Microsoft’s Majorana Qubit: What the Experts Actually Think

I have never trusted the Majorana story. I should put that bias on the table before I ask you to weigh anything else in this article. My instinct turned wary long before Microsoft etched the word onto a chip and called the result a topological qubit. The trouble is that instinct is the wrong tool here. I am not a condensed-matter physicist, and as I admitted in my earlier Majorana coverage, I cannot personally adjudicate whether a given zero-bias conductance peak is a Majorana zero mode or an ordinary Andreev state in a convincing costume. A gut feeling about Microsoft’s Majorana qubit isn’t evidence about it.

So after Microsoft announced Majorana 2 earlier this month, I started calling people who do this for a living, reading the papers and the comments on the papers, and trying to map where credible expert opinion actually sits.

Then, on June 24, Nature published a peer-reviewed Comment challenging the data behind Microsoft’s parity-readout result, along with Microsoft’s formal reply. That gave me a clean reason to write up what I found.

Before any of the detail: the weight of independent expert opinion has moved well toward skepticism without becoming unanimous, and Microsoft’s most consequential claim, that it has a working topological qubit, still lives in press releases and conference slides rather than in a peer-reviewed result.

This is the mirror image of the Q-FUD problem PostQuantum.com usually fights. Most quantum hype inflates the threat to sell a product. The Majorana story is hype in the other direction: an inflated capability claim meant to win a race for talent, funding, and credibility. The discipline that cuts through both is the same: separate what a company announced from what it has demonstrated in peer review, then ask who is convinced and what would convince them. I have tried to give Microsoft’s defenders the strongest version of their argument below, not a straw man.

What happened on June 24

Henry Legg, a condensed-matter physicist at the University of St Andrews, published a Nature Matters Arising Comment challenging the transport data behind Microsoft’s 2025 single-shot parity-readout paper. Nature ran Microsoft’s reply alongside it, with Chetan Nayak, the company’s quantum lead, as corresponding author.

(The Register covered the exchange under a headline about basic Python errors, which is both accurate and slightly unfair. I’ll come back to why.)

Legg’s argument is that the regions of the device where Microsoft claims to read out qubit parity look gapless and disordered when you examine the raw conductance (the opposite of the clean superconducting gap the interpretation requires). He also found two coding artefacts in Microsoft’s analysis pipeline. One was an antisymmetrization step that indexed bias voltage by array position instead of physical value, which matters because the bias data were not symmetric around zero. The other was plotting code that displayed only the single largest “topological” region and suppressed the others. Correcting both, Legg shows, changes which regions the protocol flags as topological and reveals that the magnetic-field value Microsoft used for its headline measurement sits in a secondary region rather than the primary one. His bottom line is that nothing in the data “proves the existence of a topological qubit or Majoranas.”

In Microsoft’s reply, Nayak argues that their case never rested on the transport data or the gap protocol at all. The evidence is the radio-frequency capacitance measurement: a flux-periodic, two-state telegraph signal that, the reply says, would not survive in a gapless system because the oscillations would wash out. On that account the topological gap protocol was only a tuning tool for finding promising operating points, and Legg is attacking the scaffolding rather than the building. On the bug, Microsoft concedes the off-by-one error but says it shifts the extracted gap by less than 5 microelectronvolts for more than 96 percent of pixels and introduces exactly one new sub-region in one device, with no change to the parity data. The region used for readout, the reply states, stays classified as gapped after the fix.

This is why “basic Python errors” undersells the disagreement. A real bug exists, and Legg found it. But the bug is not the crux. The crux is a scientific disagreement about whether three-terminal conductance data can establish a superconducting gap at all, and whether Microsoft’s capacitance signal means what the company says it means. A reader who walks away thinking the topological program collapsed over an indexing mistake has the wrong picture. So does a reader who thinks a peer-reviewed rebuttal in Nature is nothing to worry about.

Microsoft has announced this breakthrough before

When Microsoft unveiled Majorana 1 in February 2025, Satya Nadella called it “quantum’s transistor moment.” He had used almost exactly that line before. At Microsoft Ignite in September 2017, according to Microsoft’s own transcript, Nadella stood on stage with the topological team, was handed a physical “Majorana chip” by Leo Kouwenhoven, and turned to the physicist Charlie Marcus to say, “this seems like a transistor moment to me.” Eight years apart, the same metaphor, delivered both times as if it were the dawn.

The pattern runs deeper than the branding. In 2012, Kouwenhoven’s group at Delft reported the first “signatures of Majorana fermions” in a nanowire device. The language then was careful: signatures, consistent with, support for the hypothesis. Two of that paper’s co-authors, Sergey Frolov and Vincent Mourik, later became the field’s most determined critics. In 2018 came a much stronger claim, a Nature paper reporting quantized Majorana conductance as a near-smoking-gun. After Frolov and Mourik obtained and scrutinized the underlying data, Nature issued an expression of concern and then, in 2021, a full retraction. The retraction note records that the authors had applied data corrections not mentioned in the paper and that a recalibration shifted the key values by 8 percent. A Delft integrity investigation later found culpable negligence by two of the authors, while concluding that the case did not amount to a violation of scientific integrity.

That episode left a mark. In a 2021 Nature commentary, Frolov warned that a fraction of the Majorana field “is fooling itself.” A year later, Sankar Das Sarma, who co-authored the foundational theoretical proposal for these nanowire devices, published a Perspective acknowledging that the early sightings had been hasty and that real progress would require driving down disorder. By 2023, Microsoft had published its topological gap protocol paper, claiming several devices passed an automated test designed to remove human bias from the topological call. Legg’s first comment argued the protocol lacked consistent definitions and produced outcomes sensitive to arbitrary analysis choices.

Then came Majorana 1, and the gap between announcement and evidence became impossible to miss. Microsoft proclaimed the world’s first quantum processor with a topological core, eight topological qubits on a chip designed to scale to a million, built on a new material it branded a “topoconductor.” The concurrent peer-reviewed paper demonstrated something far narrower: single-shot parity readout, a measurement of whether a state holds an even or odd number of electrons. Nature‘s editors took the unusual step of attaching a note stating that the results do not represent evidence for Majorana zero modes in the devices. The working-qubit claim appeared in the press release, not the paper. Scott Aaronson flagged the discrepancy, and Winfried Hensinger of Sussex put it plainly: the peer-reviewed work contains no proof of topological qubits, but the press release speaks differently.

Majorana 2, announced this June, continued the rhythm. Microsoft reported an InAs-Pb tetron device with a characteristic parity switching time of around 20 seconds, with some instances reaching minute-scale, achieved by replacing aluminum with the higher-gap superconductor lead. Microsoft framed the result as a 1,000-fold improvement and said it now expects a scalable quantum computer by 2029, roughly halving its previous timeline. And once again, the paper presents the parity-readout measurement, not the coherent-control measurement that would actually demonstrate a qubit. Microsoft is right that the 2018 retraction involved an academic collaboration whose raw data it did not review before publication, and right that proving an emergent quasiparticle is inherently incremental.

But the through-line that bothers independent physicists is the order of operations. It’s now a pattern with Microsoft – the announcement arrives first, the evidence is contested afterward, and the most important claim never quite makes it into a peer-reviewed paper. Frolov’s framing after the latest exchange was that this is not one disputed result but “a series of papers that keep being challenged at the very basic level.”

What the skeptics actually say

The skeptics are not a fringe. When Science News and Scientific American canvassed physicists after Majorana 1, the modal response was unconvinced. The critics include people who helped build the field, and their objections are specific rather than rhetorical.

The oldest and deepest worry is disorder versus topology. Imperfections in these messy semiconductor-superconductor devices can manufacture states that mimic Majorana zero modes while being trivial Andreev bound states. Das Sarma, again, has been explicit that reducing disorder by at least another factor of two is the precondition for confidence. When one of the theory’s own architects is this cautious, that registers.

A second objection targets the gap protocol directly. Legg, with Jelena Klinovaja and Daniel Loss at Basel, showed that trivial mechanisms can pass Microsoft’s topological gap protocol. The company built the protocol precisely to remove subjective judgment from the topological call; the critique is that it carries a false-positive rate and depends on choices, like which magnetic-field range, bias range, and junction settings to use, that are not intrinsic to the device. Roman Lutchyn of Microsoft has rebutted that Legg surfaced one misleading instance out of roughly 700 and that the protocol holds up when parameters are chosen sensibly.

A third set of points is procedural. Microsoft’s published papers showed classification maps but not the raw transport data beneath them, so outside physicists could not see the gapless-looking regions for themselves until Legg extracted them from the archived dataset. And Microsoft’s reply, by leaning on the capacitance signal to argue a gap must have been present, runs into Legg’s sharpest procedural charge: that this reverses the evidence hierarchy, using the downstream result to assert the upstream prerequisite. Legg’s own characterization of the broader exercise, to The Register, was that it resembled finding an image of “Jesus in toast” after searching enough loaves.

Underneath all of it sits the simplest objection. To show you have a qubit, you have to measure it along two axes: read its state, the Z measurement, and demonstrate coherent control of a superposition, the X measurement. Microsoft’s papers, Majorana 2 included, show the Z measurement. The X data, when shown at all, has looked to several physicists like noise. At the March 2025 meeting of the American Physical Society, where Nayak and Microsoft were the most-searched topics among 14,000 attendees, Cornell’s Eun-Ah Kim said she had wanted the data to “come out screaming at me that there’s only two” states and did not see it. Klinovaja’s verdict was equally direct: “I don’t think the data are convincing.” Frolov, watching remotely, judged that the chip could not work given what was presented. Hensinger and others have described the topological approach as trailing decades behind rival methods.

Taking Microsoft’s Majorana qubit seriously

A fair survey cannot stop at the skeptics, and the defense is stronger than the most cynical takes allow. Three points carry real weight.

First, the engineering is advancing on a measurable axis that even critics mostly concede. The move from aluminum to lead produces a much larger superconducting gap, and the reported parity lifetimes have climbed by orders of magnitude across device generations. Whatever the underlying states are, the devices are getting cleaner and more stable. Kartiek Agarwal of Argonne, who has no Microsoft affiliation, called the nonlocality probing in the latest work “fantastic progress,” while still cautioning that calling the result a qubit was premature.

Second, Nayak’s account of how this kind of evidence accumulates deserves a hearing. Proving the existence of an emergent, non-local quasiparticle does not yield a single decisive photograph. His argument is that as Microsoft layers on more measurement types, the non-topological explanations have to become ever more contrived to fit the data. That is a legitimate description of how condensed-matter evidence often works, and it differs from moving goalposts. Steven Simon of Oxford, briefed on the results and with no stake in the company, offered the most quotable hedge in this whole debate: he would not bet his life that Microsoft is seeing what it thinks, “but it looks pretty good.” Anton Akhmerov at Delft, also independent, called some of Legg’s concerns “very much valid” while declining to treat them as fatal.

Third, Microsoft is one of two companies, with PsiQuantum, that DARPA advanced to the final phase of its US2QC program. That is not nothing. More than 50 technical evaluators examined Microsoft’s plans and proprietary data before the selection. But precision matters about what the vetting means. DARPA assessed whether the roadmap is credible enough to co-design and verify, not whether Majoranas exist or whether the qubit works. It is a serious program-management judgment, not an independent physics endorsement, and Microsoft’s framing of it as validation of the roadmap should be read against that boundary.

One pattern in the defenses is hard to unsee. The independent voices, Simon and Agarwal and Akhmerov and Das Sarma, cluster around “promising,” “fantastic progress,” “looks pretty good,” “the jury is out.” The unqualified, this-is-a-qubit claims come almost entirely from inside Microsoft. When you sort the commentary by who signs the paychecks, the confidence tracks the affiliation. That does not make Microsoft wrong. It tells you how much weight the broader physics community currently puts behind the strongest version of the claim, which is not much, and not yet.

Meanwhile, the rest of the field demonstrated logical qubits

The topological bet always rested on a specific promise. Conventional qubits are noisy, the argument went, and the error-correction overhead to make them useful would be brutal, on the order of millions of physical qubits for a handful of logical ones. A topologically protected qubit would bake error resistance into the hardware, so you would need far fewer of them. In 2015, that was a coherent strategic case.

It is a much weaker case in 2026, because the overhead problem the topological approach was meant to leap over is being solved the conventional way. In December 2024, Google’s Willow processor demonstrated, in a peer-reviewed Nature paper, error correction below the surface-code threshold: adding more physical qubits made the logical qubit better rather than worse, with the logical error rate suppressed by roughly half at each step up in code distance. Quantinuum’s Helios system reported dozens of error-corrected logical qubits at a physical-to-logical ratio near two-to-one, built on trapped-ion hardware with physical gate fidelities at the four-nines level. Harvard and QuEra ran 48 logical qubits on neutral atoms. IBM is building toward qLDPC-based machines that its own peer-reviewed work says need roughly ten times fewer qubits than the surface code.

Set against demonstrated logical qubits on three other platforms, Microsoft has demonstrated zero logical qubits on its own hardware, and its physical qubit is the one under dispute in Nature. The assessment that topological computing trails decades behind other approaches reads as harsh until you line up the peer-reviewed results, at which point it stops sounding like a cheap shot.

Microsoft’s counter is that it is playing a different game on a longer clock. If the physics holds, a real if, a topological qubit could sidestep the overhead problem rather than grinding through it, and a few thousand stable logical qubits would matter more than a noisy intermediate-scale machine. That is a defensible long-horizon bet. It is also, right now, a bet on a qubit that has not been shown to exist in a peer-reviewed venue, while rival platforms are publishing logical-qubit demonstrations on hardware that is, in some cases, commercially accessible.

What would change my mind

Here is what would move me, concretely:

• Full qubit-control data, both X and Z measurements showing coherent control rather than parity readout alone, published in a peer-reviewed venue and not a blog.
• Independent replication by a group with no Microsoft affiliation.
• A demonstration of braiding or fusion, or a two-qubit logical operation, that does what topological theory predicts and a trivial system cannot easily fake.
• Release of the complete raw transport data, so outside physicists can check the gap claims directly instead of reconstructing them from archives.
• A version of the topological gap protocol that survives the false-positive critique under systematic, adversarial testing.

None of these is unreasonable, and Microsoft may well deliver some of them. If it does, the skeptical consensus should soften, and I will say so in print.

I started by admitting I never trusted the Majorana story and that my gut is not evidence. Having done the work, my considered position is narrower than my instinct was. The materials science is real and improving. The existence of Majorana zero modes in these devices is contested by serious people for serious reasons. And the working topological qubit, the thing Microsoft has now announced in one form or another across the better part of a decade, has still not appeared in a peer-reviewed result. Eight years after the first “transistor moment,” the signal worth watching is simple: whether the next claim shows up in a journal instead of a keynote.