On June 22, 2026, Executive Order 14413 — Ushering in the Next Frontier of Quantum Innovation — was signed. Coverage of it focused almost entirely on the politics.
That's a shame, because the more interesting reading is technical. Policy documents like this get drafted with heavy input from people who run quantum programs, and what they ask for is a fairly candid map of what the field still can't do. Read it that way and EO 14413 is less an announcement than a problem list.
Here's what stands out.
Nobody can reliably measure how good a quantum computer is
The most quietly remarkable line in the order directs the Department of Energy to establish, within 180 days, "a national center to develop the tools and capabilities required to accurately assess the performance of quantum computing systems."
Sit with that. Roughly forty-five years after Feynman proposed quantum computers and seven years after the first quantum supremacy claim, a government is standing up an institution because we do not have a trustworthy way to answer "is this quantum computer any good?"
This is not bureaucratic throat-clearing. It's real, and anyone who has tried to compare two QPUs has run into it. The problem is that every vendor reports different numbers:
- Qubit count is nearly meaningless alone. A hundred bad qubits can be worse than twenty good ones for real work.
- Quantum Volume bundles qubit count, connectivity, and error rates into one figure — but it saturates, and it says little about how a device handles the specific circuits you care about.
- Gate fidelity is usually quoted for isolated one- and two-qubit gates, which systematically flatters a chip. Errors compound differently when gates run in parallel, and crosstalk doesn't show up in the headline number.
- Error-mitigated results can be inflated downstream by the mitigation itself. A 2026 analysis of artefactual improvements in zero-noise extrapolation found the technique can manufacture convincing gains that reflect no underlying physics at all.
- CLOPS and similar throughput metrics measure speed, not accuracy — a fast machine that returns noise is not useful.
- "Algorithmic qubits" and other vendor-defined metrics aren't comparable across vendors, which is frequently the point.
The order also asks for a mechanism to share information between agencies "to improve the Government's ability to assess commercial quantum computing capabilities" — which is a polite way of saying that vendor marketing claims are currently hard to verify independently.
If you're evaluating hardware yourself, the practical lesson is to stop trusting any single number and benchmark against your own workload. You can pull real device properties directly rather than relying on a press release:
from qiskit_ibm_runtime import QiskitRuntimeService
service = QiskitRuntimeService()
backend = service.least_busy(operational=True, simulator=False)
print(f"Backend: {backend.name}")
print(f"Qubits: {backend.num_qubits}")
print(f"Basis: {backend.basis_gates}")
# Per-qubit error rates vary enormously across a single chip
props = backend.properties()
errors = [(q, props.readout_error(q)) for q in range(backend.num_qubits)]
worst = max(errors, key=lambda x: x[1])
best = min(errors, key=lambda x: x[1])
print(f"Readout error — best qubit: {best[1]:.4f}, worst: {worst[1]:.4f}")
Run that on almost any current device and you'll see the spread between the best and worst qubit on the same chip is often a factor of five or more. That variation is invisible in every summary statistic, and it's precisely why transpilation and qubit selection matter so much. Our SDK comparison and hardware guide go deeper on picking backends.
The field hasn't picked a winning qubit technology
Something notable by its absence: the order never names a qubit modality. No preference for superconducting circuits, trapped ions, photonics, neutral atoms, or spin qubits. It refers generically to "quantum-enabling component technologies."
That's a meaningful signal. When a technology matures, procurement documents get specific. Nobody writes a national strategy for "computing devices based on some form of switching element." The deliberate vagueness reflects genuine reality: as of mid-2026, no modality has clearly won.
Each remains ahead on different axes — superconducting chips on gate speed and fabrication scale, trapped ions on fidelity and all-to-all connectivity, neutral atoms on qubit count, photonics on room-temperature operation and networking. The 2026 holographic codes experiment ran on trapped ions; Google's below-threshold error correction result ran on superconducting hardware. Both were landmark results on completely different platforms.
For anyone learning quantum computing, this is actually good news, and it has a concrete implication: don't over-specialize in one vendor's stack. The abstractions — circuits, gates, measurement, error mitigation — transfer across hardware. The vendor-specific API details may not.
"Beyond current classical capabilities" is the bar now
The order establishes the Quantum Computer for Application Development and Discovery Science (QC-ADDS) effort, aiming to deliver at least one machine to a Department of Energy facility and make it available to the scientific community. The Secretary of Energy has 90 days to identify specifications for systems capable of "transformative scientific applications... on a path towards economically significant applications and beyond current classical computer capabilities."
Two things are worth noticing in that phrasing.
First, the framing is discovery science, not commercial advantage. The goal is a scientific instrument — a machine researchers use to learn things — rather than a product that beats classical computers at a business task. That's consistent with where the evidence actually points: IBM's 2023 quantum utility result and the holographic-codes work were both physics experiments.
Second, "on a path towards" is doing real work in that sentence. It's an acknowledgment that the machine being specified is a step, not a destination.
It also explains why the bar keeps moving. Classical methods improve too: a 2026 result classically simulating Rydberg atom arrays reproduced experiments that had been treated as beyond classical reach, which is exactly the kind of moving target a national benchmarking center would exist to track. Claims of quantum advantage are only as durable as the best classical algorithm anyone has bothered to write.
Distributed quantum computing shows up as a serious goal
Tucked into the sections on networking is a request for DOE plans covering quantum networking for "distributed quantum computing."
This matters because it's an admission about scaling. Building one enormous quantum processor is extraordinarily hard — wiring, cooling, crosstalk, and yield all get worse as chips grow. Linking several smaller processors into one logical machine is an alternative route, and it depends on distributing entanglement between separate devices with high enough fidelity to be useful.
The underlying primitive is quantum teleportation, which moves a qubit's state between locations using shared entanglement plus two classical bits. It's been experimentally routine for years — but doing it fast enough, and cleanly enough, to knit processors together is not. Seeing it named as a five-year planning target is a sign the single-big-chip strategy isn't assumed to be sufficient.
The workforce gap is treated as a bottleneck
The order directs the NSF to initiate "a network of National QIST Workforce Development Institutes" within 180 days, alongside recruitment and retention strategies.
Workforce sections are easy to skim past, but the inclusion tells you something concrete: the constraint on quantum progress isn't purely hardware. There are not enough people who can write, debug, and reason about quantum programs — and that shortage is now considered severe enough to warrant dedicated institutions.
For anyone reading this site, that's the most directly actionable part of the whole document. The skills being described as scarce are learnable right now, for free, on real hardware. Our courses page collects the best structured options, and the getting started guide will get you running a circuit on a real QPU in an afternoon.
What it doesn't contain
Worth stating plainly for accuracy: the order specifies no dollar figures. It sets direction, assigns responsibilities, and imposes deadlines — 90, 120, 180, and 210 days across various sections — but appropriations come from elsewhere. Deadlines in an executive order are also not guarantees of delivery.
Treat the timelines as a statement of intent, not a schedule you can plan around.
The takeaway
Strip out the politics and EO 14413 reads as an unusually frank technical assessment:
- We cannot reliably measure quantum computer performance, and that's now urgent enough to institutionalize.
- No qubit technology has won, so bets are being hedged.
- The near-term target is scientific discovery, not commercial advantage.
- Scaling may require networking chips together rather than building one giant one.
- There aren't enough trained people, and that's a first-order bottleneck.
None of that is pessimistic. It's a field that has moved past "does this work at all" into "how do we measure it, scale it, and staff it" — which is roughly the transition classical computing went through in the 1950s.
The measurement problem is the one to watch. Benchmarking sounds dull next to qubit-count records, but you cannot engineer what you cannot measure, and every serious claim about quantum advantage ultimately rests on it. If you want to understand where the field is actually going, learn to read the error rates instead of the headlines — starting with our glossary of the terms that matter.