Quantum Computing Race 2026: Google Willow, IBM, and Microsoft Compared
Google's Willow chip completed a computation in under five minutes that would take the world's fastest supercomputer 10 septillion years — a number that outlasts the age of the universe by a factor the mind cannot hold. That result, published in Nature in December 2024, sent quantum computing stocks surging, prompted Jensen Huang to publicly retract his skepticism, and was quietly accompanied by a caveat the headlines largely skipped: the task has no known practical application. What follows is an attempt to cut through three competing announcements, three distinct technical bets, and a timeline disagreement that moved markets to figure out what is actually happening in quantum computing in 2026.
Most writing about quantum computing falls into one of two failure modes. Either it treats every hardware announcement as proof that the revolution has arrived, or it retreats to the safe ground of "still decades away" and leaves readers no better informed than before. The real picture is stranger and more specific than either account. Google, IBM, and Microsoft are not racing toward the same finish line — they are running three different races, using three different approaches, with three different definitions of winning.
This article measures those approaches against each other using the metrics that matter: error rates, qubit counts, the gap between benchmark performance and real-world utility, and the specific dates each company has committed to publicly. Where the record shows failure or dispute, that is named directly.
- Google Willow: What the Benchmark Actually Proves
- The Error Correction Milestone No One Explained Well
- IBM's Strategy: Ecosystem Over Headlines
- Microsoft Majorana 1: High-Risk Physics, High Reward
- The Criticism the Industry Prefers to Ignore
- The Three Timelines and What Each Company Has Committed To
- Who Is This For
- Verdict
- FAQ
Google Willow: What the Benchmark Actually Proves
The 10-septillion-year figure is not meaningless, but it does not mean what most coverage implied. Google's Willow chip, announced December 9, 2024, was tested using Random Circuit Sampling — a benchmark chosen specifically because it is difficult for classical computers and natural for quantum ones. As the physics community distinguishes, there is a difference between a quantum computer demonstrating it can do something classical machines cannot feasibly simulate, and demonstrating it can solve a practical problem faster than any classical machine ever could. Willow's RCS result establishes the former condition, not the latter. That distinction rarely made it into the headline.
Winfried Hensinger, a quantum computing physicist at the University of Sussex, was direct: Willow is "still well too small to do useful calculations," and quantum computers will likely require millions of qubits to tackle the problems — drug discovery, materials design, financial optimization — where the technology is expected to matter. Willow has 105. The gap between those two numbers is not a rounding error; it is roughly four years of aggressive hardware development on current trajectories, assuming no major setbacks.
Julian Kelly, Google Quantum AI's director of quantum hardware, acknowledged this plainly. Succeeding at the RCS benchmark is, in his framing, a "necessary but not sufficient" condition for a useful quantum computer. A chip that cannot win at RCS cannot win at anything harder. Willow winning at RCS means it clears the first bar — not that it has finished the race.
What Willow actually achieved technically: 105 superconducting transmon qubits with coherence times (T1) of approximately 100 microseconds — a fivefold improvement over the predecessor Sycamore chip's 20 microseconds. Average qubit connectivity of 3.47 in a square lattice. Single-qubit gate fidelity of 99.97%, two-qubit gate fidelity of 99.88%, and readout fidelity of 99.5% across the full array. These are real advances. They just do not translate directly into commercially useful computations yet.
In March 2026, Google added neutral atom hardware as a second modality alongside its superconducting approach, positioning the lab to pursue both circuit depth and qubit count in parallel. That decision signals Google's own awareness that no single qubit technology has demonstrated clear superiority across every metric that matters for fault-tolerant computation.
The Error Correction Milestone No One Explained Well
The genuinely consequential result from Willow is not the septillion-year benchmark. It is the error correction data. For nearly thirty years, quantum computing faced a fundamental problem: adding more qubits increased error rates. The two quantities moved together. Willow broke that relationship.
Google demonstrated that scaling a surface code from a 3x3 grid of qubits to a 5x5 grid to a 7x7 grid cut the logical error rate in half each time — an exponential reduction as qubit count increased rather than an exponential increase. This is what the field calls "below threshold." Getting there has been the central technical goal of superconducting quantum computing for the better part of three decades. The Willow team also demonstrated real-time error correction during computation on a superconducting system, which is the operational requirement for running any practically relevant algorithm.
"As the first system below threshold, this is the most convincing prototype for a scalable logical qubit built to date." — Hartmut Neven, Founder, Google Quantum AI
What this unlocks is not usefulness today — it is proof that scaling toward usefulness is physically possible. Before Willow, there was legitimate scientific uncertainty about whether superconducting systems could ever reach below-threshold error correction. That uncertainty is gone. The question has shifted from whether to how long.
Scaling to fault-tolerant operation still requires jumping from 105 physical qubits to likely millions. Each physical qubit in a surface code error correction scheme contributes to what is called a logical qubit — the unit of computation that is actually reliable enough to use. Getting one high-quality logical qubit currently requires roughly a thousand physical qubits, depending on the target error rate. Willow's 105 qubits can support perhaps one or two logical qubits at the quality threshold needed for useful computation. That number needs to reach thousands of logical qubits before most commercially interesting problems become tractable.
The decade closes.
IBM's Strategy: Ecosystem Over Headlines
IBM does not have a single result as cinematically dramatic as the septillion-year benchmark. What it has instead is the most mature quantum computing ecosystem of any company, a 300-person-plus Quantum Network of research institutions and enterprises, and a roadmap it has largely tracked on schedule since 2020.
At the annual Quantum Developer Conference in November 2025, IBM announced the Nighthawk processor — 120 qubits on a square lattice with four-degree qubit connectivity and 218 tunable couplers. The architecture allows circuits roughly 30 percent more complex than on the previous Heron processor while maintaining low error rates. IBM expects Nighthawk to support up to 7,500 two-qubit gates by the end of 2026, reaching 10,000 gates in 2027.
The same announcement introduced IBM Quantum Loon — an experimental chip designed specifically to test the hardware building blocks of error-corrected quantum computation. IBM is running two parallel tracks: near-term quantum advantage through Nighthawk and the HPC integration program, and long-term fault tolerance through Loon and the gross code architecture targeting the Starling processor in the late 2020s.
IBM's committed public dates:
- Quantum advantage demonstration via Nighthawk and HPC integration by the end of 2026 — IBM's own stated target, not an analyst estimate.
- Starling processor (approximately 200 logical qubits) targeted for the late 2020s as the first fault-tolerant system.
- Fault-tolerant quantum computer with a large-scale fault-tolerant architecture and quantum-centric supercomputers beyond 2033.
The approach IBM chose after 2023 is worth understanding. The company built a 1,121-qubit Condor chip and decided not to release it publicly. The reason was not failure — it was a strategic judgment that monolithic chip scaling was not the path to fault-tolerant quantum computing. IBM pivoted to modular architecture and error correction quality, which is why the qubit counts of recent IBM chips look smaller than the Condor headline number. That pivot was correct. Qubit quantity without quality is a marketing metric, not a scientific one.
What IBM offers that Google does not is breadth of access. IBM Quantum is available to external users through the cloud and at client sites. Qiskit, IBM's open-source quantum development kit, has attracted hundreds of thousands of developers. When quantum computing eventually reaches commercial relevance, the company with the largest developer ecosystem and the most enterprise relationships will have structural advantages that pure hardware leadership cannot easily overcome. IBM understands this, and has been building it for a decade.
Microsoft Majorana 1: High-Risk Physics, High Reward
Microsoft is not competing with Google and IBM in any straightforward sense. The company made a bet in the late 2000s on an entirely different physical approach to qubits — one that, if it works, would change the fundamental economics of quantum error correction. If it does not work, Microsoft will have spent roughly fifteen years pursuing a dead end.
Microsoft unveiled Majorana 1 in February 2025 — an eight-qubit processor built from a new class of materials called topoconductors, combining indium arsenide (a semiconductor) with aluminum (a superconductor). The chip is designed to host quasiparticles called Majorana zero modes, which encode quantum information in a fundamentally different way than superconducting transmons or trapped ions. The theoretical promise is that topological qubits are inherently protected from environmental noise by their physical structure, requiring far fewer redundant physical qubits per logical qubit than conventional approaches.
Microsoft's published roadmap outlines four generations of devices from the single-qubit proof-of-concept to a topological qubit array supporting lattice surgery on two logical qubits. The company was selected as one of two organizations advanced to the final phase of DARPA's US2QC program, which provides independent validation of the approach's plausibility toward a utility-scale machine by 2033.
The Criticism the Industry Prefers to Ignore
Microsoft's topological qubit program has a documented credibility problem that advocates in the field are reluctant to state plainly. In 2018, the company identified and corrected a fabrication issue that had produced false experimental signals — a retraction that cost the program years and raised questions about internal verification standards that were never fully resolved in the public record.
Those questions returned in June 2026, when Nature published a formal peer-reviewed critique of the Majorana 1 paper by Henry Legg, a physicist at the University of St. Andrews. Legg's analysis, published in Nature's "Matters Arising" section — the journal's designated venue for formal challenges to published work — argued that Microsoft did not conclusively demonstrate a working topological qubit, and that the software tool used to validate its results contained coding errors that concealed data from peer reviewers. Legg's core position: the experimental signatures Microsoft pointed to as evidence of Majorana zero modes can be explained by more conventional physics.
Microsoft published a rebuttal in the same issue, disputing Legg's interpretation. The company stated that Legg's critique "does not constitute a substantial scientific challenge to our findings." Physicist Jason Alicea of Caltech said the validation protocol "has merit." But Winfried Hensinger, from the University of Sussex, was blunter about the earlier version of the paper: "The peer-reviewed publication is quite clear that it contains no proof for topological qubits. But the press release speaks differently. In academia, that's a big no-no."
This is the failure Microsoft advocates actively avoid naming. The topological gap protocol — the test Microsoft uses to confirm a device is in the topological phase where Majorana modes exist — is itself contested. Until Microsoft can demonstrate that gap directly, rather than inferring it from a measurement that presupposes the gap's existence, the scientific foundation of the entire Majorana program remains an open question rather than a settled premise. Microsoft may ultimately be vindicated. On the evidence published to this date, that vindication has not arrived.
The company announced Majorana 2 at Microsoft Build in late June 2026, doubling down on a technology now formally under scientific dispute in Nature. Either that confidence is well-founded and the critics are wrong, or Microsoft is building its quantum roadmap on something that has not been proven to exist in the way claimed. Those are the only two possibilities, and neither has been resolved.
The Three Timelines and What Each Company Has Committed To
The best way to evaluate competing quantum programs is to ignore what each company says it will eventually achieve and focus on what each company has committed to publicly, by when, with what evidence.
| Company | Current Hardware | Near-Term Commitment | Fault-Tolerant Target |
|---|---|---|---|
| Willow (105 qubits, superconducting) | Long-lived logical qubit as next milestone (no specific date) | End of the 2020s | |
| IBM | Nighthawk (120 qubits), Heron r3 (156 qubits) | Quantum advantage via HPC by end of 2026 | Starling (~200 logical qubits) by late 2020s; fault-tolerant supercomputer by 2029 |
| Microsoft | Majorana 1 (8 topological qubits, under dispute); Majorana 2 (announced June 2026) | Scaling topological qubits; DARPA US2QC program target 2033 | Utility-scale machine by 2033 |
One number that does not appear in most quantum computing coverage: a surface code quantum computer capable of running Shor's algorithm to crack 2048-bit RSA encryption would require an estimated 4,000 logical qubits — which translates to roughly four million physical qubits at current error rates. No company is close. Google's Willow is at approximately two logical qubits. IBM's most advanced systems are in the range of a few dozen. Microsoft's topological approach, if the physics holds, could potentially reach comparable logical qubit counts with fewer physical qubits — but the physics is what is currently under dispute.
Jensen Huang's January 2025 comment that practical quantum computers are twenty years away crashed quantum stocks. His March 2025 retraction at Nvidia's GTC Conference — where he invited the quantum industry to explain why he was wrong — was unusual enough that he opened the session by calling it "the first event in history where a company CEO invites all of the guests to explain why he was wrong." By June 2025, Huang had reversed position at GTC Paris, describing quantum as reaching an "inflection point" with real-world problem-solving "in the coming years." Nvidia's venture arm subsequently invested in Quantinuum, QuEra, and PsiQuantum. That reversal, in six months, from "twenty years" to "inflection point" tells you something about how much uncertainty remains in any timeline estimate — including the optimistic ones.
Who Is This For
- You are a researcher or PhD student deciding whether to build a career in quantum computing. The field is real, the hardware progress is documented, and the job market is growing — but the timelines are long and the failure rate of specific technical approaches is high. IBM's ecosystem has the most entry points. Google's hardware is the most technically credible on benchmarks. Microsoft's approach has the highest potential upside and the highest documented risk.
- You run enterprise technology procurement and someone in your organization is asking whether to begin quantum readiness planning. The honest answer is yes, but not for deployment — for talent pipeline development, algorithm scouting, and understanding which of your computationally intensive workloads (logistics optimization, drug simulation, portfolio risk) would benefit first when hardware matures. IBM Quantum's cloud access is the most practical entry point for hands-on exploration now.
- You are tracking this as an investor and trying to understand whether 2026 is the inflection. It is probably not the year quantum computers replace anything classical. It is the year the hardware credibility of the superconducting approach was established beyond reasonable scientific doubt, which is a different and more durable signal than a benchmark number.
Verdict
Google leads on hardware credibility. The below-threshold error correction result from Willow is the most significant experimental quantum result of the decade, and the company's roadmap is the most technically coherent of the three. If you are evaluating which company is closest to demonstrating something genuinely useful, Google is ahead on the physics.
IBM leads on practical access and commercial infrastructure. The Nighthawk architecture, the Qiskit developer ecosystem, and the explicit commitment to quantum advantage by the end of 2026 via HPC integration give IBM the strongest near-term commercial position. For enterprise organizations exploring quantum now, IBM is the most productive starting point.
Microsoft's position is genuinely uncertain in a way that is not typical scientific humility — it is active scientific dispute in a top peer-reviewed journal, echoing a pattern the company has navigated before. Topological qubits, if the Majorana physics holds up, would make Microsoft the long-term winner by a wide margin. If the physics does not hold up, the company has spent fifteen years building toward something that does not exist. That bet is not irrational — it is a high-variance strategy from a company with enough resources to absorb the risk. But calling it a leading position in 2026 requires ignoring the June 2026 Nature critique, and that critique should not be ignored.
The race does not have a single winner yet. What 2026 has established is that quantum computing is not a promise anymore — it is an engineering problem with a known error correction path, contested timelines, and one approach whose fundamental physics remains under formal scientific dispute.
The question nobody in the field asks openly is what happens if all three roadmaps slip by five years simultaneously. The companies targeting fault tolerance by 2029 would push to 2034. IBM's 2026 advantage demonstration would become a 2031 goal. None of the individual roadmaps contain a mechanism for catching up if the error rate targets prove harder to hit than the current benchmarks suggest. The history of semiconductor roadmaps — where predicted timelines for each new process node consistently underestimated difficulty as the physics got harder — is not reassuring on this point. Quantum computing may not follow the same compounding trajectory as classical chips. Nothing in the current evidence rules out that it will take longer than anyone publicly projects.
FAQ
What did Google's Willow chip actually prove in quantum computing?
Willow proved that superconducting quantum systems can reduce error rates as qubit count increases, breaking a thirty-year barrier called the error correction threshold. It also completed a specific benchmark task in under five minutes that would take a classical supercomputer longer than the age of the universe — but that benchmark has no known practical application, and Willow has not yet solved any commercially relevant problem faster than a classical computer.
How does IBM's quantum strategy differ from Google's?
IBM prioritizes ecosystem depth and near-term commercial access over headline benchmark results. While Google focuses on hardware records, IBM runs a network of hundreds of enterprise and academic partners, offers cloud-accessible quantum computers, and has committed to demonstrating quantum advantage on real-world problems via its Nighthawk processor and HPC integration by the end of 2026. IBM's approach bets that the company with the largest developer base wins when practical quantum computing arrives.
Is Microsoft's Majorana 1 topological qubit claim legitimate?
It is contested. A peer-reviewed critique published in Nature in June 2026 by physicist Henry Legg argues that Microsoft did not conclusively demonstrate working topological qubits, and that the validation software contained coding errors. Microsoft published a rebuttal in the same issue. Independent researchers at the APS Global Summit noted that stronger experimental evidence — specifically X-measurements demonstrating quantum superposition — has not been independently confirmed. The dispute echoes a 2018 fabrication problem that required correction.
When will quantum computers be useful for real-world problems?
Commercial viability for most enterprises is widely projected for the early 2030s. IBM has committed to demonstrating quantum advantage by the end of 2026 through quantum-HPC hybrid systems. Google's roadmap targets a fault-tolerant machine by the late 2020s. DARPA's quantum benchmarking program is focused on utility-scale performance by 2033. Each of these targets carries meaningful execution risk, and quantum computing timelines have historically proven optimistic.
What is quantum error correction and why does it matter?
Quantum error correction is the technique of grouping many unreliable physical qubits into a single reliable logical qubit. It matters because individual qubits lose their quantum state quickly through environmental interference, making computations unreliable above a few hundred operations. Without error correction, quantum computers cannot run the deep circuits required for useful algorithms. Google's Willow was the first superconducting system to demonstrate that adding more qubits reduces rather than increases error rates — the foundational requirement for large-scale error correction.
Google Willow vs IBM Nighthawk — which is the better quantum computer?
They are not directly comparable because they target different things. Willow leads on error correction milestones and raw benchmark performance. Nighthawk leads on circuit complexity for practical workloads, supporting up to 5,000 two-qubit gates and integration with classical HPC systems. For researchers pushing physics boundaries, Willow is the more advanced system. For enterprises exploring near-term quantum applications, Nighthawk's cloud-accessible architecture and IBM's software ecosystem make it the more practical choice.
Can quantum computers break encryption today?
No. Cracking 2048-bit RSA encryption using Shor's algorithm would require an estimated four million physical qubits at current error rates, or roughly four thousand high-quality logical qubits. Google's Willow supports approximately two logical qubits. The threat to current encryption is real but not imminent — security agencies estimate that cryptographically relevant quantum computers are at least a decade away, which is why post-quantum cryptography standards from NIST are being rolled out now rather than later.
Why did Jensen Huang reverse his quantum computing skepticism?
At CES 2025, Huang said useful quantum computers were likely twenty years away, tanking quantum stocks. By March 2025, he publicly admitted his comments were wrong at Nvidia's GTC Conference. By June 2025 at GTC Paris, he described quantum as reaching an inflection point. Nvidia then invested in Quantinuum, QuEra, and PsiQuantum. The reversal reflects both the pace of hardware progress — Willow's below-threshold error correction result arrived between his January comment and his March retraction — and Nvidia's strategic interest in positioning its GPUs as the classical processors that every quantum computer will require to function.