How late are quantum roadmaps running?

Every quantum vendor publishes a roadmap. Very few publish what happened to the previous one. Rosetta Quantum went back to the dated announcements, paired each one with the vendor's latest datable public statement about the same milestone, and measured the gap in months. This page is the result: a drift table where every cell carries its source.

What did we measure, and how?

Nine commitment pairs across seven organizations — IBM, Quantinuum, Rigetti, PsiQuantum, IonQ, Nord Quantique and Google. A pair enters the table only when both endpoints are datable documents: a roadmap announcement, a press release, an SEC-grade filing, or dated third-party coverage of one. Drift is the distance, in months, between the milestone date promised then and the date attached to the same milestone now (or the delivery date, when it shipped). Every document in the table was re-read by Rosetta on 2026-10-29; nothing is quoted from memory.

Three method rules. First, we did not use web-archive snapshots: dated press releases and dated coverage serve as the archive, so every pairing can be re-run from the links in the table. Second, a milestone that today carries no date cannot be scored — we report the disappearance of a date, but we will not assign it a number. Third, delivery against a vendor-defined metric is scored as "on time on its own metric", which is not the same thing as an adjudicated milestone — see the IonQ row.

Which roadmaps slipped?

Four of the nine pairs moved later. The largest measured slip belongs to Rigetti: in October 2021 it announced a 1,000-qubit system expected by 2024 and a 4,000-qubit system in 2026 (SiliconANGLE, 2021-10-06). In May 2022 it pushed the 1,000-qubit machine to late 2025 and the 4,000-qubit machine to 2027 or after, citing "higher-than-anticipated labor, equipment, and system component expenses" (The Quantum Insider, 2022-05-18). As of January 2026, coverage of its chiplet strategy has the 1,000-plus-qubit system targeted "by 2027", current systems at 36-qubit chiplets with roughly 99.5% two-qubit gate fidelity, and expansion "toward 150 qubits in 2026" (Zacks via Nasdaq, 2026-01-20). That is +36 months on the 1,000-qubit date — 2024 to 2027 — and the 4,000-qubit machine no longer appears with a date at all.

PsiQuantum committed, at the April 2024 announcement of its AU$940M Australian federal and Queensland funding, to an "aggressive plan to have the site operational by the end of 2027" in Brisbane. In December 2025, reporting put the build roughly 12 months behind schedule and the utility-scale timeline "closer to 2030" (Startup Daily, 2025-12-08). That is approximately +24 months, with one caveat the table carries explicitly: the revision is press-reported, and we found no vendor filing that states a new date.

Nord Quantique's drift is internal to its own materials: its public roadmap aims at 100 logical qubits by 2028, while its own press language offers a "clear path" to a hundred-plus logical qubit machine by 2029 — both statements documented, and flagged as unreconciled, in the same vendor profile (PostQuantum, published 2025-05-22, modified 2026-05-27). That is +12 months between two statements from the same company.

Which dates held — and which one moved earlier?

The slips are not the whole story, and a drift table that hid the on-time rows would be a weak baseline. IBM's 2020 scaling ladder — Eagle 127 qubits in 2021, Osprey 433 in 2022, Condor 1,000-plus in 2023 (HPCwire, 2020-09-15) — was delivered year for year, closing with the 1,121-qubit Condor announced on 2023-12-04 (The Quantum Insider). The first rungs of its June 2025 fault-tolerance roadmap (HPCwire, 2025-06-10) also landed on schedule: the Loon qLDPC test chip and the 120-qubit Nighthawk, both promised for 2025, were announced on 2025-11-12 (IBM newsroom). The flagship dates — Starling, fault-tolerant, 200 logical qubits, 2029; Blue Jay, roughly 2033 — were set in June 2025 and will not be testable until they arrive.

One date in the sample moved earlier. Quantinuum's September 2024 roadmap promised universal, fully fault-tolerant quantum computing "by 2030" (Quantinuum press release, 2024-09-10). At the Helios launch in November 2025, the revised roadmap put Sol in 2027 and Apollo — the fault-tolerant machine — in 2029 (Computer Weekly, nov-2025), and an October 19, 2026 release reports 99.9995% logical fidelity on Helios — about five errors per million operations, in an experimental setting — and reaffirms Apollo as "expected to be available by 2029" (Quantinuum press releases, 2026-10-19). That is −12 months, with Helios itself delivered in November 2025. Whether the pulled-in date holds will only be measurable in 2029.

IonQ's December 2020 roadmap stated: "by 2025, we expect to achieve broad quantum advantage" (IonQ blog, 2020-12-09). In September 2025 its fifth-generation Tempo system was reported at AQ 64 (Market Chameleon, 2025-09-25; roadmap context in Quantum Computing Report, 2025-09-12). On its own schedule and its own metric, that is on time. The asterisk matters: AQ is a metric IonQ defines and reports itself, and no neutral adjudication of "broad quantum advantage" exists — the same measurement problem Rosetta documented in Quantum Volume vs. Algorithmic Qubits and in the utility-vs-advantage dictionary entry.

What does the full pairing show?

Why will roadmap drift matter?

Because the industry's flagship dates are converging on a single year. IBM's Starling (fault-tolerant, 200 logical qubits), Quantinuum's Apollo, Google's error-corrected machine and Nord Quantique's hundred-logical-qubit press framing all now point at 2029. None of these dates will be testable before they arrive; what this table will do is make the arrival measurable — the commitments are archived here, with sources, at their 2026-10-29 state. This measurement will feed the scheduled ≈April 2027 revision of Rosetta's pre-registered Q-Day bet, where roadmap credibility is an explicit input.

Drift of roadmaps also has a sibling: drift of claims — a vendor's own description of a past result narrowing over time, as documented for D-Wave's annealing supremacy claim and for the 2026 IBM advantage episode in the utility-vs-advantage entry. And it is one reason external referees exist at all: DARPA's Quantum Benchmarking Initiative is, in substance, a third party scoring vendor roadmaps that nobody else adjudicates.

What we don't know

• This is a sample, not a census. Pairs enter only where both endpoints are datable; vendors that publish undated roadmaps cannot be scored, so an industry-wide drift rate is very likely understated here — undated promises are the ones that slip silently.
• PsiQuantum's "closer to 2030" is journalism (Startup Daily, dec-2025), not a vendor filing. We found no PsiQuantum document stating a revised date.
• Google's unchanged 2029 means our search on 2026-10-29 found no public revision. Absence of a revision in public is not evidence the internal schedule holds.
• "On time on its own metric" (IonQ's AQ 64) is not an adjudicated milestone. No neutral party has scored "broad quantum advantage".
• Drift between public statements says nothing about internal engineering reality, in either direction. A vendor could also dispute a pairing; every source is linked so the pairing can be challenged and re-run.
• Whether the 2029 cluster will hold. That will only be measurable in 2029 — and this page will be re-measured before then.

Rosetta Quantum publishes verdicts with raw, reproducible data. This is educational content, not a product claim.