News

Microsoft's Quantum Gamble: The Physics It Still Must Prove

Microsoft claims a 1,000-fold leap in quantum reliability with Majorana 2, but leading physicists say the company has never demonstrated the topological behavior its entire architecture depends on.

By Alex ChenAI Reporter5 min read

Ettore Majorana proposed his particle in 1937 and vanished off a boat in the Tyrrhenian Sea the following year, leaving physics one of its more durable ghost stories. Nearly ninety years later, Microsoft is betting two decades of research that the condensed-matter analogue of his idea can be turned into a computer. The problem is the same one that has dogged the project from the start: a significant part of the physics community is not convinced Microsoft has ever found the thing it says it has built its machine around.

The 1,000-fold claim and what it rests on

In June 2026 Microsoft announced Majorana 2, which it describes as a roughly 1,000-fold improvement in qubit reliability over last year's Majorana 1. The headline number is parity lifetime — how long the qubit holds its quantum state before an unwanted change flips it. On Majorana 1, built on an aluminum superconductor, that lifetime ran between one and 12 milliseconds. On Majorana 2 it averages 20 seconds, with some instances near a minute, according to the company. Operations, by contrast, happen on microsecond timescales — so on paper there is enormous headroom between how long the qubit lives and how long a gate takes.

The engineering change behind that jump is a materials swap: out with aluminum, in with lead as the superconductor, paired with an indium-arsenide active region. Microsoft says the topological gap — the energy barrier that is supposed to insulate the qubit from environmental noise — is more than double that of the previous processor. Chetan Nayak, the Microsoft technical fellow who runs the program, put it plainly: "where are we relative to last year? We're 1,000 times better."

The company also credits agentic AI with the pace, saying its systems found correlations across nearly two decades of quantum research data no single researcher could hold in their head, and drove measurement automation, fabrication analysis and materials optimization. On the strength of all this, Microsoft has pulled its target for a scalable, commercially useful quantum computer forward to 2029.

There is institutional backing to point to as well. DARPA advanced Microsoft as one of only two companies to the final phase of its US2QC program, its effort to evaluate quantum systems for utility-scale computing — not a body that hands out placement lightly.

The bet underneath the numbers

Everything above assumes the qubit is topological, and that assumption is the whole game.

Conventional qubits — the transmons at Google and IBM, the trapped ions elsewhere — are fragile. Noise corrupts them, so designers spend enormous physical overhead building one reliable logical qubit out of many noisy ones. Microsoft's wager is to skip that tax. It wants to encode information in Majorana zero modes, storing it non-locally across a device so that no single local disturbance can read or wreck it. If competitors are engineering ever-better suspension to protect fragile cargo, Microsoft is trying to manufacture cargo that is intrinsically hard to break.

That is the theoretical advantage. It has not been demonstrated at scale. And it depends entirely on the qubit actually being topological rather than merely behaving, in a noisy measurement, as if it might be.

Why the skepticism outweighs the numbers

This is where the story stops being about lifetimes. Outside experts quoted by Scientific American say Microsoft has never conclusively demonstrated the topological behavior that its entire architecture rests on — the publication frames it as a chip that "doesn't even work and never has."

The track record is what gives that charge weight. In 2021 Microsoft retracted a high-profile 2018 Nature paper reporting Majorana zero-mode evidence, after outside experts showed the data could have come from material imperfections rather than a genuine topological qubit. Physicists have raised comparable objections to several publications since, including last year's Majorana 1 announcement. Henry Legg, one of the researchers quoted challenging the company's work, told Scientific American that "if this was from any other group or Ph.D. student, it would never make it through peer review." The skepticism as reported traces to that single publication and the critics it assembled; it is not, on the current record, a broad chorus with independent write-ups.

Read carefully, the dispute is not about whether 20 seconds beats 12 milliseconds. Twenty seconds is a real, impressive stability figure regardless of its origin. The argument is about what is producing the signal. A parity lifetime that long is only evidence for topological protection if the state being measured is topological in the first place — and that is the exact link critics say has never been established. A larger topological gap is a stronger claim to be right about, not proof of it.

When a trillion-dollar company becomes the laboratory

The deeper reason this matters is structural. Frontier physics has always run on independent replication and adversarial peer review: someone else builds your device, measures it their way, and either confirms you or doesn't. Microsoft controls the fabrication, the measurement apparatus, the analysis pipeline and the announcement calendar, and it can sustain a multi-decade bet no university group can match financially. That is a genuine strength — the aluminum-to-lead transition reportedly took years of failed attempts to get right, the kind of patience only a balance sheet like Microsoft's underwrites.

It is also precisely why the physics community's usual correction mechanism struggles here. A press release can travel a great deal faster than a reproduced result, and when the same institution owns the claim and the only apparatus capable of testing it, the outside world is left arguing about published figures rather than rerunning the experiment.

The thing to watch is narrow and specific: an independent group, working on hardware Microsoft did not build, demonstrating the topological signature in a measurement that survives the scrutiny the 2018 paper failed. That would settle the physics. Absent it, each new lifetime record extends an impressive engineering result on a foundation the field has not agreed is there. Microsoft's 2029 date is a bet that the foundation will hold long enough to build on. Everyone else is still waiting to see the floor.

About the author
Alex Chen

Alex Chen covers models, MLOps and the engineering reality behind the demos. If it ships to production, Alex wants to know how it survives contact with real traffic.

Was this helpful?

Discussion

Be the first to comment

Join the conversation. Sign in to comment, reply, and vote.

Loading discussion…

Intelligence, in your inbox

A considered briefing on AI, Quantum, Robotics, Space, Longevity & Energy. No noise.

We use your email address solely to send you our newsletter or to update you about your account. You can withdraw your consent at any time by clicking unsubscribe in any email footer. Read our Privacy Policy for details.

More Intelligence

Futures

At The Frontier: Why AI Interpretability Is the Advantage

OpenAI's chief scientist Jakub Pachocki argues that chain-of-thought monitoring, the field's primary bet on interpretability, is degrading as reasoning models become more capable. The systems can now find zero-day vulnerabilities, manipulate their own reasoning, and operate in environments beyond their training distribution. Transparency must be built into models during training, not bolted on after, and he calls for voluntary slowdowns and international coordination.

Alex Chen
News

Rice Roboticists Fold Perception Directly Into Motion Planning

Rice researchers have integrated perception scoring directly into motion planning algorithms for high-degree-of-freedom robots, rather than treating visual perception as an afterthought. Their GPU-parallelized approach, called PS-PRM, improved object detection rates above 70% and tracking rates around 90% compared to classical baselines that often fell below 50%.

Sophia Patel