Quantum computing roadmap 2026 to 2030: IBM, Google, rivals

If you run technology or strategy at a large company, the short version is this: IBM, Google, Quantinuum and Microsoft all now point to 2029 as the year a useful, error-corrected quantum computer arrives. None of them has one today. Their roadmaps are worth reading because they tell you which technical milestones to watch between now and then, and those milestones are what should move your budget, not the press releases.

This quantum computing roadmap summary reflects what was public on 23 September 2026. We will update it as milestones land or slip.

Where does each company stand in September 2026?

Each of the four has shipped something real, but they are at different distances from fault tolerance, and one of them is still arguing about whether its basic building block works.

The table hides one important difference. IBM, Google and Quantinuum are scaling qubits that everyone agrees exist and work. Microsoft is betting on a new kind of qubit whose existence is still being debated in the physics literature. Microsoft may still succeed, but its roadmap carries a different kind of risk.

IBM’s roadmap: Nighthawk, Kookaburra and Starling

IBM has the most detailed public plan, and so far it has delivered close to schedule. In June 2025 the company laid out a year-by-year path to Starling, a fault-tolerant machine to be built in Poughkeepsie, New York, that should run 100 million quantum gates on 200 logical qubits by 2029.

The path goes through four chips. Loon (2025) tests the long-range connections that IBM’s error-correcting codes need. Kookaburra (2026) is supposed to be the first module that stores information in a qLDPC memory and processes it with an attached logical unit. Cockatoo (2027) links two such modules. Starling puts it together.

IBM’s bet is on qLDPC codes (the family it calls bivariate bicycle codes) instead of the surface code most of the field uses. The pitch is efficiency: IBM says the approach needs far fewer physical qubits per logical qubit. The cost is harder wiring, because qLDPC codes need connections between qubits that are not neighbors on the chip.

On 12 November 2025 IBM delivered Nighthawk, a 120-qubit processor with 218 tunable couplers, and the Loon test chip. It also said it had moved its main quantum wafer production to a 300 mm facility in Albany, New York, and that it expects the first verified cases of quantum advantage by the end of 2026. To keep those claims honest, IBM set up an open “quantum advantage tracker” with Algorithmiq, the Flatiron Institute and BlueQubit, where candidate results can be checked against the best classical methods.

As of this writing, IBM has not announced Kookaburra’s delivery. That is not yet a slip: IBM has usually shown new processors at its November developer conference. If November passes without it, treat that as the first real delay in the fault-tolerance plan.

Google’s roadmap: Willow and a second bet on neutral atoms

Google published the result that convinced many skeptics that error correction can work at scale. Its 105-qubit Willow chip, described in Nature in December 2024, showed that logical errors fell as the error-correcting code grew from a 3x3 to a 7x7 grid of qubits. That is the “below threshold” behavior every roadmap depends on.

In October 2025 Google followed with Quantum Echoes, an algorithm that ran about 13,000 times faster on Willow than the best classical estimate for that specific task, and whose results can be checked on other hardware. It is a physics result, not a business application, but it is the most credible advantage claim so far.

Google publishes fewer dates than IBM. Its roadmap has six milestones. The third is a long-lived logical qubit, one that fails around once in a million operations. We found no announcement that Google has declared that milestone complete.

The bigger news came on 24 March 2026. Hartmut Neven, who founded Google Quantum AI, announced a neutral-atom hardware team in Boulder, Colorado, led by the atomic physicist Adam Kaufman. His explanation was unusually plain: superconducting chips are easier to scale in circuit depth, neutral atoms are easier to scale in qubit count (arrays of around ten thousand atoms), and Google wants both. The same post says Google expects commercially relevant superconducting machines by the end of this decade.

For an executive, the useful reading is that Google itself is hedging on which hardware will win. You should too.

Quantinuum’s roadmap: Helios, Sol and Apollo

Quantinuum reports some of the highest gate fidelities in the industry and is now a listed company (Nasdaq: QNT), so it reports progress every quarter.

Helios, launched on 5 November 2025, uses 98 trapped barium ions. Quantinuum reported two-qubit gate fidelity of 99.921% and 48 error-corrected logical qubits, with Amgen, BMW, JPMorgan Chase and SoftBank among the launch customers. In its second-quarter 2026 results (11 August 2026), the company said it had demonstrated “near five-nines logical fidelity” on Helios with a new family of error-correcting codes, and announced Helios access through Oracle Cloud Infrastructure.

Next comes Sol in 2027, Quantinuum’s first commercial system with a two-dimensional ion-trap chip. The company says the trap chip is back from fabrication and in validation. Apollo, the fully fault-tolerant machine, “remains on schedule for 2029.”

Trapped ions have a known weakness: they are slow. Operations take much longer than on superconducting chips, so a trapped-ion machine with fewer errors may still finish a long calculation later. When you compare vendors, ask about time to solution, not only qubit counts and fidelities.

Microsoft’s roadmap: Majorana 2 and its critics

Microsoft is the outlier. It is trying to build topological qubits, which in theory are protected from many errors by the physics of the device itself. If they work, they could need much less error correction than other approaches. The question since Majorana 1 in February 2025 has been whether they work at all.

On 2 June 2026 Microsoft announced Majorana 2. The company says a new materials stack (lead instead of aluminum, plus a new semiconductor region) makes the qubits 1,000 times more reliable than the previous generation, and that it now expects a scalable quantum computer by 2029, half its original timeline.

Independent physicists were not persuaded. In Scientific American, Henry Legg of the University of St Andrews and Sergey Frolov of the University of Pittsburgh said the results rest on limited data in a preprint that has not passed peer review. The disagreement is about basic evidence, not about engineering details.

Microsoft is also hedging. With Atom Computing it is building Magne, a neutral-atom machine in Lyngby, Denmark, that Microsoft describes as the first operational deployment of a machine powered by logical qubits. Construction started in autumn 2025 and operations are expected by late 2026. Whether Magne switches on this year is a concrete test of Microsoft’s error-correction software, separate from the Majorana debate.

Why do all four roadmaps say 2029?

Part of it is engineering. Error correction below threshold has now been shown on superconducting and trapped-ion hardware, and the remaining work (more qubits, better wiring, faster decoding) looks hard but not mysterious. Part of it is competition: nobody wants to be the company that said 2033 when its rivals said 2029.

The US government is pushing in the same direction. On 17 September 2026 the Department of Energy announced Genesis Q, a competition with up to $215 million (subject to appropriations) for a first-generation fault-tolerant, scientifically useful quantum computer by 2028. The final phase asks for at least 100 logical qubits running 100,000 operations. The Register noted that no one has yet built a machine at that level.

The 2029 date also matters for security. Google (March 2026) and Microsoft (June 2026) both set 2029 as the deadline to finish their own post-quantum cryptography migrations. Our article on when Q-Day might arrive covers the attack estimates. A useful machine for chemistry and a machine that breaks RSA are not the same thing, but the companies building the hardware are acting as if the second could follow the first sooner than they once thought.

How should executives read a quantum computing roadmap?

Read it as a list of tests the vendor has set for itself, and check each one only when there is a paper or a machine outsiders can use. Four habits help.

First, look for logical error rates, not just logical qubit counts. Forty-eight logical qubits that each fail once in a hundred operations cannot run a long algorithm.

Second, ask what classical methods can do on the same problem. Several advantage claims since 2019 were later matched by better classical algorithms. Our review of whether quantum computing is useful yet goes through the cases.

Third, separate delivered hardware from planned hardware. Nighthawk, Willow and Helios exist. Kookaburra, Sol, Starling and Apollo do not yet.

Fourth, watch for slips. A one-quarter delay is normal. A missed year on a chip that sits on the critical path to 2029 changes the picture.

Which signals should change your plans?

We track these events because each one is public, dated and easy to verify.

If two or three of these happen on time, the 2029 dates start to look solid, and the case for starting pilots in chemistry, materials and some optimization problems gets stronger. If they slip, you have lost nothing by waiting on the opportunity side. The security side is different: the migration to post-quantum cryptography should be moving now in either case, and the post-quantum cryptography migration page explains what that involves.

What to do with this now

Most companies do not need a quantum hardware opinion. They need a short list of the milestones above, a named person who checks them twice a year, and an agreed action for each one. That is what we build in a quantum strategy and roadmap engagement: your own plan, tied to these external signals, with a separate track for the cryptography work that can’t wait. If you want a quicker first look, the quantum readiness assessment covers both sides in four to six weeks.