Microsoft Built a Better Quantum Chip. Its Own AI Found the Bugs.

Microsoft announced something remarkable at its Build conference Tuesday: a quantum chip whose qubits can maintain their quantum state for 20 seconds, occasionally for as long as a minute. For context, common superconducting qubits the kind Google and IBM use measure their lifetimes in microseconds. Microsoft says its new Majorana 2 chip is 1,000 times more reliable than its predecessor, according to the company's announcement. The company now expects to have a scalable quantum computer by 2029, cutting its previous timeline in half.

Every outlet covering the announcement led with those numbers. They are the press-release figures. They are not the story.

Buried in Microsoft's own announcement is a sentence that should interest every VC funding quantum hardware and every founder trying to build it: "scientists and engineers have been using the agentic AI capabilities in Microsoft Discovery to manage workflows, automate measurements, optimize fabrication, pinpoint previously unnoticed flaws and propose new solutions," according to the company blog.

The 1,000x reliability gain did not come from physicists solving the problem through conventional iteration. It came because Microsoft pointed its own AI at its own quantum program and the AI found problems the team had not seen. One of those problems was significant enough that fixing it produced a generational leap in device quality. The machine-debugger, not the materials scientists, found the bug.

Zulfi Alam, corporate vice president for quantum at Microsoft, described the process: "Finding the exact recipe, the right amount to put to get the desired energy structure, requires a lot of experimentation in the old world order. In the new world order, through simulations, you can see where the highly probable target is. And then with that knowledge, you ideally only have to experiment once," according to the company blog.

That is the replicable claim. Not that Microsoft has a better qubit. That an AI system can narrow the search space in materials science enough to turn years of trial and error into targeted experiments. If that holds across other hardware programs, the implications extend well beyond quantum computing.

Chetan Nayak, Microsoft's technical fellow and the public face of its quantum effort, was more blunt in the press materials: "Agentic AI has permeated almost everything we do," according to the announcement. He was describing a workflow in which AI agents run measurement cycles automatically, build three-dimensional maps of qubit conditions that no individual scientist could construct manually, and flag anomalies such as an uncalibrated temperature sensor quietly corrupting data.

The company also announced that Microsoft Discovery, the platform underlying the quantum team's AI tools, is now generally available for any organization, according to a separate Azure blog post. A preview version of a Discovery app can be downloaded free and run locally with a GitHub Copilot account. The quantum program is the reference case. Microsoft is now selling the playbook.

The context that Microsoft's announcement omits is the one it cannot yet escape: the physics community's prolonged skepticism about its topological qubit claims.

At the American Physical Society's Global Physics Summit in March 2025, Microsoft researcher Chetan Nayak presented evidence for the company's first-generation Majorana 1 chip. Physicists in attendance were not impressed. The key data plot looked like random jitter, not a qubit signal. Henry Legg, a physicist at the University of St Andrews, gave a talk the day before Nayak's in which he argued that Microsoft's topological gap protocol the method used to demonstrate that a device is actually topological produces inconsistent results depending on the parameter ranges chosen, according to Physics, the journal of the American Physical Society. "Any company claiming to have a topological qubit in 2025 is essentially selling a fairytale," Legg said at the time, according to Science News. Eun-Ah Kim at Cornell and Javad Shabani at NYU expressed similar doubts, Science News reported. Jason Alicea, a theorist at Caltech, said the bar for claiming discovery of a topological qubit should be much higher.

Microsoft disputes those criticisms. Roman Lutchyn, another Microsoft researcher, responded at the summit that Legg's objections were incorrect and that the company stands behind its results.

Whether Majorana 2's new data resolves those objections is not yet known. The technical paper Microsoft cites for the 20-second parity lifetime result has not undergone independent peer review. The topological gap protocol concerns Legg raised about the measurement methodology have not been addressed in any published response. Microsoft says it is still characterizing the new design.

Microsoft also changed the underlying material. Majorana 1 used aluminum nanowires. Majorana 2 uses lead, which provides better radiation shielding for fragile qubits, according to the company announcement. The switch to a lead-antimony stack is described as the result of the AI-assisted search process. Whether lead introduces new fabrication challenges at scale is an open question the company has not addressed publicly.

The 2029 timeline is a company claim. No independent quantum computing researcher was willing to corroborate it publicly. Microsoft has revised quantum timelines before.

What is not in dispute is that the AI-assisted debugging workflow is real, that it produced a measurable result, and that Microsoft is now productizing the platform that generated it. The question for the industry is whether that workflow is proprietary to Microsoft or whether it can be replicated elsewhere. If the answer is the latter, the AI-debugger model becomes the more significant announcement.