DOE Wants Fault-Tolerant Quantum by 2028. The Roadmaps Show How Hard That Target Is to Define.

Two announcements landed within twenty-four hours of each other last week. On June 22, the White House signed an executive order directing the government to pursue a fault-tolerant quantum computer and deliver it to a federal laboratory. On June 23, the Department of Energy attached a goal and a year: logical qubits in the low hundreds, error-corrected and fault-tolerant, by 2028, standing up as a scientific user facility.

The instinct is to ask whether 2028 is realistic. It is the wrong question, or at least an incomplete one. Pull the vendor roadmaps and you find that some of them already promise more logical qubits, sooner, than DOE is asking for. IonQ's public roadmap shows 800 logical qubits in 2027 and 1,600 in 2028. The harder question is the one hiding under the word "logical." A logical qubit on IonQ's roadmap and a logical qubit on IBM's are not the same object, and DOE has not yet published the rules that decide which one its 2028 facility will require. That definitional gap, not the calendar, is where this story actually lives.

What was signed, and what it says

The two orders are easy to conflate and worth separating. The first, Executive Order 14413, "Ushering in the Next Frontier of Quantum Innovation," establishes a program called the Quantum Computer for Application Development and Discovery Science Effort, or QC-ADDS. Coordinated through the Assistant to the President for Science and Technology, the order directs the government to "pursue development of a quantum computer at a scale intended to initiate the era of quantum-enabled scientific discovery," and to deliver "at least one such computer to a Department of Energy facility." The framing is national: quantum information science, the order argues, "will... bolster national security," and "as other nations move quickly to challenge American leadership, the United States must take a cohesive, whole-of-government approach to accelerate deployment and commercialization of quantum computing, sensing, and networking."

A companion order signed the same day points the other direction in time. Rather than racing toward a quantum computer, it braces the country against one. The White House fact sheet on securing the nation against advanced cryptographic attacks directs federal High Value Assets to adopt post-quantum cryptography for key establishment by the end of 2030 and for digital signatures by the end of 2031, with a NIST migration pilot due by the end of 2027. One order is an accelerator; the other is a hedge against the same physics.

QC-ADDS is the directive. The machine itself runs through a separate vehicle. On June 23, DOE's Office of Science announced Quantum Genesis, the initiative it will use to execute the order. Its own headline describes a plan to "create and deploy the world's first scientifically relevant, fault-tolerant quantum computers." The wording matters. DOE did not say "real" quantum computing or "useful" quantum computing; it said scientifically relevant, deployed as a user facility, with the target framed as by 2028 rather than on a fixed calendar date. That precision is not pedantry. It is the difference between a deadline and an aspiration, and DOE chose its words with that distinction in mind.

Quantum Genesis is not the same thing as the Genesis Mission. The Genesis Mission, established under Executive Order 14363 in November 2025, is the broad AI-for-science umbrella DOE runs across seventeen national laboratories. It is the program behind the Lux and Solstice systems and the $293 million round of national science challenges SCN covered this spring. Quantum Genesis is the quantum pillar nested inside it. The connective tissue is Darío Gil, the DOE Under Secretary for Science who directs the Genesis Mission and arrived from IBM Research, where he ran the division that builds the quantum machines now setting the commercial pace. Wire coverage of last week's announcement blurred the two Genesis programs together; they are parent and child, not synonyms.

What "logical qubit" means

A logical qubit is not a physical one. Physical qubits (the superconducting circuits, trapped ions, or neutral atoms that hold quantum information) are fragile, and they lose their state to noise far too quickly to run a long computation. Error correction is the workaround: you bundle many physical qubits together so that the group behaves like a single, far more stable qubit, with the redundancy used to detect and fix errors faster than they accumulate. That stabilized group is a logical qubit. DOE's own language describes the target as "strong logical qubits... groupings of error-corrected physical qubits." Fault tolerance is the threshold where that scheme actually holds, the point where adding more physical qubits makes the logical qubit better rather than just bigger, so a computation can run to completion without errors swamping the result.

The cost of that stability has long been the headline number. Under surface-code assumptions, a single logical qubit can demand on the order of a thousand physical qubits, which is why low hundreds of logical qubits has historically implied hundreds of thousands of physical ones. But that ratio is code-dependent, not a law of nature. IBM's quantum low-density parity-check (qLDPC) codes cut the overhead roughly tenfold, and trapped-ion vendors claim lower still. The recent proliferation of error-correcting codes beyond the surface code is, in large part, a hunt for exactly that: schemes that buy a usable logical qubit for fewer physical ones. The trouble is that those schemes do not all produce a logical qubit of equal quality, and "logical qubit" on its own says nothing about which scheme, or what error rate, a vendor has in mind.

DOE's public Quantum Genesis page gives the target as "logical qubits numbering in the low hundreds" by 2028, deployed as a scientific user facility it calls a National Quantum Supercomputing User Facility. A sharper version of the number appears in a Department of Energy request for information issued in May, which floated a configuration of roughly 150 to 250 logical qubits at a logical error rate of 10⁻⁸ with a universal instruction set. That RFI is the closest thing to a hard specification anyone has seen. It is also, by its own terms, market research rather than a solicitation, which means none of those figures is binding yet. The host site has not been named; DOE has said only that the computer will go to a department facility.

This is the band where quantum computing stops being a physics demonstration and starts behaving like a quantum-accelerated supercomputer: enough error-corrected logical qubits to run algorithms a classical machine cannot, wired into the classical HPC and storage that turn raw qubit operations into science. The target is credible as a destination. Whether any 2028 machine clears DOE's intended bar, as opposed to its headline number, is a separate question, and it turns on specifications DOE has not published.

The yardstick

Line the public roadmaps up against DOE's 2028 goal and the numbers do not tell a single clean story. Read by logical-qubit count alone, two trapped-ion programs already promise to clear the bar before DOE wants it cleared. Read by how rigorously each vendor defines a logical qubit, the picture inverts.

Two things fall out of the table. The first is that DOE's 2028 goal is not ahead of every public roadmap. It is ahead of IBM's, which is the most concrete large-scale fault-tolerant program in the field, and behind IonQ's on raw count. The second is that the roadmaps are not measuring the same thing. IonQ's logical qubit rests on the argument that very high physical fidelity lets you treat a small group of physical qubits as a logical one, with little of the surface-code overhead IBM budgets for. IBM's logical qubit is a qLDPC-encoded unit costing several hundred physical qubits per ten logical, sized for roughly a hundred million error-free gates. Both are called "logical qubits." They are not interchangeable, and a procurement that names a count without naming a definition cannot tell them apart.

IBM remains the cleanest dated benchmark, even if it is no longer the only one. Its more-than-$10 billion, five-year investment plan, disclosed in a May Form 8-K and spanning R&D, capital spending, partnerships and manufacturing through 2030, frames Starling, a 200-logical-qubit fault-tolerant machine, for 2029 in Poughkeepsie, reached through a stepped sequence (Loon, Kookaburra, Cockatoo, a Starling proof-of-concept in 2028). On the opposite end sit Google and PsiQuantum: serious fault-tolerance programs that have deliberately declined to publish dated logical-qubit counts, on the view that a count without a rigorous definition behind it is closer to marketing than to a milestone.

Aggressive dates colliding with qubit-definition questions is a familiar pattern. When Microsoft pulled its own fault-tolerance timeline to 2029 on the strength of its topological-qubit work, physicists were still debating whether the underlying qubit had been demonstrated at all. The lesson was not that the date was wrong. It was that a date and a working, well-defined qubit are different claims, and the distance between them is where the engineering lives.

None of this settles whether a 2028 machine arrives. It reframes the risk. The question is less whether some vendor can ship a few hundred things it calls logical qubits by 2028, and more whether those qubits will be good enough, by a standard DOE has yet to set, to run the science a classical supercomputer cannot.

The September trigger

The detail that turns this from an announcement into a procurement event is buried in Section 4(c) of the order. Within 90 days, which puts the date around September 20, the Secretary of Energy "shall identify the technical specifications required for a QC-ADDS... and shall publicly release a summary of those specifications, as appropriate."

The order itself does not mandate logical-qubit counts or error-rate thresholds. It directs DOE to define "technical specifications," and leaves the content of those specs open. That is what makes the September release the document worth watching, because it is where the definitional gap gets closed, or left open, on the record. The May RFI floated 150 to 250 logical qubits at a 10⁻⁸ error rate with a universal instruction set, but an RFI is market research, not a binding specification. The September summary is the first chance to see whether DOE turns those exploratory numbers into qualification rules: a logical-qubit floor, yes, but also the error rate that defines a qualifying logical qubit, the gate depth a workload has to survive, whether the instruction set must be universal, and whether the spec is written to be modality-neutral. Each of those choices is a filter. Pin the error rate and gate depth where only a rigorously encoded logical qubit can reach by 2028, and IonQ's headline counts may not qualify even though they exceed DOE's number. Leave the definition loose, and a count that clears the bar on paper may underdeliver as a working facility.

There is a precedent for this kind of forcing function recent enough that the people building these systems lived through it. The exascale procurement RFPs of the last decade did not just buy machines. They set performance and power-efficiency requirements that influenced which accelerators got built and which vendors won the following decade of supercomputing revenue. A federal specification with a date attached tends to shape a supply chain, not only a purchase. The September release is where DOE's quantum target starts behaving like a procurement signal, and the definition it lands on will matter as much as the number.

That contest sits inside a wider one. By a 2023 CSIS tally, China had announced roughly $15.3 billion in government quantum investment against about $3.8 billion in public US funding, though the comparison comes with heavy caveats: it counts announced government spending, the underlying figures are uneven across countries, and it omits the substantial private investment that weights toward the US. The EU's Quantum Flagship runs €1 billion over ten years alongside quantum integration into EuroHPC. Domestically, the 2028 goal runs in parallel with the $2 billion the Commerce Department committed to domestic quantum supply through foundry and equity arrangements this spring. The 2028 target is an industrial position as much as a scientific one.

EO 14413 carries a second 2028 marker alongside the computing goal: the Secretary of War is directed to field three next-generation quantum sensor projects by September 30, 2028. What the order sets is an effort and a 90-day clock for specifications, not a binding delivery deadline; the 2028 system goal is DOE's, attached through Quantum Genesis. The September summary is where the headline number meets a definition, and where it becomes possible to say which roadmaps are bidding on the same machine DOE thinks it is buying.