Quantum Computing Watch · October 8, 2026
IBM, Google, and Microsoft are all chasing fault-tolerant quantum computing, but only one has a dated target, an independent test, and funded manufacturing. We sort what is demonstrated from what is promised.
Core Research Question
Who has the most credible path from today's hardware to useful fault-tolerant quantum computing?
1. Why Logical Qubits, Not Qubit Counts, Are the Scorecard
Physical qubits are error-prone. Quantum error correction groups many of them into a smaller number of logical qubits whose error rates are far lower, and a fault-tolerant machine is one that can run long computations on logical qubits without errors accumulating. That is why headline qubit counts say little: what matters is how many logical qubits a system has, how reliable their operations are, and how many physical qubits each one costs.
This first Quantum Computing Watch applies that lens to the three most closely watched programs. IBM, Google and Microsoft take different architectural bets and make different kinds of claims, so we sort each into what is demonstrated, what is targeted and what remains unproven.
The CODEW angle: In 2026, the industry is increasingly judged on engineering milestones, error correction and manufacturing, not raw qubit counts. A dated target is a promise; a measured logical result is evidence.
2. IBM: The Most Specific Roadmap, and the First Independent Test
IBM has the most explicit schedule. Its plan is Quantum Starling in 2029, which it says will run 100 million quantum operations on 200 logical qubits, about 20,000 times more than today's Heron-generation systems, followed by Blue Jay in 2033 with more than 2,000 logical qubits and a billion operations. The intermediate steps are named: Loon (2025) to test qLDPC code components, Kookaburra (2026) as the first modular processor storing and processing encoded information, and Cockatoo (2027) linking modules. IBM also published a real-time decoder design, Relay-BP, aimed at one of the field's main worries: whether qLDPC codes can be decoded fast enough.
On October 7, IBM said it was selected for Stage C of DARPA's Quantum Benchmarking Initiative, where independent experts test hardware to judge whether a fault-tolerant machine can be built. DARPA's goal is a system by 2033 whose computational value exceeds its cost. IBM says its progress is on course, including demonstrations of core fault-tolerance hardware components and decoding advances. In June, it committed more than $10 billion over five years to R&D, manufacturing and M&A, and a $1 billion contribution to Anderon, a planned quantum wafer foundry.
The caveat is that IBM's own materials are the evidence for most delivery claims. We did not find independent confirmation that Kookaburra has shipped, and IBM says only that partners will demonstrate quantum advantage in 2026, which remains a claim to test.
The CODEW angle: IBM is the only one of the three that pairs a dated logical-qubit target with a funded manufacturing plan and an independent verification process. It also has the most to lose if intermediate dates slip.
3. Google: The Strongest Error-Correction Evidence, Without a Date
Google frames its plan as six milestones rather than a calendar. It reached beyond-classical computation in 2019, and a logical-qubit prototype in 2023, and its December 2024 Willow chip showed below-threshold error correction: growing the surface code from distance 3 to 5 to 7 reduced logical errors each time. The next milestone is a long-lived logical qubit, which Google defines as one that can perform a million computational steps with less than one error. After that come a logical gate, 100 logical qubits tiled together, and finally a machine of 1 million physical qubits.
The limits are as important as the result. Willow demonstrated a preserved logical qubit, not logical gates, and logical error rates of roughly one in a thousand per cycle remain far above the roughly one-in-a-million levels Google itself describes as the target for a true fault-tolerant qubit. As far as we could find, Google's public roadmap does not attach years to its remaining milestones.
The CODEW angle: Google has the cleanest physics evidence that error correction scales on superconducting hardware, but an undated roadmap is harder to hold to account than IBM's. Its next milestone, the long-lived logical qubit, is the one to watch.
4. Microsoft: Two Tracks, One Disputed
Microsoft runs a high-risk topological hardware program alongside lower-risk partnerships. The topological track was announced as Majorana 1 in February 2025 and immediately contested. On June 2 at Build, Microsoft unveiled Majorana 2, swapping aluminum for lead and reporting parity lifetimes of about 20 seconds, up from milliseconds. Critics note that the new paper presents only one of the two measurement types needed to show a qubit, and Nature published a formal critique of the 2025 paper in June, according to reports, with Microsoft's reply. No independent replication has been reported.
The second track is about logical qubits on other people's hardware. Microsoft's qubit-virtualization software produced 12 logical qubits with Quantinuum and 24 entangled logical qubits on Atom Computing's neutral atoms in 2024. Its Magne machine, a "Level 2 Resilient" system pairing more than 1,200 Atom Computing physical qubits with Microsoft's software, is being built for QuNorth in Copenhagen and is expected to come online in early 2027. Reports say Microsoft targets a scalable machine by 2029, though the date for the topological path is disputed.
The CODEW angle: Microsoft offers the widest range of outcomes: a possible architectural leap if topological qubits are validated, and a pragmatic neutral-atom route that does not depend on them.
5. The Fault-Tolerance Scorecard
What each company has shown, based on public statements and published reporting as of October 8, 2026:
| Company | Architecture | Target | Demonstrated | Not yet shown |
| IBM | Superconducting, qLDPC codes, modular | Starling 2029 (200 logical qubits, 100M operations); Blue Jay 2033 | Core fault-tolerance components and decoding progress (company-reported); DARPA Stage C entry | Starling-scale logical operations; Kookaburra delivery not independently confirmed |
| Superconducting, surface code | Six milestones; 1M physical qubits at the end; no dates found | Below-threshold error correction on Willow (2024); logical-qubit prototype (2023) | Long-lived logical qubit; logical gates; any logical-qubit error near 10^-6 | |
| Microsoft | Topological (Majorana) plus neutral-atom and ion partnerships | Reportedly scalable machine by 2029; Magne early 2027 | 12 logical qubits (Quantinuum) and 24 entangled logical qubits (Atom), 2024; Majorana 2 chip (June 2026) | Independently validated topological qubit; Magne operation |
The CODEW angle: Treat the right-hand column as the real roadmap. Dated targets are marketing until a logical gate or long-lived logical qubit is measured.
6. Which Milestones Matter Most
Four milestones separate the three programs from a useful machine. First, logical gate fidelity: Google's roadmap treats logical gates as a separate milestone, and none of the three has published logical-gate performance near the million-operation level. Second, decoding in real time, where IBM has published a design but a working system is the test. Third, overhead: the physical qubits needed per logical qubit decides cost, and qLDPC codes, as IBM uses, aim to cut it. Fourth, independent verification, which is what DARPA's Stage C is for.
Quantum advantage claims should also be read carefully. IBM expects partners to demonstrate it in 2026 and Google reported a verifiable advantage result last year, but neither is the same as running a useful computation on logical qubits.
The CODEW angle: Watch for logical-gate results and DARPA's verification outcomes, since they are harder to market around than qubit counts.
The CODEW Angle
Who has the most credible path to useful fault-tolerant quantum computing? On today's evidence, IBM has the most auditable path, Google the strongest physics, and Microsoft the highest variance.
That is our assessment, not a prediction. IBM's combination of a dated target, a decoder design, manufacturing investment, and a place in DARPA's verification program makes its claims the easiest to check. Google has the most convincing demonstration that adding qubits reduces errors, but has not committed to dates. Microsoft's partner-based logical-qubit work is real, while its topological bet remains contested.
The Quantum Lens for this series runs from hardware to error correction, software, infrastructure, applications, commercialization and capital. The next editions will test it against DARPA's benchmark, quantum chip manufacturing and the software stack.
Sources
→ IBM: Advances to Stage C of DARPA Quantum Benchmarking Initiative
→ IBM: Commits more than $10 billion to quantum computing
→ The Quantum Insider: IBM's modular, scalable full-stack quantum roadmap
→ ITWeb: IBM firms up quantum roadmap with 2029 deadline
→ Google Quantum AI: Our quantum computing roadmap
→ Scott Aaronson (Shtetl-Optimized, via Harvard TagTeam): The Google Willow thing
→ Microsoft Quantum: From research to product engineering (Magne)
→ Science News: Microsoft's quantum chip got an upgrade. Critics are still skeptical
→ Science: Debate erupts around Microsoft's quantum computing claims
→ WindowsForum: Majorana 1 dispute and peer-review critique (secondary)
→ Andes Qubit: Quantum hardware roadmaps 2026 to 2030 (secondary)
The CODEW Stat
200 logical qubits · 100M operations · Starling 2029 · $10B+ IBM commitment · 0 independent topological-qubit confirmations
IBM figures are company targets; the Microsoft line reflects reporting that no independent replication has been published.
Editorial Note
Quantum Computing Watch tracks the transition of quantum computing from research toward fault-tolerant systems, commercial workloads, infrastructure and deployment, through the lens of technology, companies, capital, infrastructure and commercialization. It does not claim quantum advantage without credible evidence.
This edition draws on company announcements, roadmaps, scientific reporting and secondary analysis as of October 8, 2026.
Educational content only. Not investment advice. Roadmap dates are company targets and may change. Several delivery claims are company-reported and not independently verified by The CODEW. The Majorana dispute reflects published criticism and company responses, and some details come from secondary reporting.
Reviewed by Erwin Castro
on
Thursday, October 08, 2026
Rating:
No comments: