The document, in plain English
On May 15, 2026, the Department of Energy's Office of Science issued a formal Request for Information seeking vendors capable of delivering a scientifically useful fault-tolerant quantum computer to a national laboratory by 2028. The RFI asks respondents to describe the architecture, error-correction approach, performance benchmarks, deployment timeline, and operational support model for a system that would be integrated into the national lab compute environment alongside classical HPC.
That is the entire document, more or less. There is no money attached. There is no commitment from DOE to issue a solicitation. There is no preferred vendor named. By the formal standards of federal procurement, the RFI is the lowest-stakes thing the government can publish.
It is also, in my view, the most important quantum document of 2026.
Why an RFI is more durable than a press release
The reason this matters is that an RFI is not a press release. It is a federal procurement document. RFIs precede solicitations by twelve to eighteen months. They are written by procurement officers and program managers who have been told — usually by the program office, sometimes by Congress, occasionally by both — that a program is coming and that the agency needs to start identifying which vendors can credibly respond.
When a federal agency writes the phrase "fault-tolerant" into an RFI with a specific year attached, it is doing two things simultaneously. It is signaling to the field that a procurement is coming on roughly that timeline. And it is beginning the process of writing the comp spec that the eventual solicitation will use to score bids.
The vendors who answer this RFI seriously will be the vendors whose architectures shape the eventual spec. The vendors whose architectures shape the spec will be the vendors who win. This is how federal procurement has worked for sixty years across every major capability acquisition, from radar to nuclear submarines to the F-35. The quantum sector has not had a document like this until last Thursday.
The 2028 filter
The single most analytically useful thing the RFI does is publish a date. "A fault-tolerant quantum computer at a national laboratory by 2028" is a binary filter applied to every public roadmap in the sector. Two years of design, fabrication, validation, and deployment. The fault-tolerance requirement implies error correction at scale, which implies hundreds of logical qubits operating below threshold, which implies the kind of physical-qubit counts and gate-fidelity floors that only a handful of vendors can plausibly hit on this timeline.
Working through the public roadmaps:
IBM Starling targets 200 logical qubits by 2029. That is a year late for the DOE date, but the architecture and the fabrication base are mature enough that an accelerated 2028 demonstration is credible. IBM also has the customer side of this trade already running through Cleveland Clinic, RIKEN, and multiple national-lab partnerships. Of any vendor in the field, IBM has the most developed deployment muscle. They are the favorite to respond credibly.
Quantinuum Apollo targets fault tolerance at a similar 2029 vintage. The QCCD trapped-ion architecture has the highest gate fidelities in the industry today (99.99 percent claimed on best systems). Quantinuum's integration with Honeywell's manufacturing apparatus is the closest thing the quantum sector has to a defense-prime supply chain, and the upcoming IPO will give them the balance sheet to pull the date forward. They are the second favorite.
PsiQuantum targets utility-scale photonics at the Illinois Quantum and Microelectronics Park (IQMP) on a 2028-2029 timeline, with the Chicago site already under construction. Photonic architecture is the most ambitious bet on the board: if it works, the qubit count scales fast; if it doesn't, the timeline slips a decade. They are the only credible third response, and the only respondent whose architecture matches the DOE date precisely.
Working down from there, the math gets harder. Atom Computing and QuEra have the most aggressive neutral-atom roadmaps in the public field, but neither has demonstrated the error-correction stack at scale required for a 2028 fault-tolerant system. IonQ Tempo is targeted at 2027 but is not fault-tolerant; the 10,000-physical-qubit single-chip narrative is a 2027-2028 aspiration that has slipped before. Atom, Pasqal, Diraq, Photonic Inc., Nord Quantique, and the Stage A QBI cohort are mostly 2029-2030 stories. D-Wave and Rigetti are not in this race; the annealer and the superconducting-NISQ architectures respectively cannot be repackaged as fault-tolerant systems on a two-year horizon.
That leaves three names with a credible 2028 response, six names with a credible 2029-2030 response that would have to argue for date relief, and the rest of the field looking for a different procurement vehicle.
The procurement spec, written in advance
DOE RFIs that name a date and a technology this specifically are usually written against a target architecture. The language in the May 15 RFI — fault-tolerant operation at scale, integration with national lab classical compute, scientifically useful applications in chemistry and materials — maps cleanly to the published IBM Starling and Quantinuum Apollo roadmaps. Not coincidentally. RFI language is shaped by months of pre-procurement conversations with the most likely respondents, and both IBM and Quantinuum have been in serious engagement with the national lab system for years.
The implication is that the RFI is, functionally, the comp spec for a Starling-versus-Apollo bake-off. PsiQuantum is the wild card. Other respondents are invited to participate, but the architectural reference points are written against the favorites. This is not a value judgment about the DOE process. It is how federal acquisition works when the program office has already done its homework.
What this means for the cohort
The valuation consequence is mostly subtractive. The federal procurement narrative has been a meaningful component of the multiple every public quantum name carries. QBTS, IONQ, RGTI, and ARQQ all reference government adjacency in their investor materials. The Stage A QBI vendors (Anyon, Nord Quantique, Diraq, Quantum Brilliance, Photonic Inc., Quobly, Silicon Quantum Computing, Quantum Source, IBM and Quantinuum themselves, and a handful of others) trade in part on the expectation that the DARPA process is a leading indicator of subsequent agency procurement.
The DOE RFI rewrites that expectation. It says: the next major federal procurement for a working machine, on a published timeline, will be fought between two or three architectures. The rest of the field is welcome to compete on a longer horizon, in a different category, or in a different country. Every other narrative — sovereign demand, R&D contracts, government test beds — still applies. But the single-largest piece of federal procurement signal in the sector now points to two vendors and one wild card.
That is not necessarily bearish for the rest of the field. It is, however, a tightening of the bull case that the multiple compression in QBTS, RGTI, and similar names already reflects. The DOE date does not create the multiple compression. It validates it.
The IBM case
IBM is the position with the most upside from the RFI and the least share-price consequence from the outcome. Quantum is immaterial to IBM's stock — the segment is sub-one-percent of revenue, the R&D budget is sub-three-percent of total. A DOE contract at one to two billion dollars across a five-year period is interesting margin but irrelevant to the parent-company multiple.
What it would matter for is the credibility of the Starling roadmap, which has implications for the commercial customer pipeline. Cleveland Clinic, RIKEN, U Tokyo, the named-account pipeline through the IBM Quantum Platform — these customers buy the roadmap as much as they buy the current hardware. A DOE contract stamps the roadmap with federal validation in a way no other customer can. That tightens the funnel for the next tier of enterprise wins.
Our view: IBM is the favorite to win at least one of the two slots if DOE awards multiple respondents (which RFIs of this scale often do). We hold the IBM weighting unchanged. The asymmetry is real but the public stock isn't the right instrument.
The Quantinuum case
Quantinuum is the position with the most direct share-price exposure to the RFI outcome. The IPO has not priced yet but the book is being built against a thesis that includes federal procurement upside. If Apollo is named or implied as a preferred respondent, the IPO prices higher. If Apollo is sidelined for IBM or PsiQuantum, the IPO prices lower.
The QCCD architecture has the highest fidelity floors in the public field, which is a genuine technical advantage for the fault-tolerance regime that DOE has described. The Honeywell manufacturing base, the U.S.-domestic supply chain for ion-trap chips, and the Defense Trusted Foundry adjacency through Honeywell are specific reasons DOE would prefer Quantinuum to a non-US competitor in a final head-to-head. These are the structural advantages a procurement officer values.
The risk is fidelity-claim verification. The 99.99 percent two-qubit number is subject to scope conditions that DOE will press on, and an independent IV&V against the DOE's definition of fault tolerance is the test the IPO book has not yet priced. Our view: Quantinuum is the most exposed name to a positive or negative resolution of the RFI process. If the IPO comes at fourteen billion or below, that is the level at which the asymmetric upside compensates for the verification risk.
The PsiQuantum case
PsiQuantum is the only privately-held name in the credible-respondent group. The company has been the most public about a 2028 timeline for a long time, and the Illinois site is the most concrete piece of national-scale quantum infrastructure under construction in the United States.
What makes PsiQuantum the wild card rather than a co-favorite is the photonic architecture itself. The error-correction overhead is high, the cryogenic single-photon detector requirements are demanding, and the cumulative system complexity is higher than competing modalities. If the architecture works at the scale PsiQuantum has promised, it is the only one that crosses the million-physical-qubit threshold this decade. If it doesn't, the company is in an extremely difficult position because the architecture cannot easily be repackaged.
For private-market investors, the PsiQuantum thesis after this RFI is the same as before: high-conviction, binary, and on a timeline that the DOE has now publicly shortened. The position size should reflect the binary nature. Our view is that anyone holding PsiQuantum exposure through secondary or growth-fund vehicles should expect the carrying values to move materially in either direction over the next eighteen months based on the IQMP commissioning progress.
Who loses
The most exposed names are the public Stage A QBI vendors whose valuations have included a federal-procurement option that this RFI just priced out. We would include in this group most of the Stage A field that does not have an obvious 2028 response. The specific public names — Atom Computing, QuEra (assumed listed by 2027), Diraq, Nord Quantique, Quantum Brilliance — are not bad businesses. They are simply not in the 2028 trade.
The Stage A names that retain a federal procurement narrative are the ones with sovereign-adjacency arguments outside the DOE process: IQM (German and Finnish sovereign demand, EU programs), Riverlane (error-correction software for any respondent), and the U.S. national-lab in-house programs (Sandia, Oak Ridge). These are real businesses with real customers. They are simply not in the 2028 trade either.
The counterargument
The case against this thesis has three forms.
One. "It's just an RFI. No contract attached. Don't overweight it." This reads federal procurement the way a retail investor reads a 10-K: only the numbers matter. But federal acquisition is shaped at the RFI stage. By the time a solicitation drops, the comp spec is locked. The vendors who shape the RFI win the solicitation. Treating the RFI as procedural noise is the most common analytical mistake outside-the-Beltway investors make.
Two. "DOE will widen the field to include emerging architectures." This is possible — DOE has done so before — but the 2028 timeline doesn't allow it. Architectures not at TRL 5 by mid-2027 cannot credibly respond to a solicitation that requires deployment in early 2028. The timeline pre-filters the field whether or not DOE wants it to.
Three. "Multiple parallel programs reduce the importance of any one." True directionally — DARPA QBI, AFRL Quantum Networks, DOE Quantum Networks, and several agency-specific programs are all running. We agree that the DOE RFI is one signal in a constellation. We disagree that it is fungible with the others. The DOE RFI is the first program to combine a published 2028 fault-tolerance target with a national-lab deployment site. Nothing in the constellation has the same specificity.
What we're watching
The conversion test is whether this RFI converts to a solicitation by Q4 2027. Twelve to eighteen months from May 2026 puts the solicitation drop window at May 2027 through November 2027. Earlier is faster than the federal-procurement mode; later is more typical. A solicitation drop before Q3 2027 is a strong signal that DOE is moving aggressively. No solicitation by mid-2028 would invalidate the thesis.
We are also watching for: which vendors publicly confirm they intend to respond (signals confidence), the IBM Quantum Summit November 2026 roadmap update, the Quantinuum IPO pricing and aftermarket trading (proxy for market faith in the 2028 narrative), the IQMP Phase 1 commissioning milestones, and any agency-adjacent RFI or RFP that references a comparable spec.
Disclosure
Connor Reuter runs Quantum Ledger. He is an investor at Caruso Ventures, a Single Family Office. As of publication, Caruso Ventures has no direct equity positions in IBM, IonQ, D-Wave, Rigetti, Quantum Computing Inc., or Arqit. Caruso Ventures has indirect exposure to private quantum names through growth-stage fund commitments that include PsiQuantum and several Stage A QBI vendors. Views expressed are the author's and do not constitute investment advice.
The Ledger publishes one signed essay each week, plus the daily brief at six in the morning Mountain. To follow these calls over time, the archive retains the full update history; the Ledger Score is the underlying scoring framework; the DARPA QBI tracker is the closest comparable to what a DOE solicitation will look like when it drops.