Quantum Computing Companies Shift Race to Logical Qubits in 2026

A crowded field with a new scoreboard

The Quantum Insider now counts 76 major quantum computing companies. That number shows how big the industry has become, and it also hides how unevenly the work is spread. At the top are platform giants such as Amazon Braket, IBM, Google Quantum AI, Microsoft and NVIDIA. Below them is a tier of specialists, each betting on one way of building a qubit. Rigetti, IQM and Atlantic Quantum use superconducting circuits. IonQ, Quantinuum and Alpine Quantum Technologies use trapped ions. QuEra, PASQAL and Atom Computing use neutral atoms. Xanadu, PsiQuantum and Quandela use photonics. Diraq and Silicon Quantum Computing use silicon. Riverlane and Q-CTRL work on noise reduction and error correction.1

What changed in 2026 is how these companies are judged. Raw physical-qubit counts used to dominate the headlines. Now the main yardstick is the logical qubit: a single error-protected unit of information spread across many physical qubits. One industry guide calls the verified logical-qubit count "the new headline benchmark."22 That change changes the rankings, and it explains why the claims in this year's error-correction announcements deserve close reading.

The breakthrough that set the terms

Almost every 2026 analysis starts from Google's Willow result. On Willow, each step up in surface-code size (from 3×3 to 5×5 to 7×7 grids of physical qubits) cut the logical error rate by a factor of about 2.14. The largest configuration, a 101-qubit distance-7 code, reached an error rate of 0.143% per cycle.26 Its logical qubit also kept information about 2.4 times longer than the best single physical qubit on the chip. That matters because many earlier demonstrations still lost to a well-isolated individual qubit.26

The 2026 work built on that foundation in two directions: more logical qubits, and logical qubits that do actual computation instead of just storing data.

The biggest single computation came from IBM and the University of Chicago on July 30. They encoded 70 logical qubits and ran 2,415 logical two-qubit operations plus 468 logical T gates. Effective logical error rates were ten times lower than the physical ones, and the job took about 15 minutes. IBM says leading classical simulation methods would need impractically long runtimes for the same task.24 The team built a structured alternative to random circuit sampling that keeps the hardness guarantees but lets errors be detected during the run. They also published the circuits on an open Quantum Advantage Tracker.24

In June, Microsoft and Quantinuum published a peer-reviewed Nature paper on error correction during computation. Using the carbon and tesseract codes on up to 12 logical qubits, they cut errors by 11 to 800 times compared with running the same circuits on unprotected physical qubits.26 The careful coverage adds two caveats. The 800-fold figure comes from a single benchmark circuit, and some of the gains depend on throwing away runs flagged as faulty, which does not scale for free.28 Even so, the paper is notable because it survived peer review in a field full of press-release claims.28

Where the counts diverge

The biggest disagreement in the coverage is about what counts as a logical qubit. QuEra is widely credited with the record: 96 logical qubits built from 448 neutral atoms, using a high-rate [[16,6,4]] qLDPC code at roughly 4.7 physical qubits per logical one, published in Nature in January 2026.26 Some guides call these "verified" logical qubits.22 A stricter ranking describes the same result differently. It credits QuEra with two error-corrected logical qubits in distance-5 surface codes, and notes that the 96-qubit encodings kept only 0.24% to 3.4% of experimental runs.2

Quantinuum's Helios gets similarly mixed treatment. Several analyses describe 48 error-corrected logical qubits on a 98-qubit machine, a 2-to-1 ratio. They add that Quantinuum's SEC filings call this the first commercial demonstration of that efficiency.2612 Another error-correction guide calls the 48 "detected" rather than corrected.22 The stricter ranking notes that runs were discarded on Helios, and counts eight error-corrected logical qubits from a September 2026 preprint.2 Even the Microsoft–Quantinuum 12-logical-qubit milestone has different dates attached: some accounts place it in 2024, while one tracker dates it to March 2026.1311

Our reading is that the error-detection numbers in the high double digits are real engineering progress. They are not the same thing as a fault-tolerant computer with that many qubits. Results with no runs discarded deserve extra weight. Nord Quantique's single GKP logical qubit, reported in a July preprint, is one example.2 For now, raw counts tell readers less than the method behind each count.

Roadmaps converge on 2029

The big vendors may disagree on qubit types, but they agree on the date. IBM, Google, Quantinuum and Microsoft all now point to 2029 for a useful error-corrected machine. None of them has one today.12

IBM's roadmap has the most dates on it. It delivered the 120-qubit Nighthawk processor and the Loon test chip in November 2025, moved main wafer production to a 300 mm facility in Albany, and said it expects the first verified quantum advantage by the end of 2026.12 Next comes Kookaburra, a qLDPC memory module due in 2026, followed by Cockatoo in 2027 and Starling in 2029. Starling targets 200 logical qubits and 100 million gates.1216 In August, IBM linked and cooled the first two cryogenic modules Starling will run in.16

Google publishes fewer dates. Its third milestone is a long-lived logical qubit that fails about once in a million operations, and it has not declared that milestone complete.12 Its biggest strategic move came on March 24. Hartmut Neven announced a neutral-atom hardware team in Boulder led by Adam Kaufman. His reasoning: superconducting chips scale more easily in circuit depth, while atom arrays scale more easily in qubit count.12 That one of the strongest superconducting labs is hedging on neutral atoms is, in our view, one of the year's most telling signals.

Quantinuum plans Sol for 2027, targeting about 100 logical qubits near "five nines" fidelity. Apollo follows in 2029, aiming for hundreds of logical qubits.14 In August the company reported "near five-nines logical fidelity" on Helios with a new family of codes, and made Helios available through Oracle Cloud Infrastructure.12

The 2029 date matters for security as well as science. Google and Microsoft have each set 2029 as their own deadline for finishing post-quantum cryptography migrations.12 Still, analysts stress that a machine able to break today's encryption would need thousands of logical qubits. That puts it several hardware generations beyond any current roadmap.19

Efficiency becomes the battleground

A quieter trend may matter more over time: error correction is getting cheaper. qLDPC codes promise far better ratios of physical to logical qubits than the surface code, as long as hardware can support long-range connections.22 Alice & Bob says its "Elevator Codes" could deliver 100 logical qubits from 1,500 physical ones.2 IQM says its "barbell codes" cut logical error rates by up to three orders of magnitude while using up to eight times fewer qubits than the surface code. Those claims still need wider scrutiny. IonQ reported a real-time decoder on a single ordinary CPU that kept up with 408 logical qubits in simulation. In October it reported qLDPC logical operations up to 74 times faster. Both results are company-reported, and one is still an unreviewed preprint.29

Governments are starting to define what counts as success. The U.S. Department of Energy's $215 million Quantum Genesis competition effectively sets fault tolerance at 100 logical qubits.29 DARPA's Quantum Benchmarking Initiative is sorting companies into stages: IonQ and Atom Computing reached Stage C in October, while Quantinuum, QuEra, Xanadu and others sit at Stage B.2 Independent checks matter because claims can collapse. In May, Flatiron Institute researchers used tensor networks on a laptop to reproduce a result that had been billed as classically intractable.29

The money follows the milestones

The financial story has moved as fast as the technical one. Quantinuum listed on Nasdaq in June, raising $1.68 billion at $60 a share.216 Valuation estimates differ: one account puts it at about $14 billion, and another at roughly $15 billion, against $30.9 million in 2025 revenue.166 Infleqtion listed on the NYSE in February, Pasqal began trading on Nasdaq in late August, and IQM and Xanadu have also gone public.210 Consolidation is underway too: D-Wave closed its $550 million purchase of Quantum Circuits in January, and IonQ's $1.8 billion SkyWater acquisition is still pending.2

IonQ has the clearest revenue story among the pure plays: $130 million in 2025 GAAP revenue, up 202%.2 For most of the 76 companies, though, valuations rest on roadmaps rather than revenue.

The verdict

The 2026 coverage agrees on one main point: error correction has moved from a physics question to an engineering one, across superconducting, trapped-ion and neutral-atom hardware.14 Where it disagrees is on how to count, and that disagreement is the story readers should follow. The milestones to watch are no longer qubit totals. They are IBM getting Kookaburra working, the Magne system in Denmark coming online (its timing is reported as late 2026 or early 2027), Google declaring a long-lived logical qubit, and logical-qubit results achieved without throwing away most of the runs.1229 The 2029 target is plausible on paper. Over the next three years, companies will need to show they can deliver error-corrected logical qubits at scale, not just announce them.