Quantum Computing: The Next Compute Cycle — Economics, Winners & Investment Map

One governing rule runs beneath all three frameworks, stated once here and applied to every workload claim in §06: a quantum-advantage claim is only as good as the classical baseline it is checked against, and DARPA itself has noted that rigorous classical-baseline comparison is still missing for many proposed quantum applications. Every use-case claim in this report is held to that standard explicitly, not assumed.

The cluster above is loud, but it is not the anchor citation this page is built on. That comes from DARPA's Quantum Benchmarking Initiative (QBI), which is independently evaluating roughly 18 architectures against a 2033 deadline for one precise target: utility-scale quantum computing — a machine whose computational value exceeds its cost. That definition, not a qubit count, is the thesis of this entire page, and it is the same instinct that made Compute as an Asset Class work for classical GPU financing: the investable question is never the size of the machine, it is what the machine is worth once it is running.

DARPA is explicit that QBI is not a winnowing competition — multiple, one, or zero architectures may eventually qualify, and each is evaluated on its own merits rather than against each other. Eleven of the 18 original Stage A entrants advanced to Stage B on November 6, 2025 — Atom Computing, Diraq, IBM, IonQ, Nord Quantique, Photonic Inc., Quantinuum, Quantum Motion, QuEra Computing, Silicon Quantum Computing and Xanadu. Google, Rigetti and HP Enterprise did not advance in that cohort, though DARPA does not treat that as disqualifying, and Google's Stage A window was only around two months long by construction — which complicates reading much into its absence. Full stage map, including the related US2QC track and the new HARQ heterogeneous-architecture program, is in §04. Status: Demonstrated (selection itself) · Evidence: Government (IV&V).

Read the rest of this page through that lens. A Nasdaq ticker, a SPAC close, or a CHIPS Act letter of intent is a capital-formation event — informative about who can fund the race and on what terms (§10–§11), but it is not itself evidence that the underlying machine has crossed DARPA's utility-scale bar. That evidence, such as it exists in 2026, is built one link at a time in the next two sections.

A qubit count answers a marketing question. It does not answer an investment question, because a physical qubit is not a unit of computation — it is a unit of noisy computation that has to survive fidelity loss, error-correction overhead, and decoherence before it produces a logical result anyone can trust. The Quantum Utility Bridge lays out that survival chain explicitly, end to end, so a claim like "Company X has 1,000 qubits" can be located on the chain rather than mistaken for a finish line.

Quantum computing did not have a qubit-count problem in 2026. It had a link-six problem — nobody discloses executable circuit depth the way everybody discloses qubit count, and link six is the one that actually decides what a machine can do.

The next section takes link three — QEC overhead — and builds it out as its own exhibit, because it is the link most often collapsed into a single misleading ratio.

The single most quoted — and most frequently mis-cited — number in quantum computing coverage is a physical-to-logical qubit ratio. The problem is not that these ratios are wrong; each is accurate on its own terms. The problem is that different experiments measure different things at different error targets on different hardware, and a reader who compresses them into one ranked list has manufactured a comparison the underlying data does not support. This exhibit exists to be read one row at a time.

The Willow extrapolation is a projection, not an achieved result — treat it as one. The 2024 Nature paper's own Figure 1d extrapolates that reaching a 10⁻⁶ logical error rate would require a distance-27 surface code using 1,457 physical qubits — a number the paper itself frames as a scaling projection from its measured Λ = 2.14 error-suppression factor, not a system anyone has built. Status: Projected · Evidence: Peer-Reviewed (as an explicit extrapolation, not a claimed result). The paper also notes the reverse holds: halving the physical error rate would improve distance-27 logical performance by four orders of magnitude — meaning near-term hardware gains carry outsized leverage through error correction, which is a genuine reason for optimism about the trajectory even though the 1,457-qubit machine does not exist.

On the two fidelity claims that most often get compared incorrectly. Quantinuum's 99.921% two-qubit gate fidelity is a system-wide, all-pairs average measured on Helios, an actual commercially available 98-qubit machine — Available + Company Disclosure. IonQ's 99.99% figure comes from its Electronic Qubit Control technology demonstrated on small-scale R&D prototype hardware intended to underpin a future 2026-generation system, not a measurement available to a customer today — Announced/Projected (for commercial deployment) + Company Disclosure. Both figures are real and both are honestly reported by their companies. Neither should be read as "the higher number wins": one describes a production system at scale, the other a lab demonstration on a small setup, and the appropriate like-for-like comparator to Helios is IonQ's own production Forte system, which the company has separately reported at roughly 99.65% two-qubit fidelity.

Seven distinct qubit modalities are being pursued commercially in 2026 — superconducting (IBM, Rigetti, Google, IQM), trapped-ion (Quantinuum, IonQ), neutral-atom (QuEra, Atom Computing, Infleqtion), photonic (PsiQuantum, Xanadu, Photonic Inc.), silicon-spin (Diraq), topological (Microsoft) and annealing (D-Wave, a fundamentally different computational model aimed at optimization rather than general-purpose gate-based computing). DARPA's Quantum Benchmarking Initiative is the closest thing to an independent, cross-modality referee any of them have — which is precisely why its own language about what QBI is and is not matters more than any single company's roadmap slide.

Hardware progress and commercial readiness are not the same claim, and 2026's two best market studies triangulate on exactly that gap using completely different methods. BCG's enterprise-spend modeling (§08) finds value concentrated in a narrow set of high-impact use cases accessible to roughly 100–200 logical qubits — chemistry and materials simulation specifically, not a general threshold. Separately, IQM's State of Quantum 2026 report — independently researched and authored by The Quantum Insider, based on a validated survey of 107 senior practitioners across AMER, EMEA and APAC plus 19 leadership interviews at organizations including Airbus, BMW, Moderna and Argonne National Laboratory — finds that 89% of surveyed enterprises report hands-on quantum work, but only 10% report limited production use and just 3% have reached scaled production deployment. A new composite Quantum Readiness Index scores the global market 58 out of 100 — the "Developing" tier on a five-tier scale from Aware to Leading. Status: Demonstrated (survey results) · Evidence: Independent Research.

Two completely different methods — top-down enterprise-spend modeling versus bottom-up practitioner survey — landing on the same conclusion is a stronger signal than either alone. Neither is promotional: BCG is modeling a market it wants clients to invest in, and IQM is a commercial quantum vendor sponsoring survey research conducted independently by The Quantum Insider — a relationship worth naming directly, since IQM is also a subject of the capital-markets exhibits in §10–§11. The independence of the research process, not the absence of a commercial sponsor, is what makes the finding usable.

The report's sharper finding sits inside the aggregate number: hiring, budget and pilot activity are all running ahead of proprietary output and scaled deployment. Only 9% of organizations maintain a resourced quantum intellectual-property program — meaning most of the 89% doing "hands-on" work have not yet produced anything defensible from it. That is the algorithm gap in one statistic: enterprises can access quantum hardware today; very few have found an algorithm on it that clears the classical-baseline bar stated in Framework at a Glance.

BCG's frequently-cited "100–200 logical qubits" figure is not a general-purpose threshold for quantum advantage — it is a resource estimate specific to chemistry and materials simulation, the single use case BCG's base-case $2.5B-by-2030 market scenario is built around (§08). Applying it to optimization, finance or cryptography would misstate what BCG's own modeling supports. The table below separates workloads by the resource, algorithm and classical comparator each actually requires.

Quantinuum's multi-year partnership with Oracle, announced alongside its Q2 2026 earnings, is the clearest 2026 exhibit of how enterprises are actually meant to reach quantum hardware: not by buying a machine, but by renting access to one embedded inside infrastructure they already use. The deal places a Helios system directly inside an Oracle OCI AI data center — the first Helios sited on U.S. soil outside Quantinuum's own facilities, integrated with OCI compute, storage, networking and security for hybrid quantum-classical workflows. This is Announced/Planned, not an operating installation: Oracle is purchasing the system under a multi-year agreement, cloud-activation revenue is modest in 2026, and the bulk of revenue recognition is expected on system delivery in early 2027. Status: Announced/Planned · Evidence: Company Disclosure (joint press release).

This is one of at least two competing commercialization archetypes visible in 2026. Quantinuum and IonQ both push toward cloud-delivered QaaS — Quantinuum's Nexus platform now serves 180 organizations, per its Q2 2026 disclosure — while IQM pursues an on-premises deployment model that gives customers direct ownership and control of their quantum infrastructure, evidenced by 26 systems sold and 17 delivered globally, including a first major U.S. government delivery to Oak Ridge National Laboratory. Neither model has established itself as dominant; enterprises appear to be buying both access modes simultaneously depending on workload sensitivity and integration requirements.

McKinsey's enterprise-spend data gives the clearest picture of how seriously buyers are treating this, independent of which access model they choose: 33% of surveyed companies allocate more than $10 million annually to quantum computing initiatives, 7% spend more than $50 million, and the largest individual disclosed budget reaches $200 million. Status: Demonstrated (survey data) · Evidence: Independent Research. Set against the algorithm gap in §05 — where only 3% of hands-on enterprises report scaled production deployment — this spend looks less like proof of near-term ROI and more like real-option positioning: enterprises paying to keep a seat at the table for whichever architecture and algorithm combination crosses the utility-scale bar first.

Search "quantum computing market size" and the numbers on the first page span three orders of magnitude, because they are not measuring the same thing. This exhibit keeps three independently sourced 2026 reports separate on purpose — reconciling what each one actually measures is more useful than picking the biggest number.

Four layers make up the quantum value stack: hardware and components (qubits, cryogenics, control electronics, specialized fabrication), QEC, control and systems integration (the error-correction codes and classical control stacks built in §02–§03), algorithms, software and QaaS (the layer §05–§07 describe as still gapped), and applications (the commercial end-use layer §06 tries to identify first movers in). Today's scarcity rents sit almost entirely in the first layer — hardware and components — reflected in every capital-raising route documented in §10: IPO proceeds, SPAC proceeds, CHIPS Act equity, and M&A are all currently flowing toward companies building physical machines, not companies selling algorithms or applications.

Five distinct routes to capital are all live simultaneously in quantum computing in 2026: a traditional IPO, a SPAC merger, ongoing public-market access for already-listed pure-plays, a novel government minority-equity instrument, and strategic M&A. This exhibit keeps its columns to what is actually observable in each disclosure — capital source, instrument, capital raised, implied valuation, dilution, strategic terms, use of proceeds and balance-sheet runway. It deliberately does not attempt a "cost of capital per business model" figure: venture funding, government equity, SPAC proceeds and traditional IPO proceeds are not reducible to one comparable WACC-style number, and inventing pseudo-precision there would undercut the exhibit rather than strengthen it.

A bare revenue comparison across these companies invites a false conclusion. IonQ's $80.1M Q2 2026 revenue looks like ten times Quantinuum's $8.0M — but IonQ's own disclosures show that figure blends compute, networking, sensing and acquired-business revenue (the company discloses roughly 60% commercial, 50% international and 25% multi-product for the quarter, without breaking out how much is core quantum-compute revenue specifically), while Quantinuum's and D-Wave's revenue sit closer to pure compute/QaaS. Two methodology rules make this exhibit audit-proof rather than another vendor-friendly ranking: (1) every classification below is tagged Reported / Explicitly Disclosed / Estimated / Not Disclosed, and a company's own stated breakdown is never silently blended with an outside estimate; (2) no government-versus-commercial split is inferred just because a customer's identity is known — where a company has not disclosed the split, the cell says "Not Disclosed," not a guess.

On enterprise value versus market cap. Every one of these companies now carries a large net-cash position relative to revenue, which makes market cap ÷ revenue a weak standalone metric — but building an Enterprise Value comparison responsibly requires verified share-count and market-cap data this page has not independently confirmed for all five names in this pass. Rather than publish an approximate EV figure, the table below sticks to disclosed cash, spend and burn figures, and flags where a runway estimate is A.L. Capital's own arithmetic rather than company guidance.

The obvious next question after "which architecture wins?" is "which stock is that?" — and this page deliberately does not answer it with a rating. Each pure-play quantum equity — IONQ ↗, QNT ↗, RGTI ↗, QBTS ↗ and IQMX ↗ — already carries its own full equity view and conviction rating elsewhere on this site; this page cross-links to each rather than re-rating it. What follows instead are two taxonomies. The first — exposure archetypes — describes what business a company is actually in. The second — a financing overlay — describes how it is capitalized. They are kept apart because "government-supported" is a funding characteristic, not an economic exposure, and conflating the two muddies both: two companies in the same exposure archetype can sit on opposite ends of the financing overlay, and treating "government-backed" as if it described a business model would be exactly that conflation.

A falsification section that lists company-specific disclosure changes as evidence against the thesis is not actually testing the thesis — it is testing whether individual managements stay consistent, which is a different and much weaker claim. The seven falsifiers below attack the Quantum Utility Bridge chain built in §02 directly, link by link. If several of these start showing up simultaneously across independent architectures, that is a reason to revisit this page's thesis; a single company's earnings miss or a single quarter's stock move is not.

This page issues no individual equity ratings of its own — Quantinuum, IonQ, Rigetti, D-Wave and IQM (plus IBM) each already carry a full equity view and conviction rating elsewhere on this site, cross-linked throughout §12 rather than restated here. What follows is every publicly investable route into this page's subject that this research pass could verify — equities and the ETFs that wrap them — read against the archetype (§12, Exhibit 12) and financing-overlay (§12, Exhibit 13) taxonomies above, not a recommendation to hold any of them.

Position sizing and time horizon should follow from the classical-baseline discipline stated in Framework at a Glance and tested against the seven falsifiers in §13 — not from which company issued the loudest press release in a given quarter, or which vehicle in the table above is most convenient to buy. A reader applying this page to a specific ticker or fund should locate it on both §12 taxonomies independently, then bring their own valuation work; this page deliberately stops short of doing that work for them.

What sizing that exposure correctly actually requires, without naming a ticker, is three checks. Correlation, not diversification: a portfolio that already holds AI-infrastructure exposure (per Compute as an Asset Class) is not adding an offsetting risk factor by adding quantum-computing exposure on top of it — both cycles draw on the same pool of capital-markets enthusiasm for compute-adjacent technology, so a drawdown in one is a plausible catalyst for a drawdown in the other, not a hedge against it. Duration and liquidity: every publicly listed name in §12 is pre-profit or early-profit on an adjusted basis (§11), priced on multi-year roadmap execution rather than current earnings, and several carry the government-policy dependence documented in §10 on a timeline this page's author does not control. Concentration at the thesis level, not the position level: a bet on an entire nine-link chain clearing an economic bar is a different risk than a single company's earnings call, and sizing it like an ordinary equity position understates how much of the outcome still depends on links this page tags Projected rather than Demonstrated (§02). None of this argues against holding the exposure — it is the argument for why sizing it is a portfolio-construction question rather than a stock-picking one.

A higher coefficient points toward the diversified-incumbent or UCITS-ETF end of the spectrum above; a lower one tolerates the concentrated pure-play end. Translating that coefficient into an actual position size, alongside the rest of a portfolio, is what a Strategic Session does next.

This page sits alongside three other Intelligence pages that touch adjacent, but distinct, compute and capital-formation questions. We map them explicitly so a reader knows which question each page answers, rather than mistaking overlapping subject matter for disagreement.