Clouds of Uncertainty Dog Microsoft’s Majorana Qubit Claims

FeatureClouds of Uncertainty Dog Microsoft’s Majorana Qubit ClaimsA peer-reviewed critique in Nature alleges that Microsoft’s 2025 Majorana result depended on coding errors and on data the company presented selectively. Microsoft maintains that the errors were trivial and that its physics is sound. The more fundamental disagreement is one that no software correction can resolve. On 24 June, the journal Nature published a formal challenge to one of the most prominent claims in modern quantum computing. The critique was written by Henry Legg, a condensed-matter physicist at the University of St Andrews in Scotland whose research centres on the semiconductor-superconductor nanowires that Microsoft’s approach depends on, and it appeared in the journal’s Matters Arising section, which is reserved for contesting previously published research. Its subject is the February 2025 paper by Microsoft Azure Quantum that anchored the company’s Majorana 1 chip announcement. Microsoft was granted a right of reply, published in the same issue, allowing readers to weigh the accusation and the rebuttal together. In this articleWhat Microsoft claimedWhat a topological qubit requiresHow Microsoft’s chip actually worksThe critique in two partsMicrosoft’s responseThe core disagreementA second chip, and the same questionTwo tracks, and only one is its ownClouds that have not clearedA caveat, and a measure of credit Nature accepted the paper on 20 April but did not publish it until late June. By then, Microsoft had already introduced a successor chip and accelerated its target for a commercially useful quantum computer to 2029. The timing ensured that the critique reached the public during a period of renewed promotion, which has made it considerably harder to dismiss. What Microsoft claimed The February 2025 paper reported that Microsoft had performed single-shot parity readout of a device tuned into a topological phase, the regime in which Majorana modes are theorised to exist. The hardware consisted of an indium arsenide nanowire coupled to aluminium and cooled to near absolute zero. Chief executive Satya Nadella described a clear path to a processor of one million qubits, while the paper itself framed a useful machine as achievable in “years, not decades.” These are substantial commitments to rest on a single class of measurement. Whether that promise holds depends entirely on whether the device is truly topological. Understanding why the claim is contested requires a brief account of how a topological qubit is constructed. What a topological qubit requires The term “Majorana” is used in two distinct senses, and the two are frequently conflated. In 1937, Ettore Majorana predicted a particle that would be its own antiparticle, and in particle physics the neutrino remains the leading candidate, though that has not been experimentally confirmed. Microsoft is not pursuing that particle. Its objective is the Majorana zero mode, a collective excitation that can form at the end of a precisely engineered wire. The concept is most easily grasped by imagining a single electron divided into two halves, with each half confined to one end of the wire. The electron’s quantum state then belongs to the pair rather than to either location. Information is stored as parity, meaning whether the shared state is occupied or empty, and because that information is distributed across both ends, a disturbance at one end cannot read or corrupt it. Accessing the information requires acting on both ends at once, a condition that becomes exponentially more demanding as the ends are separated. Figure 1. How a topological qubit stores information. A single electron is split into two Majorana zero modes, one at each end of a semiconductor nanowire coated with a superconductor and cooled to near absolute zero. The qubit value is held as the shared parity of the pair rather than at either location, so a disturbance at one end cannot read or corrupt it. This property explains why Microsoft has pursued the approach for nearly two decades. Conventional platforms such as superconducting transmons and trapped ions store information locally and depend on extensive error correction to preserve it, whereas a topological qubit is intended to be protected by its physical structure. A single topological qubit requires at least four Majorana modes, and logic operations would be performed by braiding, in which the modes are moved around one another so that the outcome depends on the topology of their paths rather than on the precise trajectory. Braiding alone yields only part of the gate set a universal machine needs, so a further, non-topological operation would still be required. The approach carries one decisive condition. The protection exists only if the wire possesses a clean energy gap, known as the topological gap, that suppresses competing low-energy states. Without that gap, parity ceases to be a reliable means of storing information. Compounding the difficulty, several conventional phenomena can imitate the signatures of Majorana modes, among them disorder within the wire and accidental quantum dots, which are small concentrations of charge that produce comparable conductance features. Distinguishing an authentic, well-separated pair from a convincing imitation is the central experimental challenge of the field, and it is precisely the question on which Legg and Microsoft disagree. How Microsoft’s chip actually works Microsoft’s hardware is not yet a working qubit but a test device built to look for Majorana modes. It consists of a nanowire of indium arsenide, a semiconductor, laid against a strip of aluminium, a superconductor, and cooled to a few thousandths of a degree above absolute zero. A set of electrostatic gates beneath the wire tunes the density of electrons until, the company says, the wire enters the topological phase in which a Majorana zero mode forms at each end. Because controlled braiding of the modes is still out of reach, Microsoft does not yet manipulate them to perform logic. Instead it attempts to read the device’s parity, meaning whether the shared state holds an even or an odd number of electrons, using an interferometric single-shot measurement. The method links the two ends of the wire into a loop and probes the device’s quantum capacitance with radio-frequency reflectometry. Microsoft reports that the resulting signal settles into two distinct values that alternate with a regular magnetic-flux period, the single flux quantum (h/2e) expected of such a loop, which it reads as the signature of a genuine topological state. Choosing where on the device to take that measurement is the task of the Topological Gap Protocol, an automated procedure intended to find regions with a clean energy gap and to reject the false positives that have misled the field before. The whole construction therefore rests on two supports, a tuning protocol that selects the operating point and a capacitance signal that is recorded there. Whether either support holds is exactly what Henry Legg’s critique sets out to test, which is where the dispute begins. The critique in two parts Legg’s argument has two components that are best considered separately. The first concerns the software Microsoft used to validate its devices. The second concerns the content of the underlying measurements. How the validation software is questioned Microsoft assesses its devices using the Topological Gap Protocol, an automated test designed to identify suitable operating regions and to exclude the false positives that have undermined earlier work in the field. Legg contends that the test is unstable. When it is supplied with different but equally reasonable parameters, its classification can shift between “gapped” and “gapless,” and in one device the region Microsoft presents as suitable is judged unsuitable under an alternative setting. In some instances, a single data point determines the result. Legg also identifies what he describes as two errors in Microsoft’s data processing. The first, he says, instructed the analysis to display only the largest promising region and to omit the remainder, while the second allegedly reordered a data array by its index position rather than by the physical voltage it represented, an error that mattered because the voltages were not symmetric about zero. Correcting both, Legg reports, reveals additional regions that Microsoft did not examine and shows that one device’s readout occurred in a secondary region rather than the principal one the test had identified. The implication is significant. When a reviewer asked whether other regions existed in which the method did not succeed, Microsoft replied that the region it used was the only one to pass within the range it had searched. Legg argues that the statement was inaccurate, because, in his analysis, the omitted regions did exist, and that the central result therefore never received the verification the reviewers had requested. The data Microsoft did not publish The second component is the more consequential. Microsoft published no raw transport data, providing only the protocol’s classification maps, so Legg obtained the underlying measurements from the company’s own public data repository. In his assessment, they do not resemble a clean superconductor. The wires display an abundance of low-energy states, conductance extends across the full voltage range, the peaks are poorly defined, and the two ends of a wire behave inconsistently. He further reports what he reads as the characteristic indications of quantum dots, including regions of negative local conductance and a breakdown of the symmetry that a properly gapped device should preserve, which, if correct, would sit uneasily with Microsoft’s statement to reviewers that it had observed no such dots. Summarising his findings for reporters, Legg compared examining Microsoft’s device to opening a precision Swiss watch and finding “a chaotic jumble of mismatched parts,” and concluded that the company may be centuries, not decades, from a working machine. Microsoft’s response Microsoft’s reply, with quantum hardware lead Chetan Nayak as corresponding author, does not address Legg’s points individually. Instead it reframes the dispute, and it does so explicitly. The protocol, the company states, was never intended as evidence, but served only as a tuning procedure for selecting operating points. The evidence, Microsoft maintains, lies elsewhere, in radio-frequency measurements of the device’s quantum capacitance that produce a two-state signal oscillating with a regular magnetic-flux period. A gapless, disordered wire could not sustain that signal, the company argues, because it would degrade into noise, and the magnitude of the capacitance step indicates a well-developed gap. On the technical specifics, Microsoft offers conventional explanations. Conductance within the gap does not necessarily indicate its absence, the company says, because at the junction transparency it employs, a single subgap state can broaden into a feature that merely appears continuous. The symmetric components Legg highlights are consistent with the expected physics rather than evidence of gaplessness, and negative conductance has several possible origins, of which quantum dots are only one. The coding errors, Microsoft states, amount to a single off-by-one-pixel discrepancy that alters the figures negligibly, introduces one additional region in one device, and leaves the readout location classified as gapped. Its principal counter-argument is that Legg proposes no alternative model capable of reproducing the capacitance signal, and therefore identifies anomalies without accounting for the result. The core disagreement Beneath the technical detail, the two parties are not evaluating the same evidence, which is why neither can resolve the matter conclusively. Microsoft reasons forward from the capacitance signal: the signal was observed, the signal requires a gap, and therefore a gap was present. Legg reasons in the opposite direction, holding that the gap must be demonstrated independently and that the transport data do not establish it. He told The Register that Microsoft had dismissed substantive problems as minor bugs while rearranging its hierarchy of evidence, in effect treating the result as proof of its own preconditions. The following table presents the principal points of disagreement. It sets each contested issue beside Legg’s case and Microsoft’s reply. The issueLegg’s caseMicrosoft’s replyRole of the protocolCentral, since it selects where readout is claimed and so determines every resultA tuning tool only, with no role in interpreting the capacitance data that carry the conclusionsConsistency of the testThe same region shifts between suitable and unsuitable on arbitrary settingsThe test is probabilistic with a bounded false-positive rate, and the thresholds were set to keep it lowCoding errorsTwo errors allegedly hid additional regions from view and sent one device’s readout to a secondary regionA single off-by-one-pixel discrepancy that does not change the readout region’s classificationStatement to refereesLegg says it was inaccurate, because other passing regions existed but were omittedConsistent, because those regions lay outside the range that was searchedThe raw dataDisordered and gapless in appearance, with the signatures of quantum dotsSubgap conductance reflects broadened discrete states, not a missing gapWhat establishes the qubitA gap is required, and the data do not demonstrate oneA gapless device could not produce the observed flux-periodic signal A second chip, and the same question Microsoft has not slowed its programme. On 2 June, at its Build conference, the company introduced Majorana 2, a second-generation topological chip that it describes as a thousand times more stable than its predecessor. Whereas Majorana 1 maintained its states for milliseconds, Microsoft now reports a mean qubit lifetime of approximately 20 seconds, with some instances reaching a full minute, in addition to microsecond gate operations and a qubit measuring roughly one hundredth of a millimetre. The most significant change lies in the materials. The chip replaces the aluminium superconductor with lead, which Nayak says shields the fragile states from cosmic radiation and, according to Microsoft’s figures, more than doubles the topological gap. The chip was developed with assistance from Microsoft Discovery, the company’s agentic-AI research platform, and the company has accelerated its timeline for a scalable system to 2029. The figures are notable, and the materials advance appears substantive. The accompanying paper, however, does not demonstrate a working qubit. It documents a long-lived parity measured on a single wire of one qubit within a small four-qubit array, and reports only Z-direction measurements, omitting the complementary X measurements required before the device qualifies as a qubit, let alone before a logic gate can be performed. It relies on a new radio-frequency tuning method that Microsoft characterises as separate from the disputed protocol, and it does not establish consistent performance across multiple identical chips. Prominent physicists were quick to criticise it. Legg told The Register that the announcement had not changed his assessment of Microsoft’s work, and, speaking to Scientific American, he faulted the new preprint for resting on only a handful of devices without public evidence of reproducibility. The question of reproducibility that has accompanied this field since 2012 remains unresolved. The open question is how Legg’s findings bear on Majorana 2, and whether any of his criticisms have been addressed in the newer work. His critique concerns the Majorana 1 transport data and the protocol used to validate them, whereas the Majorana 2 paper rests on a different device and a tuning method the company describes as separate from that protocol, so the two do not directly engage. The publication dates compound the difficulty, since the formal scientific record is still litigating a chip from early 2025 even as Microsoft promotes its successor, an unfortunate clash of timing rather than a true meeting of arguments. For now, the evidence points more towards relocation than resolution, since the objections that attached to Majorana 1, the reliance on the protocol and the absence of an independent demonstration of the topological phase, carry over to the new device. Legg has already dismissed the chip in the press as unproven, but a formal critique of the kind he directed at its predecessor can only follow once the new work clears peer review, since Majorana 2 currently exists as a preprint. Figure 2. The claim-and-pushback cycle, from the first nanowire signatures in 2012 to the June 2026 exchange in Nature. Two tracks, and only one is its own It is easy to forget, amid the Majorana dispute, that Microsoft runs two quantum programmes in parallel. The first and more immediate one does not use Microsoft’s own hardware at all.

Through Azure Quantum the company has partnered with Quantinuum, whose trapped-ion machines ran Microsoft’s qubit-virtualisation software to produce four logical qubits with record reliability in 2024, and with Atom Computing, whose neutral-atom array Microsoft used to entangle twenty-four logical qubits later that year. In each case the qubits are someone else’s atoms or ions, and Microsoft’s contribution is the error-correction and control software wrapped around them. The second track is the topological programme examined above, and it is the only hardware Microsoft builds itself. That division is the real shape of the company’s position. The results Microsoft can stand behind today rest on partners’ machines, while the Majorana moonshot, the part that is distinctly its own, remains unproven. This is less a criticism than a clarification. The near-term partnerships give Microsoft a credible presence in the current era of noisy machines, and the topological bet gives it a shot at a decisive long-term advantage if the physics holds. The two tracks hedge each other, yet they should not be confused, because the confident demonstrations belong to the borrowed hardware and the open questions belong to the chip Microsoft is trying to make its own. Clouds that have not cleared The exchange resolves a limited set of questions. The coding error is acknowledged and corrected, and the dispute over it has narrowed to whether it was material. The underlying physics, by contrast, is unchanged. Each side maintains that the other has conducted an incomplete analysis, and no independent authority can compel a resolution. For the reader, the reasonable conclusion is that the central claim is neither confirmed nor refuted, but considerably less secure than the original announcement implied. That uncertainty is the substance of the story, and it is why “breakthrough” was a premature description. A topological qubit is only as reliable as the gap that protects it, and the question of whether that gap was present has reverted from a settled premise to an open one. Until Microsoft demonstrates the gap directly, rather than inferring it from a signal that presupposes it, the uncertainty surrounding the Majorana programme is unlikely to dissipate. The company may ultimately be vindicated. On the evidence published to date, however, its case rests on a measurement whose foundation its critics, now formally on record in Nature, maintain they cannot identify. A caveat, and a measure of credit Two points deserve emphasis in closing. The first is that experimental claims at this frontier are notoriously difficult to establish, because the signatures of exotic physics are faint, ambiguous and readily mimicked by mundane effects. That this dispute is being conducted as a published critique and reply in a peer-reviewed journal is not a failure of the process but the process operating exactly as intended, with adversarial scrutiny doing the work it is designed to do. The second is that Microsoft’s strategy merits real credit. While much of the industry has competed on raw qubit counts, assembling the ever larger arrays of noisy devices that have historically defined the noisy intermediate-scale quantum (NISQ) era, Microsoft has held to its stated north star of first making a single, high-quality qubit work before scaling to larger systems. That is a disciplined and defensible order of operations, and it reflects a long-term conviction rather than a quarterly headline. Microsoft now runs this as two tracks at once. The patient topological bet sits alongside the partner-based machines on Quantinuum and Atom Computing hardware, so the discipline praised here describes the Majorana effort specifically rather than the company as a whole. It is also among the most ambitious bets in the field, a genuine Majorana moonshot, and that ambition is the source of its appeal. The topological approach is harder and slower than its rivals precisely because it aims at hardware-level protection that the others must engineer in software, and should the physics ultimately hold, that advantage could allow Microsoft to advance from the back of the field with unusual speed. For all the uncertainty documented above, this remains the programme with perhaps the greatest capacity to surprise. Editorial note. This article reports a scientific disagreement that is being conducted through the peer-reviewed literature.

Dr Henry Legg’s criticisms are arguments published in Nature’s Matters Arising section and reported here as his contested analysis, not as established fact. They are not allegations of fraud, research misconduct, or bad faith, and we make no such allegation against Microsoft, Dr Chetan Nayak, Mr Satya Nadella, or anyone else. Microsoft rejects the criticisms, and its peer-reviewed reply, published in the same issue of Nature, is summarised above. No court, regulator, or research-integrity body has adjudicated this dispute. Any reference in this article to ‘errors,’ ‘omissions,’ or data that was ‘selectively presented’ or ‘not examined’ reflects the parties’ competing scientific positions rather than any finding of wrongdoing. Not investment advice. This article is for information and analysis only. Nothing in it is a recommendation to buy, sell, or hold any security. The quantum-computing claims discussed here, including Microsoft’s, remain scientifically contested and should not be used as a basis for financial decisions. Sources: H. F. Legg, “On the robustness of topological gap detection via transport,” Nature 654, E22-E26 (2026); Microsoft Quantum’s reply in the same issue; Microsoft Azure Quantum, “Interferometric single-shot parity measurement in InAs-Al hybrid devices,” Nature 638, 651-655 (2025); Microsoft, “Majorana 2” announcement, 2 June 2026; additional reporting by The Register, The Verge, Scientific American, SiliconANGLE and DCD, June 2026; and the University of St Andrews, Critique published by Nature challenges Microsoft’s quantum computing claims. Stay current. See today’s quantum computing news on Quantum Zeitgeist for the latest breakthroughs in qubits, hardware, algorithms, and industry deals. Tags: