apparent advantage: the brand, the manufacturing, the enterprise relationships, the capital. It built the machine and licensed the operating system from a company of roughly forty people, retaining hardware and conceding the right to license MS-DOS to other manufacturers. Within a decade the hardware was a commodity assembled by a dozen interchangeable vendors competing on price, and the economics of personal computing belonged to the company that owned the layer applications were written against.

Microsoft's position was never primarily technical. Competing operating systems were, at various points, credibly better. What Microsoft owned was the accumulated fact that the world's software had been written against its interface — and every additional application deepened the reason for the next developer to target it, and the next user to require it. The moat was not in the code. It was in everything built on top of the code.

The pattern recurs with consistency: mainframes to IBM's integrated stack, the PC to Microsoft, mobile to iOS and Android, accelerated computing to NVIDIA's CUDA. In each case the layer that captured durable rents was the one developers could not practically leave. [ARG] 

NVIDIA is the most instructive modern instance because it is the closest structural analogue. Its position in AI compute is not primarily explained by transistor-level superiority; competing silicon has repeatedly benchmarked competitively and repeatedly failed to displace it, because displacement would require roughly two decades of libraries, frameworks, teaching materials, and trained engineers to rewrite themselves against a different interface. CUDA launched in 2006. The position it created did not look decisive until the deep-learning era, roughly a decade later. Platform layers look unmonetized for a long time, and then they do not. [ARG] 

24.2 The Layer Is Currently Unclaimed

"Our mission is to unlock broad quantum advantage by building software infrastructure that empowers developers to use quantum computing to solve the world's toughest computational problems."  — Horizon Quantum, stated mission, horizonquantum.com [FACT]

Here is the observation on which this entire report turns: in quantum computing, that layer does not yet have an owner. [ARG] 

This is genuinely unusual. By the time most paradigms are visible enough to invest in, the abstraction layer has already been claimed and the entry window has closed. Quantum computing is in the rare interval where the machines are advancing rapidly, enormous capital is flowing into hardware, and the position that historically captures the most durable value is still contested.

Consider who might have claimed it and why they have not:

• IBM's Qiskit has the largest community in quantum software. It is also free, open-source, and structurally coupled to IBM's own superconducting roadmap and hardware assumptions. Qiskit is not a business competing for the platform layer; it is demand generation for IBM hardware. That is a coherent strategy, but a different one.
• Xanadu's PennyLane is hardware-agnostic and genuinely strong in quantum machine learning. It is a Python framework for expressing differentiable quantum circuits — a library, not a language stack — and likewise free and open-source, serving Xanadu's photonic hardware ambitions.
• Classiq addresses algorithm synthesis but not the full chain from ordinary classical code through compilation to deployed, metered execution.
• Hardware vendors generally ship SDKs to make their own machines usable. None has an incentive to build the layer that makes a customer indifferent between their machine and a competitor's.

The structural reason the layer remains unclaimed is that claiming it requires solving a problem almost nobody else is attempting. Everyone else is building better tools for describing quantum circuits. Horizon is attempting to make circuits an implementation detail — to build the general-purpose programming layer that sits above them. Fitzsimons has stated the scope directly: useful quantum computing requires reinventing roughly eighty years of computer science, including kernels, dynamic memory allocation, I/O, and networking. That is not a more ambitious version of what competitors are doing. It is a different undertaking. [ARG] 

24.3 What Horizon Does That Competitors Do Not

Six capabilities. Each is examined for what it actually is, why it is technically hard, what competitors do instead, and why the gap persists.

24.3.1 Turing-Complete Languages, Not Circuit Descriptions

Nearly all quantum software today produces a circuit: a fixed, directed sequence of gates, fully determined before execution begins, applied to qubits, followed by measurement at the end. Variational frameworks vary the parameters of that circuit between runs, but the structure — which gate acts on which qubit, in what order — is fixed at compile time. [FACT] 

Helium and Hydrogen are Turing-complete. They express conditionals evaluated during execution on mid-circuit measurement outcomes; definite and indefinite loops, including loops whose iteration count cannot be known in advance; subroutine calls; concurrent classical-quantum computation; and external I/O. Hydrogen represents control flow as flow-chart blocks containing instruction lists, joined by conditional jumps, with Horizon's execution infrastructure supplying branch information at runtime so programs respond dynamically. [FACT] 

Why this is genuinely hard: branching mid-computation requires measuring a qubit, transmitting the result to a classical processor, evaluating a condition, and returning an instruction — all within the coherence time of the qubits still holding the computation. On superconducting hardware that budget is microseconds. This is why the latency constraint of roughly one nanosecond per foot is not an implementation detail: at these timescales, physical distance between the classical controller and the quantum processor is a hard limit on what the language can express. A language feature is only real if the runtime beneath it can execute it in time. [ARG] 

The consequence competitors cannot escape: quantum error correction — the entire fault-tolerant era the industry is building toward — requires mid-circuit measurement with classical feedforward. Syndrome extraction is measuring ancilla qubits mid-computation and conditionally applying corrections based on the result. A framework that can only express static circuits is structurally incapable of expressing error-corrected computation in its native form. Every roadmap in quantum computing terminates in fault tolerance, and the dominant programming abstractions cannot describe it. Horizon built the languages that can before the hardware needing them arrived. [ARG] 

IBM has added dynamic-circuit capability to Qiskit, but it is an extension to a circuit-centric model on IBM's own hardware, not a general-purpose language stack. PennyLane is a Python library for constructing differentiable quantum circuits — powerful for variational and machine-learning workloads, structurally a circuit builder. Neither begins from "this is a programming language" and derives circuits as an output. [ARG] 

24.3.2 Ingesting Classical Code That Already Exists

A large fraction of any real quantum algorithm is classical arithmetic performed on data held in superposition — evaluating a function inside a Grover oracle, computing a phase in phase estimation, applying a cost function in optimization. That classical computation must be expressed as reversible quantum operations, because quantum evolution is unitary. [FACT] 

Ordinary classical logic is irreversible: an AND gate destroys information about its inputs. Converting an arbitrary classical function into reversible form requires ancilla qubits, Toffoli-gate decomposition, and careful uncomputation to release scratch space without corrupting the superposition — with real space-time tradeoffs at every step. Horizon's own description of doing this by hand: a task that may otherwise require anywhere from dozens to many millions of operations. For a non-trivial function this is specialist work measured in weeks, and it must be redone when the target hardware changes. [FACT] 

The c2q toolchain compiles C and C++ functions directly into partial quantum circuits, made available as callable subroutines within Helium via a single line of code. A c2q specification file declares the entry point into the classical source and the optimisation settings governing performance tradeoffs. [FACT] 

Why this is commercially decisive rather than merely convenient: consider a pharmaceutical company with a validated computational chemistry pipeline — geometry optimisation, integral evaluation, property prediction — built over decades, scientifically load-bearing, in a regulated environment where revalidation is expensive and slow. Every competing framework asks that organisation to have someone rewrite the relevant components as quantum circuits. Horizon asks it to point a compiler at code it already trusts. In regulated industries the ability to preserve an existing validated codebase is frequently the difference between adoption and refusal — and no competitor offers a path from an existing classical codebase into quantum execution at all. [ARG] 

24.3.3 Owning and Operating Hardware in Two Modalities

On systems lacking real-time classical feedback, an algorithm requiring a mid-circuit branch is executed by running the whole circuit many times and discarding every run where the measurement did not produce the needed outcome. This is post-selection. Its cost compounds: each additional branch point multiplies the discard rate, so the number of runs required grows exponentially with the number of adaptive decisions. Many algorithms are not slow under post-selection — they are infeasible. [ARG] 

Per the Q2 2026 earnings release, Triple Alpha on Ember-1 supports real-time execution of complete programs, eliminating the need for post-selected execution, with pulse- and gate-level access supporting a broad range of quantum operations and workflows. Pulse-level control means direct manipulation of the analogue signals sent from control electronics to the processor — beneath the gate abstraction entirely. [FACT] 

Why owning the machine is the only route: a cloud API exposes what the vendor chose to expose, on the vendor's latency budget, mediated by software layers that Fitzsimons noted can obscure what a machine is actually capable of. Instrumenting a control system, closing a feedback loop in microseconds, or testing behaviour at the edge of hardware tolerance requires physical possession. He also noted a benefit unavailable to any renter: freedom to test capabilities aggressively without risking damage to a scarce machine. [FACT] 

Ember-1 pairs a Rigetti Novera 9-qubit superconducting processor with Quantum Machines OPX1000 control electronics and a Maybell dilution refrigerator, live in Singapore. The AQ256 is a 256-qubit IonQ trapped-ion system, contracted at $35 million, targeted for Dublin in 2027. These architectures differ in qubit realisation, native gate sets, connectivity graphs, error mechanisms, and operating timescales — trapped ions offer all-to-all connectivity and long coherence with slow gates; superconducting qubits are fast and cheap per shot with constrained connectivity. [FACT] 

A compiler developed at pulse level against both, by one team, is a categorically stronger claim to hardware-agnosticism than targeting many backends through vendor APIs. No other software pure-play in quantum computing owns and operates hardware in two modalities. The barrier is not cleverness — it is that a software company must be willing to buy a $35 million quantum computer, and almost none will. [ARG] 

24.3.4 A Patent Over the Core Mechanism

US Patent 11,842,177 B2 — filed June 3, 2021, granted December 12, 2023, assigned to Horizon Quantum Computing Pte. Ltd., inventors Joseph Francis Fitzsimons and Si-Hui Tan. The granted claims describe a compiler that obtains a program in a unified language "that is effectively a classical language, as opposed to a quantum language," performs code refactoring, converts the refactored code into intermediate representations comprising quantum data structures, and emits gate-level code conforming to the instruction set and gate-locality constraints of a specific target quantum processor. A continuation (US 2024/0111506 A1) is pending; a European application (EP 4012552 A1) was filed. [FACT] 

The claim language is not adjacent to the commercial differentiator — it is the commercial differentiator, described in patent terms: compile classical code, target arbitrary hardware. The principal comparators are open-source projects with no exclusionary position at all. Whether Horizon would ever litigate is a separate question, and industry norms lean toward trade secrecy and open publication rather than enforcement; the honest reading is that this is a defensive and signalling asset with real but untested teeth. [ARG] 

24.3.5 An End State No Competitor Has Articulated

Horizon states that beyond Triple Alpha's existing capabilities it is developing techniques for automatically constructing quantum-accelerated algorithms from classical code — methods that go beyond translating instructions into quantum-compatible form and instead generate implementations that exploit quantum speedups where they exist. The company describes the goal as taking programs written in conventional languages such as C or Python and transforming them into accelerated quantum applications, requiring automated algorithm synthesis paired with an optimising compiler. [FACT] 

This is the fourth layer above Hydrogen, Helium, and Beryllium, and it is the actual destination. If it works even partially, the addressable population moves from the few hundred people worldwide who can formulate quantum algorithms to the tens of millions who can write ordinary code. No competitor has articulated this end state, let alone organised a language stack beneath it as a staged path toward it. Horizon's four layers are not four products — they are the visible increments of one long programme, and each shipped layer is evidence the programme is proceeding. [ARG] 

24.3.6 Founders Who Originated the Science

Fitzsimons co-authored "Universal Blind Quantum Computation" at FOCS 2009 — a top-two venue in theoretical computer science — with the experimental demonstration published in Science in 2012 alongside Anton Zeilinger, who received the 2022 Nobel Prize in Physics. Field literature describes verification via blind quantum computing as initiated by Fitzsimons and Kashefi. Si-Hui Tan holds a Caltech BSc and an MIT PhD under Seth Lloyd, was an MIT Presidential Fellow, and is co-inventor on the patent; the two have collaborated since 2013. [FACT] 

The relevance is specific. Blind and verifiable computation is the theory of computing correctly on a machine you do not control and cannot fully observe. That is structurally identical to compiling for hardware whose noise, connectivity, and drift you cannot fully characterise. Horizon's two most senior technical people spent a decade on the theoretical version before building the engineering version, and are named together on the patent. This is not a founder with credentials plus a hired scientist — it is a research partnership that produced a company. [ARG] 

24.3.7 Why the Combination Is the Asset

Any single capability above is interesting. The combination is what is rare, and each element is load-bearing for the others. Turing-complete languages are theoretically elegant and practically unverified without hardware to run them on — the languages require the testbeds. Owned hardware without the language stack is an expensive lab. c2q without Turing-complete control flow has nowhere to put the subroutines it generates. The patent without a shipping product protects nothing. The fourth layer is unreachable without the three beneath it. And the founders' research background is what made the whole architecture conceivable before the hardware existed to demand it. [ARG] 

What Horizon has assembled is a complete vertical attempt at the platform layer — languages, compiler, classical-code ingestion, owned multi-modal hardware, protected IP, and a metered delivery model — held by 57 people while no incumbent contests the ground.

24.4 Why the Position Is Difficult to Copy

Time. Horizon has been building this since 2018 — eight years of compiler work preceded by roughly a decade of the founders' underlying research. A competitor starting now begins eight years behind on a problem whose difficulty is precisely that it cannot be shortcut. [ARG] 

Physics. The nanosecond-per-foot latency constraint is not an engineering preference; it is the speed of light. Concurrent classical processing inside a running quantum program requires the classical control system to sit physically adjacent to the processor. A software vendor working through cloud APIs cannot develop or validate that capability at all — not slowly, but not at all. Access to the problem requires owning machines, and very few software companies will buy a $35 million quantum computer. [ARG] 

Incentive. The hardware vendors best positioned to build this layer have the least reason to. IBM's tooling exists to sell IBM machines; a genuinely hardware-agnostic layer would make its customers indifferent to IBM hardware. This is the classic incumbent's dilemma, and it is why the position tends to fall to a specialist outsider — as it fell to Microsoft rather than IBM in 1981. [ARG] 

24.5 The Shape of the Opportunity

What an investor is buying at roughly $1.08 billion is not a software business with visible revenue. It is a claim on the abstraction layer of a computing paradigm, at the stage before that layer has an owner, held by the team with the strongest technical provenance in the field and protected by an issued patent. [ARG] 

• If quantum computing becomes commercially significant and needs a general-purpose programming layer, the entity that owns that layer occupies the position CUDA occupies in accelerated computing and Windows occupied in personal computing — and the current valuation will look like an entry point rather than a price.
• If quantum computing becomes commercially significant but never needs more than hand-optimised circuits for a handful of narrow algorithms, Horizon is a well-capitalised compiler company with excellent researchers and a limited market.
• If quantum computing does not become commercially significant, no software layer matters and the entire sector fails together — a risk Horizon shares with every company in it and cannot hedge.

The critical observation is that the bull case does not require Horizon to win a hardware race, pick the winning modality, or out-engineer IBM on circuit optimisation. It requires the industry to eventually need a programming layer above circuits, and for Horizon to have built it first while the ground was uncontested. That is a materially wider set of futures in which this investment works than any single-modality hardware bet offers. [ARG] 

And unlike Microsoft in 1981 — which needed IBM to hand it the opportunity — Horizon is not waiting to be chosen. It bought the machines.

24.6 What This Section Does Not Claim

Stated so the argument can be assessed honestly rather than accepted rhetorically:

• Horizon has no revenue, no named customer, and no disclosed pricing. Every element of this thesis is positioning, not performance.
• The comparators are free. Horizon must prove that a proprietary layer delivers enough value to displace open-source tools that cost nothing — the assumption with the least supporting evidence in this report.
• The analogy has limits. Microsoft had a shipping market of millions of PCs; quantum computing has no equivalent installed base, and the timing of one is genuinely unknown.
• First to build is not the same as first to monetise. Computing history contains many companies that built the right layer and watched someone else collect the rents.
• Beryllium reached early access at the end of Q2 2026 with a small group. The network effects described here do not yet exist at any meaningful scale.

I hold the position because the barriers above are real, the ground is genuinely uncontested, and the cost of the option is roughly $1.08 billion against a paradigm-layer outcome. That is a judgment about asymmetry, not a prediction — and Section 29 sets out exactly what would have to appear before it could be called anything more. [ARG] 

24.7 Why We Believe Horizon Will Succeed

This report has been deliberately restrained about what the evidence proves. This section states what we believe, why, and on what terms it can be judged wrong. It is opinion, tagged accordingly — but it is opinion built on the preceding forty pages of primary-source work rather than on enthusiasm.

First, the definition. By success we mean a specific, assessable outcome: that Triple Alpha becomes the standard general-purpose programming layer through which a material share of commercial quantum workloads are expressed and executed, generating metered revenue at platform scale. Not merely surviving. Not being acquired for its patents. Owning the layer.

We believe that outcome is likely, for seven reasons that compound rather than merely accumulate. [ARG] 

One: the bar is lower than the sector's. Almost every quantum investment requires a specific physical bet to pay off — that superconducting beats trapped ion, or that a particular error-correction scheme scales. Horizon requires none of that. It requires quantum computing to work at all, in any modality, and to eventually need a programming layer above circuits. Horizon already compiles across four architectures and 30+ backends. The number of futures in which the thesis works is a superset of the futures in which any single hardware vendor's thesis works — and it fails only in the scenario where the entire sector fails, which is a risk borne equally by every participant. [ARG] 

Two: the hard part is built and shipping, not promised. Hydrogen and Helium are in production inside Triple Alpha today, Turing-complete, with C/C++ subroutine ingestion working. Beryllium moved from December 2025 preview to early access at the end of Q2 2026 — on schedule. Ember-1 went from announcement to inauguration to external users inside twelve months. The AQ256 went from announcement to signed $35 million contract with a named site. Across five consecutive Form 6-Ks, this company has said what it would do and then done it. The execution risk that normally dominates deep-technology investments has been partially retired by observed delivery. [ARG] 

Three: the barriers protecting the position are structural, not competitive. Time, physics, and incentive (Section 24.4) are not obstacles a well-funded rival overcomes with effort. A competitor cannot compress eight years of compiler work; cannot develop real-time execution through a cloud API at any budget, because the constraint is the speed of light; and if it is a hardware vendor, it actively does not want a layer that makes customers indifferent to its machines. These are not leads that erode with competition. They are the kind that widen. [ARG] 

Four: the team has already solved the harder version of this problem. Fitzsimons and Tan spent a decade proving how to compute correctly and privately on machines you neither control nor fully observe — then spent eight more building the compiler that does it commercially, and hold the granted patent covering the mechanism. This is not a team learning the domain on investors' capital. It is the team that wrote the theory, executing the engineering, on their own prior work. [ARG] 

Five: they are funded past the decisive milestone. $113.3 million in cash against roughly $9.3 million of half-year operating burn, with the AQ256 already contracted at a fixed price and IonQ contractually responsible for delivery and benchmark verification. Horizon does not need to raise money to reach the Dublin installation that tests its central claim. Most deep-technology companies fail because they run out of runway before the proving event; Horizon has bought its runway past it. [ARG] 

Six: demand is already visible, and deliberately unconverted. The 20-F documents inbound early-access interest from 40-plus major corporations, 80 universities, 10 quantum software companies, and 15 national laboratories and government agencies. Management is declining to convert it into services revenue because doing so would misprice the platform. A company with no interest and no revenue is a company nobody wants. A company with documented interest at that scale and no revenue by choice is a different situation entirely — and the market appears to be pricing the first. [ARG] 

Seven: the ground is uncontested, and structurally likely to remain so. Every plausible incumbent has a reason not to build this layer — Qiskit sells IBM hardware, PennyLane serves Xanadu's photonics, vendor SDKs exist to differentiate machines, and the open-source players have no commercial model that rewards owning it. Horizon is not racing a competitor to the position. It is walking toward a position nobody else is structurally motivated to occupy, and it has been walking since 2018. [ARG] 

Compounding matters more than the individual points. A company with a strong team but no moat gets copied. A company with a moat but no funding dies before it matters. A company with both but an unclear market builds something nobody needs. Horizon has the team that originated the science, a moat grounded in physics and time rather than mere effort, capital sufficient to reach its proving event, documented latent demand, an issued patent, and no incumbent contesting the ground. We have found no other company in quantum computing where all six hold simultaneously. [ARG] 

24.8 On the Sector Itself: We Believe Quantum Computing Will Succeed

We should be direct about something a conditional framing would obscure. We do not merely assume quantum computing will succeed for the sake of argument — we believe it will. [ARG] 

The reasons are visible in this report rather than asserted around it. Resource requirements for the canonical hard problems are falling fast, and falling through better algorithms and architecture rather than only through more qubits — a 27-fold reduction on FeMoco from a single change in qubit encoding. Hardware has moved from laboratory demonstration to commercial product: IonQ is selling sixth-generation 256-qubit chip-based systems to buyers who pay $35 million and take delivery. Error correction has moved from theoretical construction to engineering roadmap with dates attached. Governments across three blocs are funding at national-programme scale, which is a judgment by parties with access to classified assessments. And the theoretical foundation — that a machine built from quantum mechanics simulates quantum mechanics with an advantage no classical machine can match — is not a hypothesis awaiting confirmation. It is the reason the field exists, and it has never been the weak link. [ARG] 

On timing, we will not offer a date, but neither will we claim the question is opaque. Our perspective comes from a decade of following this ecosystem — beginning with tracking the professors whose work would inevitably reach the capital markets and produce the first wave of quantum startups, then watching those companies through the long, careful, often unglamorous progress that followed. What we observe now is different in kind from what we observed in the first eight of those years. The ecosystem has entered a phase of accelerating and compounding maturity: hardware moving from demonstration to product with delivery schedules and purchase contracts; resource requirements for the canonical problems falling through architectural insight rather than brute force; error correction leaving the whiteboard for engineering roadmaps; capital, sovereign and private, arriving at scale; and the founding generation of academic researchers now running operating companies with public filings and customers. Each of these was a bottleneck we watched persist for years. They are resolving in parallel, and they are resolving faster than they did. [ARG] 

That maturity is the condition under which Horizon's position converts from architecture into revenue. A software layer is worth little in an ecosystem of laboratory demonstrations and worth a great deal in an ecosystem of deployed machines running commercial workloads. We expect the acceleration we are now observing to reap extensive benefits for Horizon specifically — because it is positioned at precisely the layer that a maturing ecosystem needs next and does not yet have. [ARG] 

So the full statement of our view is this: we believe quantum computing will succeed, and we believe Horizon will be the layer through which it is programmed. The second conviction does not depend on the first being certain — that is what makes the position asymmetric — but we hold both, and it would be less than honest to state only the one that requires less of us. This is the most asymmetric position we have found in the sector, and it is why we are long. [ARG] 

What would prove us wrong, stated as clearly as the conviction: a shipped, credible, free general-purpose quantum programming layer from a well-resourced incumbent; a slipped or failed AQ256 installation that reveals the compiler cannot deliver on production hardware; Beryllium failing to attract developers after general availability; or a sustained widening of losses that forces dilutive financing before the platform monetizes. Section 29 tracks each of these as a formal trigger, and Section 24.6 states what this thesis does not claim. We would rather be checked against those markers than believed.

25. The Bear Case, Answered

No sell-side analyst has published a bear thesis on Horizon Quantum. Coverage consists of two Buy ratings — Needham at $20 (June 3, 2026) and Craig-Hallum at $21 (August 17, 2026). The negative ratings appearing on aggregator screens come from Wall Street Zen and Weiss Ratings, rules-based quantitative services that do not employ analysts covering this name; Morningstar states explicitly that its rating on HQ is model-generated and that the company is not formally covered by an analyst. [FACT] 

This creates a problem for a bull report. Absent a published bear thesis, it is tempting to declare the field clear. We think that is a mistake. The honest exercise is to construct the strongest bear case available from the actual evidence — the case a competent skeptic would make — and answer it. What follows is our attempt to argue against ourselves as well as we can, and then respond. Where the bear has the better of it, we say so.

25.1 "The company lost $115 million in a quarter and missed EPS by 2,344%."

The bear case: Q2 2026 GAAP net loss was $115.2 million, or $2.20 per share, against a consensus estimate of $0.09 — among the largest proportional misses on the market. Quantitative rating services responded by moving the stock to Sell and Strong Sell. [FACT] 

Our answer: roughly $108.3 million of that loss is a non-cash charge from remeasuring warrant liabilities at fair value — an accounting consequence of the share price rising during the quarter. It consumed no cash, funded no expense, and reflects no operational change. CFO Greg Gould confirmed on the call that remaining warrants will continue to be remeasured quarterly, so this line will keep swinging in both directions with the stock. [FACT] 

The mechanically important point: this charge is worse when the stock does better. Any bear citing both "the stock is weak" and "look at the GAAP loss" is double-counting, because those facts are inversely related. The $0.09 consensus estimate also demonstrates that the analysts modelling this company did not forecast warrant remeasurement — which is unforecastable, since it depends on an unknowable future share price. The miss is an artifact of an unmodellable line, not operational underperformance. The quantitative downgrades follow mechanically from the same artifact: an algorithm scoring revenue, earnings, and margin will penalise a pre-revenue company with a large non-cash loss, and the rating carries no information beyond "this company has no revenue" — which this report states repeatedly. [ARG] 

Where the bear retains ground: operating loss of $7.2 million and adjusted EBITDA loss of $5.5 million are real, cash, and widening. Those are the numbers that matter, and they are moving the wrong way — which is Objection 2.

25.2 "Losses keep widening, and the decisive catalyst is still a year away."

The bear case: operating loss has run $4.87M to $4.55M to $6.50M to $7.18M — up sharply over the last two quarters — while adjusted EBITDA loss has risen in every sequential step: $3.01M to $3.03M to $4.10M to $5.46M. Revenue over the same period: zero. Meanwhile the event that would test whether the software layer delivers on production hardware — the AQ256 installation — is at least a year out, requires a facility not yet built, and carries facility capex that has never been separately disclosed. Hardware installations routinely slip. [FACT] 

Our answer on spending: the expenditure is identifiable and matches the strategy rather than drifting. R&D rose 117% on headcount and hardware testbed setup; G&A rose 236% on the costs of becoming a public company; sales and marketing — the line that would signal desperation if it dominated — is the smallest at $0.4 million. Headcount went from 34 to 57. This is the expenditure profile of a company building a product and a testbed, not one burning cash on customer acquisition it cannot convert. Management described Q2 opex as a reasonable baseline for near-term planning. [FACT] 

Our answer on the catalyst: the contract structure transfers much of the execution risk. IonQ is contractually responsible for procuring, constructing, installing, and verifying the system against specified performance benchmarks, at a fixed $35 million. Horizon is not managing a hardware build; it is taking delivery of one. It is also funded through the milestone — $113.3 million against roughly $9.3 million of half-year operating burn — so it need not raise capital to reach the test. And Dublin is not the only evidence: Ember-1 is operating today and has produced a real technical result in real-time execution without post-selection. The 2027 event is the scale test of something already demonstrated at nine qubits. [ARG] 

Where the bear retains ground: "reasonable baseline" is a statement of intent, not a result, and three consecutive quarters of rising adjusted EBITDA loss is a trend rather than noise. Dublin facility capex beyond the system cost is genuinely undisclosed and is the most likely source of burn acceleration. Both a narrowing adjusted EBITDA loss and Dublin milestone progress are formal triggers in Section 29 precisely because we cannot yet claim either.

25.3 "It has no revenue, no customers, and no published pricing. Everything is a promise."

The bear case: FY2025 revenue was $38,873 — down from $263,785 in FY2024. H1 2026 revenue was zero. No named customer, no disclosed contract, no published price list. Every element of the thesis is positioning rather than performance. [FACT] 

Our answer: the revenue absence is a disclosed strategic choice with a stated rationale, not an inability to sell. Fitzsimons said directly that revenue available today would come from professional services bundled with system access, and that booking it would misrepresent the step-change the platform is built for. Against that, the 20-F documents inbound early-access interest from more than 40 major corporations, 80 universities, 10 quantum software companies, and 15 national laboratories and government agencies. A company nobody wants and a company declining to monetise demand are different situations with identical revenue lines. [FACT] 

The revenue model is specified rather than vague: usage-based metering through Horizon's proprietary stack, with applications deployed as API endpoints. That scales with customer workload rather than with headcount — precisely the property professional-services revenue lacks, and the reason the refusal is consistent with the design of the business rather than a narrative preference. [ARG] 

Where the bear retains ground: this is the strongest objection in the list and we do not fully defeat it. Declining to monetise removes the single data point that would let anyone verify market pull, and it asks shareholders to fund years of losses on management's judgment about timing. Inbound interest is not a pipeline. We accept this as a genuine cost of the strategy, not a misunderstanding by the market.

25.4 "Competitors are better funded, and the main alternatives are free."

The bear case: Classiq (approximately $180–200M), Riverlane (approximately $195M), and Q-CTRL (approximately $190M) have each raised more than Horizon's roughly $120 million de-SPAC gross plus $28.7 million of warrant proceeds — and Horizon carries a $35 million hardware commitment they do not. Meanwhile Qiskit, Cirq, PennyLane, and tket cost nothing and hold overwhelming developer mindshare. [FACT] 

Our answer on capital: the relevant test is not who raised most but who is funded past their proving event, and Horizon is. Its binding constraints are time and physics — an eight-year compiler head start, and the speed-of-light limit on cloud-mediated real-time control — neither of which additional capital relieves. A competitor with $200 million cannot buy back eight years, and cannot develop real-time execution through an API at any budget. [ARG] 

Our answer on free alternatives: the open-source tools are free because they are demand generation for their sponsors' hardware. Qiskit exists to sell IBM machines; PennyLane serves Xanadu's photonics. None has an incentive to build a genuinely hardware-agnostic layer, because such a layer makes customers indifferent to the sponsor's hardware. That is the incumbent's dilemma, and it is why the position historically falls to a specialist outsider rather than to the largest incumbent. [ARG] 

Where the bear retains ground: "free" is a real competitive weapon regardless of motive, and mindshare compounds. If circuits remain the durable interface rather than being superseded by a general-purpose layer, Horizon's adoption deficit is decisive rather than incidental. This is the single least-evidenced assumption in the report, and Section 24.6 says so.

25.5 "This is a bet on quantum computing working at all — and that remains unproven."

The bear case: no commercially valuable problem has been demonstrated running better on a quantum computer than on a classical one. Industrially relevant chemistry simulation is independently placed in the 2033–2036 window. If broad commercial advantage never arrives, no software layer matters and the equity is worth little. [FACT] 

Our answer: this is the one risk Horizon cannot hedge, and we do not pretend otherwise. Our view is stated openly in Section 24.8 rather than buried — we believe quantum computing will succeed, on the evidence of falling resource requirements achieved through architecture rather than brute force, hardware moving from demonstration to purchasable product, error correction on dated engineering roadmaps, and sovereign funding across three blocs.

The structural point in Horizon's favour is that it requires no particular modality to win. It compiles across four architectures and 30+ backends. It fails only in the scenario where the entire sector fails — a risk every participant shares equally — and it captures value in every scenario where any modality succeeds. That is a materially wider set of futures than any single-hardware position offers. [ARG] 

Where the bear retains ground: entirely. This objection cannot be answered with evidence available today, only with judgment. An investor who believes quantum computing will not reach commercial advantage should not own this stock, and should not own anything else in the sector either.

25.6 What the Bear Gets Right

Stated plainly, so that nothing above reads as reflexive defence. The bear is correct that revenue is zero and the strategy keeping it there removes the market's ability to verify demand; that losses are running at roughly double year-earlier levels with adjusted EBITDA loss rising in each of the last three sequential quarters, and management's baseline is intent rather than result; that the decisive catalyst is more than a year away with undisclosed facility capex attached; that the principal alternatives are free and hold the developer community; and that sector-level risk is real and unhedgeable. [ARG] 

What we dispute is not the facts but the inference. Every item above is a cost of building a platform position before the market for it exists. The bear reads them as evidence the position will not materialise. We read them as the price of holding it — and the distinction is testable against the triggers in Section 29 rather than being a matter of taste. [ARG] 

26. Valuation Context

 

HQ traded as high as $45.00 and as low as $8.29 over the trailing 52 weeks. The same Level 3 warrant-valuation model that used a $27.76 stock price as of June 30, 2026 implies the stock sits roughly 28% below that quarter-end mark at the August 21, 2026 close of $19.93 — having risen approximately 61% from the ~$12.40 level of mid-August. [FACT] 

The stock moved sharply in the weeks around this report. Investing.com's Aug 11 coverage of the Canaccord Genuity conference put shares at $12.83 premarket (up from a $12.56 prior close), describing a 17.6% decline over the trailing week. By Aug 12, Morningstar showed a close of $16.35 (+9.36%); the stock closed at $16.77 on Aug 18 and at $19.93 on Aug 21 — approximately 61% above the Aug 10–11 level in under two weeks. Craig-Hallum's Aug 17 Buy initiation supplies at least a partial catalyst, though it postdates the Aug 12 jump and so does not explain the earlier leg. [FACT] 

Sell-side coverage has expanded and now runs in both directions. Needham (N. Quinn Bolton) initiated June 3, 2026 with a Buy and a $20.00 price target. Craig-Hallum initiated August 17, 2026 with a Buy and a $21.00 target. The rally through August 21 has essentially closed the gap to both: at $19.93 the stock sits fractionally below Needham's target and roughly 5% below Craig-Hallum's. Against those, quantitative/model-driven services have been negative: Wall Street Zen downgraded HQ from Sell to Strong Sell on August 8, 2026, and Weiss Ratings moved from "sell (e)" to "sell (e+)" on June 17, 2026. Q2 earnings-call Q&A also included named analysts from Rosenblatt Securities and StoneX, whose published ratings I have not located. [FACT] 

The split is worth stating plainly rather than cherry-picking: the two fundamental research houses covering the name are constructive with targets in the $20–21 range, while the rules-based screening services are negative — an unsurprising divergence for a pre-revenue company, since quantitative models penalize zero revenue and negative earnings almost mechanically. Neither camp should be treated as independent confirmation of the other.

Shares closed at $19.93 on August 21, 2026, up roughly 61% from the ~$12.40 level of August 10–12. Craig-Hallum's August 17 Buy initiation is a plausible partial catalyst; the earlier leg of the move is not explained by any company-specific disclosure I have located. [FACT] 

At $19.93, implied market cap is approximately $682 million using Class A shares only (34,227,495 × $19.93) or approximately $1.08 billion using all 53,972,080 shares outstanding. Data vendors use the combined Class A + Class B convention, so the higher figure is the one that will appear on quote pages. Note that the stock has now essentially reached both published analyst targets — a point the report should state plainly rather than continue citing upside that no longer exists. [INFER] 

Market cap depends on which share count is used, and the choice should be stated rather than assumed. Using only the 34,227,495 publicly traded Class A shares against the $19.93 August 21 close gives ≈$682 million. Using the full 53,972,080 shares outstanding (Class A + Class B, the convention common for dual-class issuers, and the one data vendors apply to HQ) gives ≈$1.08 billion. Both are computed directly from the Q2 6-K share count. Any apparent disagreement between vendors on this name is a denominator difference rather than a data error. [INFER] 

For sector context, IonQ — the largest and only meaningfully revenue-generating pure-play quantum hardware company — carries a market capitalization in the double-digit-billions range and reported Q1 2026 revenue of roughly $64.7 million (up sharply year-over-year), while smaller hardware peers (Rigetti, D-Wave) trade in the single-digit-billions with revenue in the low single-digit millions to tens of millions. Horizon, by contrast, is pre-revenue and roughly an order of magnitude smaller by market cap than IonQ. That comparison set is imperfect — Horizon is a software company, not a hardware seller, and software layers historically capture value on different timelines and different margin structures than the hardware beneath them. The honest reading of the gap is that Horizon is priced for platform optionality rather than demonstrated traction; whether that represents room to grow into a software-layer position or simply an appropriate discount for having no revenue is precisely the question this report cannot settle on current evidence.

Anchoring note: the ≈$1.08 billion total-share figure is the more useful reference point for peer comparison, since it matches the convention data vendors use for HQ and the convention applied to the dual-class peers it would be compared against. The rally from the mid-August low coincided with primary-source catalysts — the Canaccord conference appearance and the Craig-Hallum initiation — but it has also carried the stock to both published price targets, which removes the analyst-target support the valuation previously had. Section 30 explains why analyst targets should not be scored as independent validation in either direction.

With no meaningful revenue, this remains a pre-revenue platform bet priced on optionality — the AQ256 delivery in 2027, Beryllium's path from early access to general availability, and Ember-1's uptime/utilization trajectory are the observable near-term markers that would inform any valuation discussion, rather than a discounted-cash-flow approach that isn't yet supportable given the revenue base. [ARG] 

27. Risks and Challenges

• Execution risk on the Dublin-area facility and AQ256 integration timeline — a 2027 milestone, not a near-term catalyst, and one whose associated capex has only just begun to show up in the cash flow statement. Mitigant: the system is already under signed contract at a fixed $35M, with IonQ contractually responsible for procurement, construction, installation, and verification against performance benchmarks.
• Continued pre-revenue status with a widening (not narrowing) adjusted-EBITDA loss trend over the last three sequential quarters, and operating loss up sharply year over year. Mitigant: $113.3M cash against ~$9.3M half-year operating burn provides a multi-year floor, management has called Q2 opex a reasonable near-term base, and the Q2 6-K states going-concern doubt is alleviated for at least twelve months.
• A dilution overhang that is now quantified rather than qualitative: 765,554 public warrants, 2,884,660 private warrants, 6,341,712 options (avg. strike $3.36), and 474,784 RSUs against 53.97M shares outstanding.
• Non-cash warrant-liability volatility will continue to distort headline GAAP EPS each quarter as the share price moves — the swing from a $15.3M to $76.8M warrant liability in a single quarter illustrates the scale of this effect.
• Structural competitive risk from open-source incumbents (Qiskit, PennyLane) that compete on price (free) against Horizon's proprietary commercial model. Mitigant: neither operates its own hardware, and neither can develop against the physical-adjacency constraint described in Section 7 and the flywheel in that section — the differentiator is capability, not price.
• Enterprise adoption risk — competing vendor-native SDKs (IonQ's own tools, Rigetti's offerings) may retain developer share even where Horizon's hardware-agnostic pitch is technically differentiated. Mitigant: Horizon's stated near-term go-to-market focus is hardware manufacturers rather than broad developer share, which sidesteps the SDK popularity contest for now; 30+ supported backends and deep integrations with four vendors give it multi-modal reach no vendor-native tool has.
• Export-control and cross-border regulatory exposure given the Singapore/Ireland/U.S. footprint and the sensitivity of quantum hardware to national-security-related trade rules.

28. Honest Concessions

Three cautions about how this company is commonly covered, each corrected here against primary documents. Near-term revenue visibility is routinely overstated; Ember-1's actual scale — a nine-qubit chip — is routinely understated; and IonQ's 13.3% and 7.8% figures are routinely presented as a decline when they are the same holding measured against different denominators — percentage of Class A versus percentage of all shares outstanding. The AQ256 cost is likewise often described as a press estimate when it is disclosed in Horizon's own financial statement notes. [ARG] 

The general lesson: evidence-tier tags are only as good as the sources actually fetched, and search-snippet aggregation is not the same as reading the filing. This report works from the 20-F, both quarterly 6-K exhibits, the earnings-call transcripts, the patent text, the Section 16 filings, and the Schedule 13Ds directly. The material point is what that produces: on the primary record, the technical and contractual progress is stronger and better documented than secondary coverage conveys — the AQ256 is a signed $35M contract, the Dublin site is company-confirmed, the inbound demand is quantified, and the leadership bench is deeper. I hold a disclosed long position in Horizon Quantum Holdings, and this report was prepared, as always, to stress-test that position rather than to promote it.

29. What Would Change This View

Strengthening triggers — what would convert optionality into demonstrated progress:

• A signed, named commercial or government customer contract with disclosed economics would meaningfully strengthen the revenue-pathway thesis. The inbound interest documented in Section 20 is the raw material; conversion is the missing step.
• Beryllium's transition from early access to general availability, with disclosed developer-adoption metrics, would be the clearest observable signal of product-market progress.
• Confirmed on-schedule progress toward the Dublin-area facility (permitting, construction milestones) ahead of the 2027 AQ256 target would reduce execution-timeline risk and validate the capex plan.
• Published results from Ember-1 — uptime, external-user activity, or calibration gains from the Quantum Machines collaboration — would convert the in-house hardware argument from structural claim to measured advantage.
• A quarter in which adjusted EBITDA loss narrows rather than widens would be the first sign of operating leverage.

Weakening triggers — what would undermine it:

• A slipped Dublin or AQ256 timeline, or capex materially above plan, would damage the central 2027 catalyst.
• Continued widening of adjusted EBITDA loss beyond the Q2 base management described as reasonable, or a new equity raise on unfavorable terms, would compound the dilution already sized in Section 18.
• An updated Schedule 13D or 13G from Peak XV stating both their current share count and the denominator used would resolve the ≈18.9% vs. ≈12.0% ambiguity in Section 19. Their exact holding at the de-SPAC is in their own filing, so the percentage can be computed against whichever denominator a reader prefers.
• Insider selling once lock-ups release, particularly by Fitzsimons around the ~March 2027 expiry, would be a meaningful negative signal given that the record to date shows no open-market disposals at any price.
• Disclosure that the internal applications team's three unnamed high-value problems include a chemistry or life-sciences target would materially firm up the pathway in Section 22; disclosure that they do not would weaken it.

Tracking dashboard. The above are thresholds; the table below is the repeatable checklist I intend to run against each quarterly print, so that judgments are made against pre-committed metrics rather than reconstructed after the fact.

Indicator	Current reading (as of Aug 2026)	What I'm watching for
Revenue / bookings / named design partners	$0 revenue Q1 and Q2 2026; no named customer in any disclosure	First named contract with disclosed economics; any recurring or usage-based revenue line
Beryllium status	Early access since end of Q2 2026; small group, mostly hardware partners	General-availability date; disclosed active developers, workloads compiled, retention
Ember-1 utilization	Opened to first external users Q2 2026; no metrics disclosed	Uptime, external user counts, calibration gains from the Quantum Machines work, published workload results
Dublin / AQ256 execution	New premises being selected; installation targeted 2027	Lease/permitting/construction milestones; installation start; verified performance against spec
Adjusted EBITDA loss trend	Risen in each of the last three sequential quarters: $3.01M → $3.03M → $4.10M → $5.46M	First quarter of narrowing, or flat burn despite continued investment
Cash and burn	$113.3M at Jun 30, 2026; ~$9.3M H1 operating burn; Q2 opex called a reasonable near-term base	Whether opex holds near that base; whether Dublin capex materially accelerates investing outflow
Fully diluted share count	53,972,080 outstanding; 765,554 public + 2,884,660 private warrants; 6,341,712 options; 474,784 RSUs	Further warrant/option exercises; any new equity or convertible financing
IonQ stake vs. 5% threshold	7.8% at Jun 30, 2026 — roughly 2.8pp of headroom	Whether continued dilution pushes IonQ below the 5% floor that conditions its board-nomination right
Insider transactions	No open-market sales in any Form 4 reviewed; no open-market purchases either	First discretionary sale or purchase by any officer or director — either direction is informative
Lock-up expiries	Fitzsimons' Class B locked to ~Mar 2027; other holders' terms not itemised in reviewed disclosures	A full supply calendar; whether Fitzsimons extends, and whether PIPE/pre-SPAC holders' releases cluster near the AQ256 window

30. Signal Quality: What Counts as Evidence and What Doesn't

Section 19 lists what would strengthen the thesis. This section is its complement: developments that read as validation but shouldn't be scored as such. It exists because the common failure mode in coverage of pre-revenue companies is exactly this — treating adjacent, easily available signals as if they were evidence of commercial progress.

Signal	Why it is not sufficient on its own
Analyst price targets of $20–$21	Two Buy ratings are opinions about the same public facts I already have, not independent confirmation of adoption or revenue. Two rules-based services rate the stock negatively on the same facts (Section 16).
Conference sponsorships and technical presentations	QIP 2027 anchor sponsorship and Q2B Tokyo talks support recruiting and technical credibility. They are not pipeline. The report scores them as credibility signals only (Section 13).
More backend integrations	Going from 30 to 40 supported backends is API-level compatibility, materially weaker evidence than deep integration or paid usage. The Rigetti/IonQ/AQT/Alice & Bob tier is what matters (Section 7).
Inbound early-access interest	The 20-F's 40+ corporations, 80 universities, and 15 national labs are real and dated, but the company has explicitly chosen not to convert them into engagements. Interest is not demand fulfilled (Section 14).
A higher share price	See below — mechanically increases the warrant liability and does nothing about revenue, dilution, or execution risk.

The share-price point deserves its own statement because it cuts against intuition: because Horizon's public and private warrants are remeasured at fair value every quarter, a rising share price mechanically increases the warrant derivative liability and produces a larger non-cash loss. This is precisely what produced the $108.3M charge and the headline $115.2M Q2 net loss. CFO Greg Gould stated directly on the Q2 call that all remaining warrants will continue to be remeasured each quarter, so future price movements will keep generating non-cash gains or losses. [FACT] 

Concretely: the stock's move from roughly $12.40 in mid-August to $19.93 on August 21 is, on its own, a setup for a substantially larger warrant charge in the Q3 print than a flat or declining price would have produced. That is an accounting artifact, not an operating deterioration — but a reader who scores "stock up, analyst target raised" as unambiguously bullish and then reacts to a large Q3 GAAP loss as a negative surprise will have double-counted the same event in both directions. Adjusted EBITDA remains the right lens (Section 16).

Net position: this remains a platform-optionality thesis, not a demonstrated commercial-growth thesis. I hold it as the former and have tried throughout this report not to dress it as the latter. The distinction is not pessimism — the optionality is real, sizeable, and backed by verifiable technical execution. It simply has not yet been converted into the kind of evidence that would let anyone call it proven. [ARG] 

31. Conclusion

Sixty pages of primary-source work reduce to a single proposition. Every computing paradigm has resolved the same way — the machines commoditize and the durable economics migrate to whoever owns the layer developers write against — and in quantum computing that layer does not yet have an owner. Horizon Quantum is the only company assembling everything required to claim it, and on the evidence we have gathered we believe it will.

31.1 What the Record Actually Shows

Across five consecutive Form 6-Ks, a 20-F, two earnings-call transcripts, an investor-conference transcript, a granted patent, and the Section 16 filings, this company has said what it would do and then done it. Hydrogen and Helium ship in production, Turing-complete, with user-definable instruction sets defined in the general mathematical language of quantum operations and classical C and C++ ingestion working. Beryllium moved from December 2025 preview to early access on schedule. Ember-1 went from announcement to inauguration to external users inside twelve months, producing real-time execution that eliminates post-selection — a capability unavailable through any cloud API. The AQ256 converted from announcement to a signed $35 million contract with a named site, a named jurisdiction, and IonQ contractually responsible for delivery and benchmark verification. A Goldman Sachs managing director became CFO, a sitting Grab CFO took the Audit Committee chair, and a six-year internal product leader was promoted to Chief Product Officer. No insider has sold a share into the public market at any price.

That is an unusual density of delivery for five months as a public company, and it is the strongest available evidence about the one thing that cannot be assessed from a balance sheet: whether this team does what it says. [ARG] 

31.2 Why the Position Is Rare Rather Than Merely Good

What Horizon holds is not a lead that capital erodes. It is a combination protected by time, physics, and incentive. Eight years of compiler work cannot be compressed. The nanosecond-per-foot latency limit that makes real-time execution impossible through a cloud API is the speed of light, not an engineering preference — which means a competitor cannot develop this capability at any budget without buying quantum computers, and almost no software company will. And the hardware vendors best placed to build a hardware-agnostic layer have the least reason to, because such a layer makes their customers indifferent to their machines. This is the incumbent's dilemma that transferred the personal computer to Microsoft rather than IBM.

Around that sits a set of assets no competitor combines: an issued US patent whose claims cover compiling from a language that is effectively classical to hardware-specific gate code; owned and operating hardware in two physical modalities; founders who originated the underlying science, published it in Science alongside a subsequent Nobel laureate, and are named together as co-inventors on the patent; and a stated end state — automatic quantum acceleration of ordinary classical code — that no competitor has articulated, let alone staged a language ladder beneath.

31.3 The Opportunity Is Larger Than the Price Implies

The bottleneck in quantum computing is moving. Hardware roadmaps are engineered to relax the constraint that currently binds, and when they do, the constraint becomes the number of people who can express a valuable problem in a form a quantum machine can execute — a few hundred specialists worldwide, against tens of millions of working software developers. Closing that gap is not an incremental improvement in tooling. It is the difference between an industry serving a research community and an industry serving the software profession.

The clearest near-term expression of that value is chemistry and the health sciences. The calculations that matter most — drug metabolism at heme active sites, binding affinity against targets classical methods cannot approximate, enzyme and biocatalyst design — are defined by resource estimates that are themselves compiler outputs. A twenty-seven-fold reduction in the physical qubits required for FeMoco came from better architecture, not more hardware. A software layer that cuts those requirements again does something economically equivalent to a decade of hardware progress without building a machine, in an industry that already funds computational chemistry heavily and does not need to be persuaded of its value.

Against that, roughly $1.08 billion on all shares outstanding remains a modest price for the abstraction layer of a computing paradigm — particularly one held by the team that originated its theory, protected by an issued patent, with the ground uncontested. [ARG] 

31.4 What We Believe

We believe quantum computing will succeed. Resource requirements for the canonical hard problems are falling through architectural insight rather than brute force; hardware has moved from demonstration to purchasable product with delivery schedules and signed contracts; error correction has left the whiteboard for dated engineering roadmaps; and sovereign capital is arriving at national-programme scale across three blocs. After a decade following this ecosystem — from tracking the professors whose work would reach the capital markets, through the long unglamorous progress that followed — what we observe now is different in kind: bottlenecks we watched persist for years are resolving in parallel, and faster than they did. [ARG] 

And we believe Horizon will be the layer through which it is programmed. Not because any single capability is decisive, but because the combination is: Turing-complete languages that can express the adaptive computation error correction requires and static-circuit frameworks structurally cannot; a compiler that ingests the classical code enterprises already trust; owned hardware in two modalities to develop against; a patent over the mechanism; capital sufficient to reach the proving event without raising again; documented latent demand the company is deliberately declining to convert; and no incumbent structurally motivated to contest the ground. We have found no other company in quantum computing where all of these hold at once. [ARG] 

The risks are real and we have not minimised them. Revenue is zero and the strategy keeping it there removes the market's ability to verify demand. Adjusted EBITDA loss has widened in each of the last three sequential quarters, and both loss measures are roughly double year-earlier levels. The decisive catalyst is more than a year away with undisclosed facility capex attached. The principal alternatives are free and hold the developer community. Every one of these is catalogued at length in Sections 23, 25, 27 and 28, and Section 29 sets out precisely what would have to appear before this judgment could be called anything more than judgment.

But we read those costs as the price of building a platform position before the market for it exists, not as evidence the position will fail to materialise. On the evidence assembled here, Horizon Quantum is a genuinely unique company holding a genuinely rare position, and we expect it to succeed on a scale substantially greater than its current price reflects. That is why we are long, and why we would rather be measured against the markers in Section 29 than believed on the strength of the argument. [ARG] 

I am not giving anyone investment advice. I hope this report is useful to those following the company.

32. Source Appendix

Primary sources fetched and read directly, not via aggregator:

• Horizon Quantum Holdings Ltd., Form 6-K, Q1 2026 results incl. full balance sheet/income statement/cash flow and 9-quarter trailing summary (May 5, 2026) — sec.gov/Archives/edgar/data/2088256/000121390026052179/ea028894101ex99-1.htm.
• Horizon Quantum Holdings Ltd., Form 6-K, Q2 2026 unaudited condensed consolidated interim financial statements incl. Note 13 (Related Party Transactions — IonQ Quantum Systems Agreement, $35M, 7.8% stake) — sec.gov/Archives/edgar/data/2088256/000121390026085209/ea029986201ex99-1.htm.
• Horizon Quantum Announces Second Quarter 2026 Financial Results (full press release, incl. Beryllium/Ember-1/testbed operational detail and expense-line breakdown) — investors.horizonquantum.com/news-releases, Aug 4, 2026.
• Horizon Quantum Holdings Ltd., Form 6-K, Amanda Chew CPO appointment (Board approval Jul 28, 2026; effective Aug 17, 2026) — sec.gov/Archives/edgar/data/0002088256/000121390026084607/ea0300149-6k_horizon.htm; also company announcement, horizonquantum.com/resources/newsroom/appointment-of-amanda-chew-as-chief-product-officer, Aug 4, 2026.
• Horizon Quantum Holdings Ltd., Schedule 13D — IonQ, Inc. beneficial ownership, 13.3% of Class A shares/April 2026 basis, per Note to Row 13 based on the 20-F filed March 25, 2026 (superseded by the 7.8%-of-total-shares figure in the Jun 30, 2026 6-K) — SEC EDGAR / streetinsider.com.
• Horizon Quantum Holdings Ltd., Schedule 13D — Joseph Francis Fitzsimons beneficial ownership (filed ~March 31, 2026) — SEC EDGAR.
• Horizon Quantum Holdings Ltd., Schedule 13D — Peak XV–affiliated funds beneficial ownership, 20.3% of Class A shares/April 24, 2026 basis (filed July 9, 2026) — SEC EDGAR. Background on Peak XV's Surge seed program via The Arc (paywalled; headline/description only, not independently verified beyond that).
• Horizon Quantum Holdings Ltd., Form 6-K, Peter Oey board appointment (Apr 29, 2026; filed May 4, 2026) — sec.gov/Archives/edgar/data/2088256/000121390026051632/ea0288728-6k_horizon.htm.
• Horizon Quantum press release, "Horizon Quantum Announces Dublin as Its Second Quantum Computer Testbed Location" (June 11, 2026) — horizonquantum.com/resources/newsroom; also businesswire.com and idaireland.com.
• Horizon Quantum / IonQ joint press release, strategic agreement for 256-qubit trapped-ion system (April 9, 2026) — businesswire.com.
• Horizon Quantum and Quantum Machines Announce Strategic Collaboration to Increase Efficiency of Quantum Systems (full press release, July 29, 2026) — horizonquantum.com/resources/newsroom.
• Horizon Quantum technology and product documentation, read directly: horizonquantum.com/technology; /technology/software-infrastructure/development-infrastructure; /product/programming-languages; /triple-alpha/structure; /triple-alpha/levels-and-languages; /triple-alpha/subroutines — source for the four-layer stack, Turing-completeness, c2q classical-to-quantum subroutines, pulse-level control, and API deployment.
• Joseph Fitzsimons research record: Google Scholar profile (7,781 citations); Broadbent, Fitzsimons & Kashefi, "Universal Blind Quantum Computation," FOCS 2009; Barz, Kashefi, Broadbent, Fitzsimons, Zeilinger & Walther, Science 335:303 (2012); Barz, Fitzsimons, Kashefi & Walther, Nature Physics 9:727 (2013); Mantri, Pérez-Delgado & Fitzsimons, PRL 111:230502 (2013); Fitzsimons & Kashefi, Phys. Rev. A 96:012303 (2017); Fitzsimons, npj Quantum Information 3:23 (2017); Fitzsimons, Hajdušek & Morimae, PRL 120:040501 (2018). Funding acknowledgements: Singapore NRF Fellowship NRF-NRFF2013-01; US AFOSR grant FA2386-15-1-4082.
• US Patent 11,842,177 B2, "Systems and methods for unified computing on digital and quantum computers," filed Jun 3, 2021, granted Dec 12, 2023, assignee Horizon Quantum Computing Pte. Ltd., inventors Joseph Francis Fitzsimons and Si-Hui Tan; continuation US 2024/0111506 A1 (filed Nov 30, 2023); European application EP 4012552 A1 (filed Jun 9, 2021) — via Google Patents and Justia Patents.
• Horizon Quantum blog, "Why we're building a classical to quantum toolchain" — horizonquantum.com/updates/blog; and "Horizon Quantum to Debut Object-Oriented Language for Programming Quantum Computers" (Dec 2025) — company newsroom.
• Yahoo Finance / quarterly-update analysis, "Horizon Quantum Deepens Infrastructure Strategy as Hardware Ecosystem Expands" (May 2026) — source for the usage-based cloud pricing model and the four-layer stack progression.
• Peak XV Partners identification as the former Sequoia Capital India & Southeast Asia, and Entrepreneur First / Sequoia as early Horizon backers — via Quantum Navigator / entangledfuture.com company profile and SGInnovate.
• Horizon Quantum Triple Alpha developer video documentation (company YouTube channel): "Instruction set: Defining commands for quantum operations" — source for user-definable instruction sets, Kraus-operator command definitions, dim_in/dim_out, qudit support, POVM measurement definitions, and Python/MATLAB function handles for parameterised gates. "Hydrogen: Applying commands to quantum hardware" — source for dynamic subsystem allocation, variable binding, qubit pinning against the processor connectivity graph, and the three-language structure.
• Horizon Quantum (@horizon_quantum) post announcing "The Landscape for Quantum Computing Roles in Ireland," Dublin panel, August 20, 2026, with Dr Ray Lloyd (Horizon), Dr Felix Binder (Trinity College Dublin), and Dr Mike Dascal (Fidelity); event listing via luma.com.
• "An Introduction to Beryllium" (company technical blog, published Aug 11, 2026) — horizonquantum.com/resources/blog/an-introduction-to-beryllium.
• Joe Fitzsimons (@jfitzsimons) and QIP 2027 (@QIPConference) posts on X announcing Horizon Quantum as QIP 2027 anchor sponsor (early August 2026); corroborated by qipconference.org/2027 and the QIP 2027 sponsorship prospectus (CQT/NUS).
• Horizon Quantum (@horizon_quantum) post on X promoting the Q2B 2026 Tokyo session with Quantum Machines on real-time execution.
• Horizon Quantum Holdings Ltd., Form 20-F (Shell Company Report), filed March 25, 2026 — full business overview, risk factor cross-reference, director/officer bios, compensation, lock-up terms, IonQ Side Letter terms (5% board-nomination threshold), Maybell Quantum Industries purchase commitment, and inbound early-access demand figures — sec.gov/Archives/edgar/data/2088256/000121390026034058/ea0282908-20f_horizon.htm.
• Investing.com, full transcript of Horizon Quantum's Q2 2026 earnings call (Aug 4, 2026), including CFO Greg Gould's remarks and Q&A with analysts from Needham, Rosenblatt Securities, Craig-Hallum Capital Group, and StoneX — investing.com/news/transcripts/earnings-call-transcript-horizon-quantum-posts-q2-2026-loss-shares-fall-premarket-93CH-4834502.
• Investing.com transcript, "Horizon Quantum Computing at Canaccord Genuity: software-first quantum push" — full transcript of Canaccord Genuity's 46th Annual Growth Conference (published Aug 11, 2026), including CEO Q&A on strategy, hardware trade-offs, internal applications team, and pre-revenue positioning.
• GuruFocus / Investing.com / MarketScreener, Needham analyst coverage initiation (N. Quinn Bolton, Buy, $20 PT, June 3, 2026). Craig-Hallum initiation (Buy, $21 PT, Aug 17, 2026) per TipRanks, Yahoo Finance analyst-actions table, and Daily Political; Wall Street Zen downgrade to Strong Sell (Aug 8, 2026) and Weiss Ratings change to "sell (e+)" (Jun 17, 2026) per the same aggregated coverage summary — these two are rules-based screens, not fundamental research notes.
• Horizon Quantum press release, "Horizon Quantum Announces Participation in Upcoming Investor Conferences" (Aug 6, 2026) — businesswire.com.
• Competitive field data: Quantum Zeitgeist, "Top Quantum Software Companies 2026" (May 2026) and "Quantum Computing Companies in 2026" (Feb 2026) — stack-layer taxonomy and capital-raised figures for Classiq, Riverlane, Q-CTRL, Quantum Machines, Strangeworks. Tracxn company profile for Classiq (funding, founders, competitor ranking listing Horizon as closest competitor). The Quantum Insider, "Top Quantum Computing Companies" (Jul 2026) and "Top Quantum Computing Investors in 2026" (Jun 2026) — Classiq platform description, Riverlane/Rigetti sub-microsecond decoding on Ankaa-2 (Oct 2024), IBM Ventures portfolio (QEDMA, QunaSys, Strangeworks).
• Quantum chemistry resource estimates (third-party, not Horizon): Goings et al., "Reliably assessing the electronic structure of cytochrome P450 on today's classical computers and tomorrow's quantum computers," PNAS 119 (2022), doi:10.1073/pnas.2203533119 — ~4,900 logical qubits, ~10⁹ Toffoli gates, ~73h runtime for P450. Alice & Bob, "Quantum Resource Estimation for Ground State Energy of FeMoco and P450 on Cat Qubits" (Oct 2025) — 2.7M to ~99,000 physical qubits for FeMoco, a 27x reduction vs. the 2021 Google benchmark. Independent scope and timeline assessment via postquantum.com quantum utility map (May 2026) — 2033–2036 for industrially relevant FeMoco; quantum edge confined to the hardest 5–10% of chemistry problems.
• Section 16 insider filings via SEC EDGAR / StockTitan: Form 4 for Joseph Francis Fitzsimons (Mar 19, 2026 — 8,108,696 Legacy shares exchanged for 19,744,585 Class B at $0.00); Form 4 for Gregory M. Gould (285,300 options at $5.13 per 2.43499 Class A shares, 16-quarter vesting from Aug 15, 2025; 35,662 vested as of Mar 24, 2026); Form 3 for Catherine Michele Fitzsimons (May 2026, no holdings reported).
• Dr. Si-Hui Tan: personal academic site sihuitan.com (biography and research interests — Caltech BSc, MIT PhD under Seth Lloyd, MIT Presidential Fellowship, Niels Bohr International Academy, SG100 Women in Tech 2021, NTU College of Science Advisory Board); Google Scholar profile (1,981 citations); ResearchGate publication record including quantum homomorphic encryption work.
• Horizon Quantum Holdings Ltd., Form F-4 registration statement (File No. 333-292737, filed Jan 14, 2026, effective Feb 17, 2026) — management table, officer and director biographies, and pre-close announcement of Catherine Fitzsimons as Chief Legal and Compliance Officer alongside the incoming board.
• Peer valuation context: The Motley Fool quantum-stocks sector coverage (market-cap figures for IonQ, Rigetti, D-Wave, Quantinuum, Infleqtion, late July/early August 2026); 24/7 Wall St coverage of IonQ Q1 2026 revenue ($64.67M).
• Market data (HQ price, 52-week range): Investing.com, Robinhood, Morningstar, StockAnalysis.com — closing price of $19.93 as of Aug 21, 2026; market cap in this report is calculated from the primary share count rather than taken from these vendors directly.

Note on quotations: all quoted material in this report is drawn from primary sources — Horizon Quantum press releases and company website, SEC filings and their exhibits, the granted patent text, and published transcripts of the Q2 2026 earnings call and the Canaccord Genuity Growth Conference (August 11, 2026). Each quotation is attributed inline to its speaker or document and dated where the source permits. No quotation in this report is reconstructed, paraphrased into quotation marks, or attributed to any person on the basis of inference.

One item remains genuinely open at publication: a current Peak XV ownership percentage from a post-dilution filing. Section 19 gives a defensible range (≈12.0%–18.9%) computed from Horizon's own share counts, and Peak XV's exact holding at the de-SPAC appears in their own filing, so the percentage can be computed against whichever denominator a reader prefers — but no company- or Peak-XV-disclosed current figure exists. Everything else in this report is sourced to a primary document read in full.