Roadmaps

2029: IBM, Quantinuum, IonQ and Google Compared

IBM, Quantinuum and IonQ all aim at fault tolerance around 2029, but count logical qubits, gates and error rates differently. What each roadmap promises, and what has shipped so far.

Gabriele Masetti · 2026-10-22

The Date Four Roadmaps Share

IBM, Quantinuum and IonQ each publish a fault-tolerance roadmap, and each lands on the end of this decade. IBM names 2029 for a machine called Starling. Quantinuum names 2029 for a system called Apollo. IonQ shows a 2029 row and a 2030 row in its public table. Google, the fourth name, wrote its goal in 2021 as "within the decade", which coverage reported as 2029, and in 2026 it set 2029 as a separate date for its own cryptography migration.

2029 is a target that four organisations chose for themselves, and each one measures it in different units. A target is a company's intention, backed by a design and a budget. None of the 2029 machines exists today. The useful reading skill is to convert each promise into comparable terms and then check what each company has already shipped against its earlier dates.

What "Fault-Tolerant" Means in These Documents

A physical qubit is the hardware element, and it is noisy. A logical qubit is an error-corrected qubit encoded across many physical ones, built so that its error rate falls below that of any single physical qubit. A machine counts as fault-tolerant when it can run long computations while error correction keeps pace with the errors that arise.

The roadmaps use these words with different weight. IBM states a gate budget as well as a qubit count. Quantinuum states error-rate targets alongside qubit ranges. IonQ states totals of physical and logical qubits year by year. Google states milestones in sequence and puts an error rate on the one it describes as a long-lived logical qubit.

Ratios matter more than headline counts. A company that needs two physical qubits per logical qubit and one that needs fifty are making very different promises with the same word. Anyone comparing roadmaps should write down, for each entry, the physical qubits, the logical qubits, the logical error rate and the number of operations the machine is meant to survive.

IBM: 200 Logical Qubits and 100 Million Gates

IBM describes its goal as "delivering the first large-scale, fault-tolerant quantum computer by 2029." Its technology page gives the stated target for Starling: running 100 million quantum gates on 200 logical qubits. Coverage of the June 2025 announcement adds that Starling is planned for IBM's quantum data center in Poughkeepsie, New York.

IBM's path to Starling has named intermediate chips. As reported in June 2025, Loon was set for 2025 to test components of the qLDPC error-correcting code, including couplers that link distant qubits on one chip. Kookaburra, planned for 2026, is the first modular processor meant to store and process encoded information. Cockatoo, planned for 2027, is meant to entangle two Kookaburra modules.

Against that schedule, IBM's November 2025 release announced Loon as an experimental processor that it says demonstrates the key components needed for fault-tolerant computing. The same release reported real-time decoding of qLDPC codes in under 480 nanoseconds, which IBM says arrived a year ahead of schedule. Independent coverage repeated both figures, though it carried IBM's own description. Kookaburra, due in 2026, is the next test of that schedule.

IBM also publishes a nearer ladder for its non-fault-tolerant line. Its Nighthawk processor has 120 qubits and 218 tunable couplers and supports circuits of up to 5,000 two-qubit gates. IBM reports that Nighthawk r2 delivers accurate results on circuits containing 7,500+ gates; it expects later versions to reach 10,000 in 2027 and 15,000 by 2028.

Quantinuum: Helios Now, Sol in 2027, Apollo in 2029

Quantinuum's roadmap is named after systems. Helios, its trapped-ion machine launched in November 2025, has 98 physical qubits, and the company's launch materials report a two-qubit gate fidelity of 99.921%. In those materials Quantinuum counts 94 error-detected logical qubits entangled together and 48 fully error-corrected logical qubits at a 2:1 encoding rate. MIT Technology Review's coverage confirms 98 barium ions and the two-ions-per-logical-qubit ratio, and notes that Helios is not powerful enough for commercially useful algorithms.

Sol is the next step. Coverage of Quantinuum's August 2026 earnings call says the company remains on track to launch a 192-physical-qubit Sol in 2027. Apollo, planned for 2029, is described by the company as universal and fully fault-tolerant.

Quantinuum's own 2024 roadmap post gives the units: Apollo with thousands of physical qubits, hundreds of logical qubits, a physical error rate of 10^-4 and logical error rates better than 10^-6. In May 2026, The Next Web reported that the company had filed for a US listing under the ticker QNT, and its coverage repeats Sol for 2027 and Apollo for 2029 as company plans.

IonQ: The Largest Numbers and the Longest Reach

IonQ's public roadmap table is the most ambitious on paper. It lists 10,000 physical and 800 logical qubits for 2027, 20,000 and 1,600 for 2028, 200,000 and 8,000 for 2029, and 2,000,000 and 80,000 for 2030. The logical error rate in the table is below 10^-7 for 2026 through 2028 and below 10^-12 for 2029 and 2030.

The implied ratio is 12.5 physical qubits per logical qubit in 2027 and 2028, and 25 in 2029 and 2030. The ratio doubles while the stated logical error rate improves by five orders of magnitude, from below 10^-7 to below 10^-12. A reader can fairly ask what changes in the hardware or the code to make that jump, since the table does not say. Wording also varies across coverage: IonQ's October 2025 release cites a target of 2 million qubits by 2030, while the roadmap table separates physical from logical counts.

The delivered record is narrower. In October 2025 IonQ reported two-qubit gate fidelity above 99.99%, against a previous record of 99.97% set by Oxford Ionics, now part of IonQ. That result came from prototypes in IonQ's laboratories. The company said it planned to demonstrate 256-qubit systems in 2026, and on its first-quarter 2026 call it described two separate milestones for the 256-qubit technology: a demonstration by the end of 2026, and customer systems beginning commissioning by the end of the second quarter of 2027. The company said it was moving from component tests to system-level testing of a first engineering prototype.

Google: A Decade Goal, Six Milestones and a Prototype

Google wrote its hardware goal in May 2021. Erik Lucero, Lead Engineer at Google Quantum AI, put it this way on Google's blog: "Within the decade, Google aims to build a useful, error-corrected quantum computer." The same post sets the engineering target of 1,000,000 physical qubits working together inside a room-sized error-corrected machine. Coverage of the announcement reported the goal as 2029.

The roadmap that follows has six numbered milestones, from "Beyond classical" to "Large error-corrected quantum computer." It puts no year on each step, and Heise's December 2024 report said the team could not say when the last would arrive. In December 2024 Google announced Willow, a 105-qubit chip, and said that encoded arrays of 3x3, 5x5 and 7x7 qubits each halved the error rate, which it calls operating below threshold.

Hartmut Neven, who leads the effort, framed the result narrowly:

"As the first system below threshold, this is the most convincing prototype for a scalable logical qubit built to date." — Hartmut Neven, Founder and Lead, Google Quantum AI

The wording says prototype: it describes a single logical qubit and no machine with hundreds of them. Coverage of Willow placed it on the way to the third of the six milestones, a long-lived logical qubit, which a December 2024 report put one or two years away.

Google also has a second 2029. In March 2026 its security team wrote, "We're setting a timeline for post-quantum cryptography migration to 2029." The post cites progress in hardware, error correction and factoring estimates as context. It is a deadline for moving cryptography and gives no hardware milestone, qubit count or arrival date for a quantum computer that could break encryption.

Why the Counts Cannot Be Lined Up Directly

Two of the headline numbers can be compared only after conversion. IBM's 200 logical qubits and IonQ's 8,000 for the same year differ by a factor of forty, but IBM attaches a gate budget and IonQ attaches an error rate. A 100-million-gate machine implies a certain logical error per operation, and a below-10^-12 target implies a certain machine lifetime, yet neither company states the other's quantity.

Quantinuum's figures sit between them. Its 2024 target of logical error rates better than 10^-6 is looser than IonQ's 2029 figure by six orders of magnitude. The physical-qubit ratios also differ: Helios encodes at 2:1 in the company's own demonstration, while IonQ's table implies 25:1. A reader who sees only "hundreds of logical qubits" or "8,000 logical qubits" learns very little without the error rate behind it.

The Comparison Table

The gap between the entries is wide. IBM's promise is the smallest in qubits though not the only one stated in gates: Quantinuum describes Apollo as capable of executing circuits with millions of gates. IonQ's promise is forty times larger in logical qubits than IBM's (8,000 against 200), from a company whose published fidelity record comes from a laboratory prototype. Quantinuum's is loose on counts and firm on error rates. Google's states a physical-qubit total and no error rate.

What Has Been Delivered Against Earlier Dates

The delivered record is real and smaller than the targets. IBM announced Loon and decoding results in 2025, the year its roadmap assigned to Loon. Quantinuum shipped Helios in 2025, with 98 physical qubits, and says Sol is on track for 2027. IonQ posted a fidelity record in a laboratory in 2025, and its 256-qubit demonstration is due by the end of 2026. Google demonstrated below-threshold scaling on one logical qubit.

None of these results is a fault-tolerant computer. Each is a component demonstrated before the whole is built, and the step from a demonstrated component to a machine running a hundred million gates is the part that has not been shown. A roadmap date is also a statement about funding, supply chains and engineering schedules, and these tend to move.

How to Read a Roadmap Date

Check four things when a quantum company announces a year. First, find the unit: physical qubits, logical qubits, gates or error rate. Second, find the ratio between physical and logical qubits that the claim implies. Third, find the last dated milestone that was met, and whether the company reported it in the year it promised. Fourth, note who wrote the date: a company's roadmap, a prospectus, an investor call or an independent measurement.

By those tests, the companies compared here are not yet making the same promise, even where they share a year. IBM's is the most concrete, Quantinuum's the most error-rate-driven, IonQ's the largest, and Google's is the oldest, written in 2021 as an aim for the end of the decade. A reader should treat 2029 as four intentions to be checked against delivery, one milestone at a time. None of the four is a forecast that a fault-tolerant computer will exist then.