Quantum Computing Roadmaps Converge on 2029 as Google Cuts Errors

The week the 2029 timelines got more believable

Google's quantum team published a result in Science on October 15 showing that a technique called magic-state cultivation reduced errors about 40-fold on its Willow processor. The prepared T-states reached a reported logical fidelity of 99.99%.12 Most coverage also passed along Google's caveat: the method has to work at larger scale before it saves hardware in practice.1214

The timing makes the result more significant than it would otherwise be. IBM, Google, Quantinuum and Microsoft now all name 2029 as roughly the year a useful error-corrected machine should exist, and none of them has one today.5 Meanwhile, governments have attached binding dates to the move away from encryption that quantum computers could break.21 Hardware schedules and cryptographic deadlines are now converging on the same few years, and this result shows the bottleneck that sits between them.

What Google actually showed

Magic states are the resource that lets an error-corrected quantum computer run every operation it needs, not just a limited set. Each one can be used only once, so a machine has to keep producing them.20 Some resource estimates suggest that making these states could take up as much as 95% of a machine's capacity. That figure is a projection, not something measured on Willow.13

The cultivation method was proposed in 2024 by Craig Gidney, Noah Shutty and Cody Jones. A noisy state goes into a small region of the chip, coordinated checks throw out flawed attempts, and the surviving state is grown into a larger encoding.20 The team also moved cultivated states from a distance-3 color code to a distance-5 surface code. That step, called grafting, is evidence the technique can fit into the surface-code designs most of the field relies on.14

The careful reports differed from the eager ones mainly on the fine print. The Quantum Insider pointed out that the 40-fold figure applies only to this state-preparation task and is not a 40-fold improvement in the processor overall. It also reported that a storage test produced errors about seven times higher than simulations predicted, most likely because of leakage, which the simulations left out.13 Other reporting said only about 8% of attempts passed every check and that the experiment used 54 of Willow's 105 qubits.15 To get the protocol working, the team used reinforcement learning as a final calibration step, which improved both error rates and the share of attempts kept by about an order of magnitude.13

This is a demonstration of a mechanism, not a working capability. The cryptographic coverage also stretched the result further than it goes. One outlet said it brings threats to RSA-2048 and Bitcoin's elliptic-curve cryptography "closer to reality."19 By that outlet's own account, though, breaking RSA-2048 would require magic-state errors of about one in ten billion, roughly a million times cleaner than what Willow achieved.1915 The better reading is that cultivation was one of the main assumptions behind Gidney's May 2025 estimate that RSA-2048 could be factored with fewer than a million noisy qubits, and that assumption has now been checked on real hardware for the first time.15

IBM: on schedule so far, with one milestone still outstanding

IBM has the most detailed public plan of any company. Starling, being built in Poughkeepsie, New York, is meant to run 100 million gates on 200 logical qubits in 2029. Blue Jay, with 2,000 logical qubits and a billion gates, follows from 2033.1 IBM uses qLDPC codes rather than surface codes and says they cut error-correction overhead by about 90%.3 The intermediate chips are Loon in 2025, Kookaburra in 2026 and Cockatoo in 2027. Kookaburra is meant to be the first module that stores encoded information in a qLDPC memory and processes it with an attached logical processing unit.2

IBM has mostly delivered on time. It shipped the 120-qubit Nighthawk and the Loon test chip in November 2025 and moved its main wafer production to a 300 mm facility in Albany, New York.5 One industry tracker reports that an August 2026 revision of Nighthawk ran circuits of more than 7,500 gates accurately, which meets IBM's gate-count target for 2026.32 The quantum advantage claim is less settled. In July 2026, IBM and partners said they had achieved advantage on several problems, and one of those results was later reproduced on classical computers.9

The milestone to watch is Kookaburra. As of October 10, no primary source had reported that it had been delivered.9 IBM usually unveils processors at its November developer conference, so the gap is not yet a delay. If November passes without it, though, that would be the first real slip in the fault-tolerance plan.5 Even what Kookaburra is supposed to be depends on the source. One company profile still describes it as a 1,386-qubit chip that links into a 4,158-qubit system.6 IBM's own current material describes it as a qLDPC memory-plus-logic module.2 That gap suggests IBM's priority has moved from raw qubit counts to encoded operations.

The rivals: different bets, all roughly 2029–2030

Assessments of who is leading change depending on what gets measured. One comparison puts IBM one to three years ahead on industrialization: deployed systems, software reach and manufacturing pace. The same analysis gives Google the cleaner superconducting error-correction milestone and Quantinuum the lead in demonstrated logical qubits.10

In 2024, Quantinuum and Microsoft reported 12 logical qubits on Quantinuum's H2 machine.38 Helios, delivered in November 2025, is reported to have shipped 50 error-corrected logical qubits at 2:1 encoding.32 The company's date for full fault tolerance has moved over time. It announced a target of 2030 with its Apollo system in 202438, while more recent analysis groups Quantinuum with the other companies aiming for 2029.5

IonQ's target is far larger than anyone else's: 2 million physical qubits and 80,000 logical qubits by 2030.31 It does have a record to point to. In October 2025 it reported two-qubit gate fidelity of 99.99%.35 Even sympathetic analysis calls the timeline extremely ambitious and notes that any slip in intermediate chips would push back the 2028 and 2030 goals.39 IonQ is aiming for an order of magnitude more logical qubits than IBM in about the same time, so its plan should be read as an upper bound, not as what the industry expects.

Microsoft remains the hardest to judge. Its Majorana 2, announced in June 2026, is reported to improve qubit reliability 1,000-fold, but the device tested was a four-qubit prototype. Microsoft still targets a scalable fault-tolerant machine by 2029.32 Many physicists still dispute its topological approach.5 Elsewhere, Infleqtion demonstrated 30 entangled logical qubits on neutral atoms in September 2026, and Google itself expanded into neutral-atom hardware in March.32 Amazon's expanded partnership with QuEra aims to put a neutral-atom system with hundreds of logical qubits on its Braket cloud service by 2028.32

Cryptography no longer depends on the hardware race

Post-quantum migration has changed the most. On June 22, 2026, Executive Order 14412 required federal high-value and high-impact systems to use post-quantum key establishment by December 31, 2030, and post-quantum signatures by the end of 2031.21 Under the follow-up OMB memorandum M-26-15, every civilian agency must submit a migration plan by about October 22 this year.22 A proposed rule requiring federal contractors to meet NIST standards, including the post-quantum ones, by the end of 2030 is due around December 19.22 Under the NSA's CNSA 2.0 rules, new national security system purchases must comply starting January 1, 2027.22

The algorithms themselves are settled. NIST finalized ML-KEM, ML-DSA and SLH-DSA in August 2024.28 NIST's transition guidance, still in draft, deprecates RSA and elliptic-curve cryptography by 2030 and disallows them by 2035.21 The EU expects high-risk systems to migrate by the end of 2030.21 China stands apart: it is developing its own algorithms and has published no migration deadline.21 Cloudflare reads the 2031 signature deadline as a sign that the US government sees a real chance of a working code-breaking machine around then.28

IBM's 2029 machine would not break RSA-2048 on its own, since that is generally thought to need thousands of logical qubits.6 Migration deadlines are not set by when a machine arrives, though. Encrypted traffic recorded today can be decrypted later, which puts any long-lived data at risk now.29 For a large enterprise, just finding where cryptography is used can take 12 to 24 months.27

The bottom line

Across these reports, the hardware is ahead of where skeptics expected and behind where the press releases suggest. Google's cultivation result and IBM's gate-count progress show the engineering steps are being met. Kookaburra's status and Microsoft's four-qubit prototype show how much is still unproven. The 2029 dates are credible enough that waiting for proof is no longer a reasonable plan for anyone holding data that must stay secret into the 2030s. The policy deadlines now make that same argument with legal force.