Table of Contents
October 8, 2026 – Four more quantum companies have cleared DARPA’s review of their plans for an economically useful computer. Their hardware and system designs now face the agency’s final verification phase.
DARPA announced on October 7 that Atom Computing, Diraq, IBM and IonQ have been selected for Stage C of its Quantum Benchmarking Initiative (QBI). They join Microsoft and PsiQuantum, bringing the publicly identified group at this level to six.
The initiative’s central question is whether any quantum-computing approach can achieve utility-scale operation by 2033, meaning that the computational value it delivers exceeds its cost. Stage B examined the companies’ development plans, technical risks and proposed mitigations. Stage C brings independent verification and validation to the hardware, prototypes and system designs on which those plans depend.
Four new entrants
Atom Computing
Atom Computing’s announcement describes potential funding of up to US$300 million for its neutral-atom approach. It does not disclose an initial funded amount. The release also identifies a partner in the evaluation: Microsoft has agreed to provide algorithmic support and error-correction codes.
Atom enters with experimental work addressing a practical problem for neutral-atom machines: continuing a computation while replacing atoms that are lost. In a June preprint on quantum error correction with the toric code, the team reported tests through as many as 90 cycles of syndrome extraction, incorporating lost-qubit replacement and reservoir reloading. The evidence for lower logical error rates with a larger code came from separate, shorter tests of up to eight cycles; the long reloaded runs did not establish that improvement.
Diraq
Diraq disclosed an initial US$51 million tranche to support the next year of work, within potential funding of up to US$300 million. Its silicon spin-qubit program will deliver technical demonstrations, system updates and design work, supported by expanded research and testing infrastructure.
The company’s manufacturing argument has a concrete experimental basis. A July paper in Nature Communications reported coherent control and readout of an eight-qubit array fabricated using a 300 mm CMOS-compatible foundry process, with a controlled-phase gate demonstrated on one adjacent pair. That establishes progress in operating foundry-made devices; reliable entangling operations across the array remain to be demonstrated.
Diraq’s announcement reiterates company targets of 150,000 physical qubits in 2029 and more than two million on a single chip in 2031.
IBM
IBM’s announcement confirms its advancement to Stage C and reiterates its target of delivering a fault-tolerant system in 2029. The release does not disclose an IBM-specific award amount or Stage C duration.
IBM’s published hardware roadmap identifies that system as IBM Quantum Starling, a modular superconducting design targeting 200 logical qubits and circuits containing 100 million quantum gates. Achieving that requires multiple modules to operate together within an error-corrected architecture. The 2029 date and performance specification remain IBM’s targets; Stage C provides a mechanism for examining the engineering behind them.
IonQ
IonQ says its signed Stage C agreement has a potential value of up to US$300 million and runs through 2029. Multiple generations of its Superion systems will undergo independent evaluation. The company explicitly says that funding beyond the initial obligation depends on future appropriations and DARPA funding actions, without identifying the initial amount.
The hardware under scrutiny is significant. Superion, announced in September, uses electronic qubit control developed from Oxford Ionics technology. IonQ reported fabrication of its first integrated 256-qubit chips at SkyWater and the trapping of first ions in prototype systems. Customer deliveries are targeted for 2027.
Those milestones establish fabrication and early prototype progress. Stage C will examine how successive systems perform as IonQ scales the platform.
Microsoft and PsiQuantum already under evaluation
Microsoft and PsiQuantum entered the final phase through QBI’s predecessor, the Underexplored Systems for Utility-Scale Quantum Computing program (US2QC). DARPA announced their selection in February 2025, when agreements were still being negotiated, and explained that the phase had the same technical goals as QBI’s final stage.
Microsoft supplied a recent update on what that evaluation now involves. In its September 22 announcement of its Maryland research center, the company said it was providing a Majorana 2-based system for DARPA testing, with full access to the hardware on site. This is the evaluation of Microsoft’s proposed topological approach, separate from its work with Atom Computing’s neutral-atom machines.
PsiQuantum announced a US$125 million expanded agreement in July. The performance-based award supports evaluation of its photonic hardware designs, components, system performance and software, along with infrastructure in California and Illinois. It follows a US$31.8 million agreement announced in September 2025.
The newer US$300 million figures also require context. DARPA’s original program briefing already listed that amount as Stage C’s potential funding ceiling, subject to conditions. An agreement ceiling, an initial obligation and an expanded award measure different financial commitments.
Further admissions remain possible. DARPA explicitly expects additional Stage C promotions soon. Other companies from its November 2025 Stage B cohort, including Quantinuum, QuEra, Photonic Inc. and Xanadu, are absent from this announcement. The agency describes advancement as staggered; absence here does not establish rejection.
My Analysis
In August, I examined the apparent paradox of DARPA’s two Stage C participants: Microsoft and PsiQuantum were ahead in the evaluation process despite the substantial public hardware records of companies still in Stage B. The program history explained much of that picture. US2QC had started earlier and had deliberately examined underexplored approaches.
The broader QBI process is now bringing four more architectures into that depth of scrutiny. That is meaningful progress. It also reinforces why admission order was a poor guide to technical leadership: the companies were moving through different evaluation timelines.
I would treat Stage C selection as evidence that a development plan has survived serious examination. The practical value of the next phase lies in testing the assumptions that remain.
The economics of the complete machine
The common thread through these announcements is the cost and difficulty of turning a processor into a working system.
Diraq’s foundry compatibility matters because repeatable semiconductor fabrication could make large numbers of qubits practical to manufacture. The engineering question is how that advantage survives the control, readout, calibration and cooling requirements of the complete machine. IonQ’s Superion program makes a related argument from a different physical platform: electronic control and semiconductor manufacturing are intended to make trapped-ion systems easier to reproduce and scale.
IBM’s modular plan puts particular weight on what happens between processors. Atom’s repeated-error-correction work addresses how to keep a neutral-atom system operating while its physical constituents need replenishment. These are different engineering problems, and a physical-qubit count says little about whether a proposed solution will work at scale.
The Microsoft–Atom relationship makes the system boundary especially visible. Microsoft contributes error-correction and algorithmic capabilities to one evaluated architecture while supplying its own hardware for another. A useful assessment needs to follow those dependencies across company boundaries. Buying a processor does not resolve the questions of how its errors will be decoded, which computations it can sustain, or what the supporting infrastructure will cost.
DARPA’s economic criterion is also narrower than corporate profitability. Its published QBI FAQ includes construction and operating costs but excludes R&D expenditure. A machine could therefore satisfy the program’s utility criterion without demonstrating that its developer will recover every dollar spent getting there.
What completing Stage C would establish
The same FAQ contains a distinction that deserves more prominence in coverage of these awards: DARPA does not expect Stage C itself to build the complete utility-scale computer.
Its intended endpoint is sufficient evidence from components and prototypes, with major risks sufficiently reduced, to justify finalizing the system’s design and beginning manufacture. A company might build more during that period, but an operating utility-scale machine is not the required endpoint of Stage C.
That makes IonQ’s evaluation through 2029 compatible with QBI’s 2033 horizon. It also places the latest selections in proportion. These companies are entering the phase that will examine whether they can reach that construction decision.
The distinction matters for readers trying to interpret government validation. DARPA has evaluated information that outside observers may never see. Its FAQ says disclosure of proprietary information requires authorization. Selection provides a useful external signal, but it cannot substitute for public system results when comparing the capabilities available to customers.
The 2033 assessment keeps strengthening
DARPA’s own language has moved considerably since the program began. Its September 2024 briefing described utility-scale quantum computing by 2033 as unlikely. By March 2026, managing director Micah Stoutimore was saying that someone appeared likely to achieve it. In the October announcement, he describes an increasing expectation of success while leaving open which team or technology will get there.
That progression deserves attention because it comes from the organization conducting the evaluations. The optimistic assessment was already public in March; October continues it.
For security planners, however, economic utility and cryptanalytic capability remain different thresholds. A machine that delivers value on a materials or chemistry workload has not thereby demonstrated the resources and reliability needed to attack a particular public-key cryptosystem. This announcement gives us more reason to take hardware development seriously, without supplying a new date for Q-Day.
What buyers should ask next
For organizations evaluating quantum suppliers, Stage C is a reason to ask more specific questions about the path to a useful system:
- Which computation establishes the value? Ask for the workload, required logical performance, runtime and classical comparison that underpin the economic case.
- Which assumptions have been tested? Distinguish measured component or prototype performance from behavior projected for the integrated machine, including the supporting controls and error-correction system.
- What remains to be funded and built? Separate the initial funded work from potential later awards, and ask which technical milestones must be met before construction becomes justified.
The next consequential disclosures will show how far tested hardware can carry that argument. A roadmap earns more credibility each time an assumption becomes a measurement. Stage C is where four more companies will have to make that conversion.
