Information in this post reflects publicly available sources as of June 16, 2026.
On June 2, at the close of its Build 2026 conference, Microsoft held up a small gold-and-blue chip and made a claim that would be hard to overstate. Its quantum bits, the company said, had become 1,000 times more reliable than the previous generation. One Microsoft technical fellow put it plainly from the stage: "We're 1,000 times better." The room understood the subtext. If true, this was the moment topological quantum computing stopped being a bet and started being a machine.
Within hours, a different sentence was circulating among physicists who study exactly this technology: it doesn't work, and it never has.
Both of those reactions came from serious people looking at the same preprint. That gap is the story.
The chip, and the name
The chip is called Majorana 2, after the Majorana quasiparticle, a theoretical object that Microsoft's whole approach depends on. The naming is fitting in a way the company may not have intended. Ettore Majorana was the Italian physicist who predicted such a particle in the 1930s and then, in 1938, boarded a ship and vanished without a trace. Nobody ever proved what happened to him. The particle named after him has a similar quality: long predicted, endlessly searched for, never quite pinned down beyond dispute.
That is the knot at the center of this announcement. Microsoft is not claiming a faster version of a thing everyone agrees exists. It is claiming progress on a thing whose existence is itself contested.
What Microsoft actually announced
Strip away the framing and there is a concrete, checkable engineering change underneath. The team replaced the aluminum superconductor used in last year's Majorana 1 with lead. Lead is an unusual choice; one of the lead researchers admitted on the call that it "sounds like a crazy idea." The reasoning is that lead better shields the fragile qubit from environmental interference, including stray particles like cosmic rays.
The headline number flows from that change. A property called parity lifetime, which tracks how long the qubit holds its quantum state before random noise spoils it, reportedly jumped from milliseconds to around 20 seconds, with some instances reaching a full minute. One Microsoft framing compared it to a phone battery that, instead of dying in a day, lasts nearly three years.
On the strength of that result, Microsoft moved up its public roadmap. The company now says it aims to demonstrate "scalable, practical quantum computing" by 2029, pulling the target in from earlier estimates. It also leaned on a second story: the chip was developed with help from Microsoft Discovery, its agentic AI platform, which the team used to automate measurements, optimize fabrication, and surface design flaws.
Those are the confirmed claims. Every one of them traces back to Microsoft's own announcement and a preprint the company posted. None has been peer-reviewed.
Why physicists are not applauding
The skepticism here is specific, and it comes from named researchers with standing in the field. It is not reflexive contrarianism.
The core objection is statistical. Henry Legg, a physicist at the University of St. Andrews, points out that the striking results in the preprint appear to come from a handful of instances on a single device. His concern is the oldest one in experimental physics: a single device can show something remarkable once and never again, because the effect was an artifact rather than a real phenomenon. What convinces a community is reproduction across many devices, and that is not what the paper shows.
His verdict on the standard is blunt. If this work came from any other group or a graduate student, he argues, it would not survive peer review.
Sergey Frolov, a quantum researcher at the University of Pittsburgh, adds a second layer. He notes that Microsoft's comparable preprint from the prior summer has stayed unpublished, which he reads as a sign that top journals likely rejected it. His characterization of the field's mood is unsparing: among quantum specialists, he says, a Microsoft quantum claim now tends to draw a chuckle or a raised eyebrow rather than excitement.
The reason this lands harder than ordinary scientific caution is history.
In 2021 Microsoft retracted a paper in Nature after outside experts showed the data could have come from ordinary material imperfections rather than a genuine topological qubit. Similar disputes trailed the Majorana 1 announcement the year before this one. So when critics say the new preprint "builds on controversial results," they mean it literally. The foundation under the 1,000x number is the same foundation they have been questioning for years.
Why Microsoft is unmoved
The company's posture is worth taking seriously rather than dismissing, because it reframes what counts as proof.
Microsoft's executives argue that demanding textbook-clean particle physics is the wrong bar for an engineering program. The Quantum team's executive vice president invoked a famous comparison more than once: Bell Labs did not have to prove the electron existed in order to invent the transistor. The implication is that a working, scalable device is its own validation, and that the academic dispute over the cleanliness of the underlying measurement is a separate question from whether the machine ultimately computes.
There is also a reframing buried in that same quote. Microsoft says it did have to prove the Majorana physics was real, and believes it has. So the company is not sidestepping the question so much as insisting it has already answered it, on terms the academic critics reject.
Whether that confidence is vision or spin depends entirely on data that does not yet exist in public, peer-reviewed form. That is the honest center of this story.
Where things stand
As of now, a few things are settled and a great deal is not.
Settled: Microsoft changed a real material, posted a real preprint, reported a large jump in a measurable quantity, and moved its public roadmap to 2029. The agentic-AI fabrication story is also genuinely interesting on its own terms, independent of whether the qubit claim holds.
Unsettled: whether the measured 20-second lifetime reflects a true topological qubit or an artifact of one device on a good day. That question will not be answered by a press briefing or a roadmap. It will be answered, if at all, by the same effect showing up again and again across many devices, written up in a paper that survives peer review.
The thing to watch is not Microsoft's next announcement. It is whether independent groups, or Microsoft's own follow-up work, can reproduce the result in a form the physics community accepts. A second device showing the same lifetime would change the conversation overnight. Continued silence would tell its own story.
Summary
The deepest fact about the Majorana 2 announcement is that the disagreement is not about the size of the number. Both sides accept that Microsoft reported a 1,000x improvement. The fight is about whether the thing being improved exists in the first place. That is an unusual place for a multitrillion-dollar product roadmap to rest, and it is why a chip named after a physicist who vanished is, for now, an apt metaphor for its own central claim. The particle was predicted long ago, has been hunted for decades, and still slips away whenever someone tries to hold it up to the light and ask others to look.
This is a standalone post. Future posts covering AI and computing news and releases will appear under the In Focus label.