Four conclusions explain the month, and only one logical qubit has a measured error rate
Between Sep 24 and Oct 23, 2026, quantum computing produced precise error-correction numbers, a government verification programme and large private cheques. Four conclusions explain the month; each points to the module that holds the evidence, with status judged against Oct 24, 2026.
Four conclusions
Error correction improved on paper, with one logical qubit behind the headline
Quantinuum's 98-qubit Helios reached 5 × 10−6 logical error per round at code distance 7, resting on 3 failures in 29,000 runs of 20 rounds; it is a preprint, not peer reviewed. Infleqtion's 30 logical qubits on 80 physical come from a company release with no logical error rate, and Quantinuum says its figure and Google's 99.99% magic-state fidelity measure different things.
Speedups and cryptography estimates are projections
IonQ's “up to 74x” is 74.4x for logical Clifford gates on its larger code, evaluated for a design and not run on a machine (Oct 6). The material holds no new quantum advantage claim dated in the window, and Shor estimates (under 1 million noisy qubits for RSA-2048) predate it; 500,000 ÷ 98 ≈ 5,100 times the Helios qubit count.
Governments paid for verification and measured their own unreadiness
DARPA moved Atom Computing, Diraq, IBM and IonQ into Stage C on Oct 7, and the Department of War announced about $350M on Oct 8 (about $200M DARPA plus a conditional $150M PsiQuantum loan). GAO-27-108740 found none of 24 agencies fully prepared, with 89 recommendations; the OMB plan deadline of Oct 22 has passed, and filings are not confirmed.
Capital concentrated in a few names while revenue stayed small
Oratomic announced a $475M Series B ($775M since March), and IonQ priced a $2B equity offering and agreed to buy Elisity for up to $1.3B. Against that, Gartner estimates $869.9M of 2026 industry revenue, and IonQ, Rigetti and D-Wave reported Q2 revenue of $80.1M, $5.1M and $3.01M.
Milestones at a glance
| Company | Milestone | Metric | Date (2026) | Status (Oct 24) | Source |
|---|---|---|---|---|---|
| Quantinuum | Helios color-code memory (trapped ion) | 5 × 10−6 logical error per round, distance 7 | Oct 19 | Preprint, not peer reviewed | Quantinuum; PostQuantum |
| Infleqtion | Sqale logical qubits (neutral atom) | 30 logical on 80 physical qubits | Sep 24 | Company release; not independently verified | Infleqtion |
| Willow magic-state cultivation | 99.99% fidelity, about 40x fewer errors | Oct 15 | Science, per secondary report | Futurum | |
| IonQ | Scheduled qLDPC error correction | Up to 74.4x (Clifford), 5.0x (Toffoli) | Oct 6 | ProjectedCompany blog and preprint | IonQ |
| DARPA | Quantum Benchmarking Initiative Stage C | Four firms join Microsoft and PsiQuantum | Oct 7 | Announced | DARPA |
| GAO | GAO-27-108740 | 0 of 24 agencies fully prepared; 89 recommendations | Oct 6 | Published | GAO |
| Department of War | Quantum package | About $350M (about $200M + about $150M loan) | Oct 8 | AnnouncedLoan conditional | Department of War |
| Oratomic | Series B | $475M; $775M total since March | Oct 8 | Announced | Oratomic; Wilson Sonsini |
| IonQ | Equity offering | $2B | Oct 14 | Priced | 24/7 Wall St. |
Status labels follow the source material as of Oct 24, 2026. Counts are as reported: the Helios figure is a logical-error rate, the Infleqtion figure a logical-qubit count.
Hardware: the strongest evidence is one trapped-ion logical qubit, and three results measure three different things
Each hardware claim behind the month is checked for platform, metric and who has verified it, as of Oct 24, 2026.
Window milestones at a glance
| Item | Key figure | Date (2026) | Status and evidence | Source |
|---|---|---|---|---|
| Quantinuum Helios (trapped ion, 98 qubits) | 5 × 10−6 logical error per round | Oct 19 | Company releasePreprint, not peer reviewed | Quantinuum; PostQuantum |
| Infleqtion Sqale (neutral atom) | 30 logical qubits from 80 physical | Sep 24 | Company releaseNo preprint in the material | Infleqtion |
| Google Willow (superconducting, 105 qubits) | Magic state 99.99%, ~40x error cut | Oct 15 | Science, per secondary report | Futurum; Quantinuum |
| IonQ (trapped ion to memory) | >1,000 entanglements per second | Oct 9 | Company releasearXiv preprint | IonQ |
| DARPA QBI Stage C | 4 firms join Microsoft, PsiQuantum (6) | Oct 7 | Government release | DARPA |
By platform
On Oct 19, 2026 Quantinuum announced that its Helios trapped-ion computer reached 99.9995% logical fidelity with a color code. The blog defines the metric: a logical failure probability of 5 × 10−6 per syndrome-extraction cycle at code distance 7, falling by an average 12x per step across distances 3, 5 and 7. “Five errors per million operations” therefore means per million rounds of a memory experiment, not per gate.
Helios has 98 qubits and the experiment encoded one logical qubit. The 23-author preprint was not peer reviewed at announcement. Preprint coverage lists 13, 37 and 73 physical qubits (data plus measurement) at distances 3, 5 and 7, while the press release says “up to 37”; the material does not reconcile them, and we follow 73 for distance 7.
Sandia's Robin Blume-Kohout, not a study author, called it “a logical qubit nearly 100x better than its component physical qubits”: commentary, not independent verification. The “100x lower” claim targets Google's 2024 result. The per-round metric ignores round length: a Helios round takes about 200 ms against about 1 µs on superconducting chips, roughly 200,000x. The “30x fewer physical qubits” figure is an Apollo-scale projection, not a measurement.
On Oct 9 IonQ announced an entanglement rate above 1,000 per second between a trapped-ion qubit and a silicon-vacancy memory, over 4x the prior trapped-ion record from Duke's Chris Monroe group, so that record was under 250 per second (1,000 ÷ 4). The Qubit Report adds 87.9% fidelity, absent from IonQ's text in our material and a link measure, not a gate fidelity.
Infleqtion (NYSE: INFQ) announced that Sqale entangled 30 logical qubits on 80 physical, 10 blocks of 8 atoms, about 2.7 physical per logical (80 ÷ 30), in a circuit of about 1,000 physical operations. Evidence is a result roughly 1,000x above the uniform-random baseline; software loss correction raised retained samples 4x at a cost in accuracy.
No logical error rate, preprint or independent check appears in our material, so this is a company press release plus blog. “First neutral-atom company at 30 logical qubits on a commercial system” is the company's wording.
Quantum Computing Report adds four logical CCZ gates and an AI-found operation that cut a key entangling step from eight physical two-qubit gates to four. Earlier logical-qubit counts were 2 (2024) and 12 (2025), a 2.5x step (30 ÷ 12), all company-reported.
CEO Matt Kinsella cited 99.73% two-qubit gate fidelity in an Oct 21 interview that still calls 30 logical qubits a future goal, so the figure may predate the experiment. It is a gate fidelity, not a logical error rate, and cannot be set beside Helios.
According to Futurum, Google Quantum AI published magic-state cultivation on the 105-qubit Willow in Science on Oct 15, cutting state-preparation error about 40x. Quantinuum's David Hayes gives 99.99% fidelity for a resource consumed by logical operations, not a logical computation. We have no Google primary source.
Quantinuum says its figure measures a logical computation and Google's 99.99% measures the preparation of a resource state. Its “100x lower than Google” compares against Google's 2024 surface-code result, a different code and platform.
Superconducting chips check about once per microsecond (about 1 µs), against about 200 ms quoted for Helios: 200 ms ÷ 1 µs = 200,000. IBM's only window hardware item in our material is its Stage C selection.
DARPA announced that Atom Computing, Diraq, IBM and IonQ advanced to Stage C of its Quantum Benchmarking Initiative, joining Microsoft and PsiQuantum. The test is whether any approach reaches “utility scale”, value above cost, by 2033, moving “from plans to proof”. Quantinuum, QuEra, Photonic Inc. and Xanadu were not named; absence is not rejection.
These are company targets or prior results, not Stage C measurements, and $300M is a ceiling, not an award. Stage C is the first test in the month that does not rest on company-reported numbers.
| Performer | Platform | Funding | Hardware claim |
|---|---|---|---|
| Atom Computing | Neutral atom | Up to $300M; initial not disclosed | Magne, >1,200 qubits, early 2027 |
| Diraq | Silicon spin | $51M initial, up to $300M | 8-qubit CMOS array |
| IBM | Superconducting | Not disclosed | Starling 2029: 200 logical qubits |
| IonQ | Trapped ion | Up to $300M through 2029 | Superion 256-qubit chips |
Error correction and algorithms: only a memory experiment was demonstrated, and speedups and Shor estimates remain projections
This module separates what the window demonstrated from what is still projected, and keeps each yardstick apart.
Helios logical error per round by code distance (d)
Quantinuum preprint, Oct 19, 2026; not peer reviewed. d is the code distance. Points sit on a log scale: each gridline is 10x. Lower is better.
13, 37 and 73 physical qubits (data plus measurement) at distances 3, 5 and 7, from preprint coverage. The experiment encoded one logical qubit.
Four things to separate
5 × 10−6 logical error per round for a triangular color code at distance 7 (73 physical qubits including measurement qubits), Λ = 12. Decoding ran after the experiment with Google's Tesseract decoder, so it was not real-time.
99.99% fidelity measures the preparation of a resource state, not a logical computation, so it cannot be ranked against Helios. Figures reach us via Quantinuum and a secondary report; no Google primary source.
30 logical qubits (80 atoms, 8:3 ratio, codes [[8,3,3]] and [[8,3,2]]) ran an IQP circuit of about 1,000 operations with four logical CCZ gates and a signal about 1,000x above random noise. These small codes have no per-round logical error rate, so they cannot be ranked against Helios.
Helios, distance 7. 3 failures ÷ (29,000 runs × 20 rounds) = 3 ÷ 580,000 ≈ 5.2 × 10−6 per round. Logical two-qubit Clifford error: 7 × 10−4 at distance 3 and 7 × 10−5 at distance 5 against 9.5 × 10−4 unencoded, about 14x better (9.5 × 10−4 ÷ 7 × 10−5); distance 7 was not run.
Extrapolation (estimate). According to PostQuantum's analysis, holding Λ = 12 constant, 5 × 10−6 ÷ 127 ≈ 1.4 × 10−13 at distance 21, needing 661 physical qubits per logical qubit; Λ = 2 would need distance 53 and 4,213. Google's Craig Gidney estimated that practical error rates need distance about 21.
Cross-check (before the window). At IEEE Quantum Week (Toronto, week of Sep 14) Quantinuum's CEO reported a Helios memory failing 4.6 times per 100,000 rounds; 4.6 × 10−5 ÷ 5 × 10−6 ≈ 9.2, consistent with the claimed 10x, though that report does not name the code. Quantinuum's Sep 28 blog gives Helix a logical compute error of 2.8 × 10−4 against 8 × 10−4 physical and calls it 4.28x, but 8 ÷ 2.8 ≈ 2.9, so the baseline is unstated and we do not use 4.28x.
IonQ's blog presents work by Mark Webster and Nicolas Delfosse (arXiv 2607.16166; the material gives no posting date, so the window item is the announcement). Joint measurement of commuting logical operators, steered by classical scheduler codes, plus a logical CliNR protocol, is compared with sequential measurements. IonQ says “up to 3x” for joint measurement; Quantum Computing Report labels the gain 2.4x at L=10 and 3.0x at L=20 “vs. Viterbi”, and we follow IonQ's wording.
“Up to 74x” is the larger code on random Clifford circuits; the smaller code gives 74.4 ÷ 18.5 ≈ 4x less. The material does not say any of this ran on a machine; it is evaluated for IonQ's Walking Cat trapped-ion blueprint, so treat it as projected.
| Code | Logical Clifford | Logical Toffoli |
|---|---|---|
| Q70, 6 logical qubits | 18.5x | 4.2x |
| Q102, 22 logical qubits | 74.4x | 5.0x |
A speedup in logical operations only helps once logical qubits of this kind exist on hardware. Demonstrated in the window was one trapped-ion logical qubit held for 20 rounds at distance 7; the 74x speedup, the 90% qubit saving and every Shor estimate are projections.
The guide covers 46 talks, mostly simulation or replayed data. IQM's sum is 10,000 logical qubits × 50 checks per µs = 500 Gbit/s. Maryland timed a full superconducting control loop at 446 ns with no qubits attached. Tokyo's sandwich decoder cut time per round from 186.7 to 122.1 ns, about 35% less, in simulation. Learned decoders ranked best in simulation did worse on IBM chips.
Reported at about 20x lower logical error than belief propagation near 1% physical error, with microsecond decoding. No paper status is given and the result date is not stated.
PennyLane added ONNX-model decoders to Backline and ran NVIDIA's Ising pre-decoder on an AMD Instinct MI210 with PyMatching, online; no timings appear in the excerpt. Google's post launched a Kaggle competition on three benchmarks with its kickmix library, which fuzz-tests circuits at 60 billion gates per second on one CPU core.
A 104-qubit fault-tolerant simulation holds 245 ≈ 35.2 trillion amplitudes on 8,192 GH200 chips. Its “90% fewer physical qubits” is the protocol's promise, not a measured result.
No new advantage claim dated in the window appears in the material. On Jul 30 IBM and the University of Chicago announced a result they say meets the criteria; 16 days after the preprint, SUTD and NVIDIA computed exact probabilities for all 2,051 samples in 37.3 minutes on 256 GPUs (about 159 GPU-hours), and IBM has since revised the experiment. Google's 2024 random-circuit claim had no full classical reproduction as of Oct 5. At QCE26 (Sep 14, before the window) Google's Salvatore Mandrà said a classical run matched a 56-qubit Quantinuum H2 experiment in about two days on one A100.
Gidney's May 2025 paper puts RSA-2048 at under 1 million noisy qubits for under a week, against 20 million for 8 hours in 2019: 20 ÷ 1 = 20x fewer qubits. Google's March 2026 elliptic-curve paper uses fewer than 500,000 physical qubits, 18 or 23 minutes per key. Against the 98-qubit Helios, 500,000 ÷ 98 ≈ 5,100x. According to one explainer, IonQ's CEO claimed 2028 in September, which most researchers call optimistic.
National policy: the United States paid for verification while its agencies measured their own unreadiness, and the only fixed obligations are security deadlines
Governments spent the window preparing for the day the machines work and paying to find out whether they will. Ceilings, loans and announcements are kept apart.
Programmes and obligations at a glance
| Item | Key figure | Date | Status | Source |
|---|---|---|---|---|
| GAO-27-108740 | 0 of 24 agencies fully prepared; 89 recommendations | Oct 6, 2026 | Published | GAO |
| EO 14412 (OMB plans due Oct 22) | Key establishment by Dec 31, 2030 | Signed Jun 22, 2026 | In effect | Technology.org |
| Department of War package | ~$350M (~$200M DARPA + ~$150M loan) | Oct 8, 2026 | AnnouncedLoan conditional | Department of War |
| DOE Quantum Genesis | 8 priority problems; competition up to $215M | Oct 8; Sep 2026 | Announced | PostQuantum |
| UK / UKRI package | £78M | Date not stated | Announced | Quantum Computing Report |
| Horizon Europe QTI call | €20M | Closes Nov 17, 2026 | Open | EU Agenda |
The detail
GAO released a public version of a sensitive report issued in September 2025 and prepared with the Office of the National Cyber Director. It is a government audit, independent of the agencies it reviews. None of the 24 CFO Act agencies fully addressed three preparatory practices, because of gaps in cryptography expertise, in processes for inventories and funding, and in test plans. GAO made 89 recommendations to 23 agencies; 12 agreed, two partly agreed, seven neither agreed nor disagreed, and one disagreed with three of its four. It says most experts expect a cryptographically relevant quantum computer “possibly as soon as the 2030s”.
One agency fully addressed inventory, 22 partly and one not at all (1 + 22 + 1 = 24); 21 had partial cost estimates and three none; no agency had tested post-quantum cryptography; ONCD gave Congress roughly $7.1 billion. Seimless says GAO reissued the report on Oct 14; GAO's own page shows Oct 6, which we follow.
OMB M-26-15 (Jun 24) gave agencies 120 days, which falls on Oct 22. That date has passed; plans are not public and filings are not confirmed.
EO 14412 (signed Jun 22) requires quantum-safe key establishment by Dec 31, 2030 and signatures by Dec 31, 2031, for high-value assets only; the FAR Council contractor rule is due about Dec 19, 2026, and none had appeared by mid-October.
NSA CNSA 2.0 applies to new national-security purchases from Jan 1, 2027. EU roadmap: member states start by end-2026, high-risk uses by end-2030, and convert as much as feasible by 2035.
The US spent October measuring how unprepared its agencies are (none of 24 fully ready) and paying about $350 million to learn whether the machines will arrive.
DARPA lists up to $300 million per performer, and IonQ says its agreement runs to 2029, with later funding subject to appropriations. Four times $300 million would be $1.2 billion, against about $200 million announced. An even split would be about $50 million each (200 ÷ 4); a $51 million initial Diraq tranche is reported, but no source gives the split.
DARPA will host application workshops with DOE's Office of Science, NNSA and the Laboratory for Physical Sciences, beginning before January 2027. DOE named eight “Quantum Genesis Priority Applications” in chemistry, materials, subatomic physics and mathematics; its Q Competition, announced in September with up to $215 million planned, is separate from the $350 million. Both define questions, not results.
According to Future Markets, a market-research vendor, Commerce signed nine letters of intent for $2.013 billion of CHIPS incentives with minority equity stakes, and Anderon finalised a $1 billion foundry award. Dates are not stated; only two awards appear elsewhere, Diraq's $38 million letter of intent and Infleqtion's $100 million contingent, “not a final award”.
A £78 million package (£50 million sensing, £20 million networking supply chain, £8 million apprentices; 50 + 20 + 8 = 78), plus a £17 million EPSRC call on quantum-AI-HPC integration closing in December 2026, with awards up to £3.75 million, under the £2 billion ten-year strategy. Most of it funds sensing and networking, not computers.
A Horizon Europe call for a quantum testing infrastructure offers €20 million, about one grant of €19–20 million, closing Nov 17, 2026. A law firm reports the Commission is expected to present the EU Quantum Act on Nov 18; scope depends on the final text.
The Department of Industry awarded $12.3 million (currency not stated) in stage-two Critical Technologies grants to six projects.
Commercialisation: capital concentrated in a few names while revenue stayed small
Private and public capital arrived in the same four weeks; this module follows who raised what, who partnered with whom, and how the money compares with revenue.
Venture rounds named in the material, $ millions
Announced rounds, Sep 25 to Oct 22, 2026. Universal Quantum is a floor (more than $100M); Qilimanjaro converts €19.6M (about $21.3M); Qingxing converts about ¥100M. Our tally, not a market total.
One startup took about 76% of the disclosed venture money, and one listed company raised about 3.2 times the whole venture total from the market, against a Gartner estimate of $869.9M for the whole industry's 2026 revenue.
The window's deals at a glance
| Item | Key figure | Date (2026) | Status | Source |
|---|---|---|---|---|
| Oratomic Series B | $475M; total $775M since March | Oct 8 | Announced | Oratomic; Wilson Sonsini |
| Universal Quantum Series A | >$100M; DCVC, Firgun co-led | Oct 8 | Announced | TNW |
| Qilimanjaro Series A, first close | €19.6M | Oct 14 | Announced; round open | Qilimanjaro |
| Korea Quantum Computing SPAC (Charlton Aria) | $80M equity value | Oct 7 | Announced, not closed | TNW |
| IonQ to buy Elisity | Up to $1.3B, all stock | Oct 12 | Definitive agreement | Zacks (via Yahoo) |
| IonQ equity offering | $2B | Oct 14 | Priced | 24/7 Wall St. |
| Xanadu and GlobalFoundries | Multi-year; terms not stated | Oct 6 | Announced | Xanadu |
| Quantinuum and UWA | MoU; Helios cloud access | Oct 5 | Signed | Quantinuum (via StockTitan) |
| Fujitsu and New Mexico | Research centre; State pool $150M | Oct 9 | Exploratory | Fujitsu |
| Gartner revenue forecast | $869.9M (2026), $1.1B (2027) | Oct 8 | Estimate | TNW |
The detail
Oratomic announced that it had raised $775M in total since launching in March 2026. ARCH, Spark Capital, Khosla, Index, General Catalyst and Bezos Expeditions led; Lowercarbon, Bain Capital, Formation and Nebular are also listed. Its law firm states the Series B is $475M; the July Series A was $300M, and 300 + 475 = 775. The $5.4B valuation is reported by the Wall Street Journal, up from $1.5B (5.4 ÷ 1.5 = 3.6 times in three months); another outlet says the company did not disclose it, so we treat the valuation as reported. CEO Dolev Bluvstein told the Journal that about 10,000 qubits suffice, with a machine by 2030. That is a company claim, not verified.
Raised more than $100M, which it calls the largest UK quantum Series A. EDBI and Roblox's Dave Baszucki joined; valuation and revenue were not shared. The money funds a Singapore centre and expansion in the US, Japan and Hamburg, alongside a $77M contract for Germany's DLR (date not stated). No UK public support went into the round.
€19.6M: Indra €10M through IndraMind plus €9.6M from Spain's state entity SETT. It expects to complete the round in Q4 2026, so the final size is open. The company reports 80 staff and eight systems deployed, three at the Barcelona Supercomputing Center.
Will industrialise silicon-nitride photonics and single-photon detectors on a 300 mm line in Malta, New York. Terms are not stated.
A memorandum of understanding gives cloud access to Helios. No value is stated.
They will “explore” a research centre; the $150M belongs to the State, not Fujitsu.
Terra Quantum joined the QuEra Quantum Alliance on Oct 15. NTT DOCOMO BUSINESS signed a Classiq MoU (date not stated). Neither carries a value.
DLR's $77M contract with Universal Quantum (date not stated) is the only priced contract in the material.
We found no cloud or pharma deal with disclosed value in the window. D-Wave's renewals with AT&T, Nasdaq Verafin and Shionogi, among others, carry no value and were reported on Aug 6.
| Company | Q2 2026 revenue | Cash, Jun 30 | Source |
|---|---|---|---|
| IonQ | $80.1M | about $3B (about $2B after SkyWater) | Motley Fool |
| Rigetti | $5.1M | about $541M | Motley Fool |
| D-Wave | $3.01M | $546.2M | FinanceFeeds |
Grew 287% in Q2; part of its growth is acquired. H1 operating cash burn was $254.8M, against $85.6M a year earlier. Guidance conflicts: $280–290M excluding SkyWater, per three sources, versus $450–460M including it, per one; we follow $280–290M. The $2B was sold below the market price to one buyer, on top of 405.1M shares outstanding. Against the $13.92B market cap on Oct 16, $2B is about 14%, and Elisity's $1.3B about 9%.
D-Wave's $3.01M is flat against $3.10M a year earlier; Q2 bookings were $2.1M after $33.4M in Q1 (33.4 + 2.1 = 35.5); remaining obligations fell from $42.4M to $40.7M. Rigetti's H1 revenue of $9.5M already exceeds 2025's $7.1M. IBM does not report quantum revenue; it cites over $1B of cumulative bookings, and its CEO said profit impact comes in 2028–29.
Priced its IPO at $60 on Jun 4, raising $1.68B at about $15.6B (reported). It closed at $36.50 on Oct 23, market cap $9.697B: 36.5 ÷ 60 − 1 ≈ −39%. Morgan Stanley upgraded it to buy on Oct 15 (analyst view). Next report due Nov 5. Value over annualised latest-quarter revenue was about 52x for IonQ, 422x Quantinuum, 270x Rigetti and 540x D-Wave (New Market Pitch, Sep 24 update, estimate).
How far from practical use: one logical qubit with a measured error rate stands against a useful machine of 10,000
This module turns the month’s facts into a distance to practical use, using only those facts. Logical and physical counts are kept apart.
Logical qubits: shown against a useful-machine example
Linear scale: the Helios and Infleqtion bars are barely visible beside 10,000, which is the gap. Counts are logical qubits only.
For a Helios-type code, PostQuantum estimates 661 physical qubits per logical at distance 21 if the suppression factor stays at 12, and 4,213 if it falls to 2. Shor estimates sit lower: under 1 million noisy qubits for RSA-2048 and under 500,000 for elliptic curves, about 5,100 times the 98-qubit Helios (500,000 ÷ 98).
Speed
A Helios round takes about 200 ms against about 1 µs on superconducting chips, a factor of 200,000 (Quantinuum preprint coverage and QCE26 guide).
The month’s best number is a memory experiment, not a computation.
Decoding
Helios decoding ran after the experiment, not in real time. The 446 ns superconducting control loop was timed with no qubits attached (QCE26 guide, Oct 20).
Real-time decoding on a running logical qubit is not shown in the material.
Roadmaps are claims
IBM Starling 2029 with 200 logical qubits; Atom Computing Magne in early 2027 with more than 1,200 qubits; Oratomic’s roughly 10,000-qubit machine by 2030. All are company targets, not measurements.
IonQ’s 74.4x speedup is projected; DARPA’s Stage C tests utility scale by 2033.
What changed and what it means, by reader
Researchers
What changed. A defined per-round metric at distance 7 now exists for trapped ions; IonQ’s speedup is projected and Harvard’s decoder result has no stated paper status.
Compare only like metrics (per round, per gate, magic-state fidelity); look for peer review, real-time decoding and error rates for logical-qubit counts.
Security teams
What changed. GAO found 0 of 24 agencies fully prepared; the OMB plan deadline of Oct 22 passed with filings unconfirmed; the cryptographically relevant horizon is “possibly as soon as the 2030s”.
Treat the 2030 and 2031 EO 14412 dates and Jan 1, 2027 CNSA 2.0 as the clock, not machine milestones.
Investors
What changed. Oratomic $775M since March, IonQ $2B offering, Elisity up to $1.3B; Q2 revenue $80.1M, $5.1M and $3.01M; Quantinuum at $36.50 on Oct 23 against its $60 IPO (−39%).
Capital far exceeds revenue (Gartner estimate $869.9M for 2026); watch Quantinuum’s Nov 5 report and guidance conflicts such as IonQ’s. This is not investment advice.
Programme managers
What changed. US money went to verification and manufacturing; allies funded sensing, networking and testing.
Ceilings of up to $300M per performer are not awards; check what each programme measures.
Watch the dated obligations first: the security deadlines arrive before the machines do
Dates already past on Oct 24, 2026 are marked in effect; later ones are upcoming. Years after 2026 are company or government targets, not measurements.
Key dates
| Date | Event | State on Oct 24, 2026 | Source |
|---|---|---|---|
| Sep 24, 2026 | Infleqtion announces 30 logical qubits on Sqale | In effect | Infleqtion |
| Oct 6, 2026 | GAO-27-108740 published | In effect | GAO |
| Oct 7, 2026 | DARPA names Stage C performers | In effect | DARPA |
| Oct 8, 2026 | Department of War announces about $350M | In effect | Department of War |
| Oct 19, 2026 | Quantinuum announces the Helios color-code result | In effect | Quantinuum |
| Oct 22, 2026 | OMB M-26-15 migration plans due (filings not confirmed) | In effect | Technology.org |
| Nov 5, 2026 | Quantinuum’s next report due | Upcoming | Yahoo Finance |
| Nov 17, 2026 | Horizon Europe quantum testing call closes | Upcoming | EU Agenda |
| Nov 18, 2026 | EU Quantum Act proposal expected (reported) | Upcoming | Abreu Advogados |
| Dec 2026 | EPSRC quantum-AI-HPC call closes | Upcoming | Quantum Computing Report |
| About Dec 19, 2026 | FAR Council contractor rule due | Upcoming | Seimless |
| Jan 1, 2027 | NSA CNSA 2.0 applies to new national-security-system purchases | Upcoming | Technology.org |
| Early 2027 | Atom Computing’s Magne, over 1,200 qubits (company target) | Upcoming | Russ Fein |
| 2029 | IBM Starling, 200 logical qubits (company target) | Upcoming | PostQuantum |
| Dec 31, 2030 | EO 14412 quantum-safe key establishment | Upcoming | Technology.org |
| 2033 | DARPA’s utility-scale test horizon | Upcoming | DARPA |