Microsoft has announced two significant advances in computing at its annual developer conference, Microsoft Build, in San Francisco: a new quantum computing chip it claims is 1,000 times more reliable than its predecessor, and a suite of in-house artificial intelligence models designed to reduce the company's dependence on OpenAI.
The new chip, called Majorana 2, represents a major step in Microsoft's long-running effort to build a fault-tolerant quantum computer. At the heart of quantum computing are qubits — the basic units of quantum information — which are notoriously unstable: even minor vibrations or temperature changes can cause errors. Microsoft says the qubits on Majorana 2 survive for an average of 20 seconds, compared to mere milliseconds on the original Majorana chip released last year, comparing the improvement to upgrading from a phone that needs daily charging to one that lasts several years. The key material change was replacing aluminium with lead as the superconductor, a switch developed with the help of Microsoft's own AI tools for materials science. The company now says it expects to have commercially useful quantum machines by 2029 — the same target year recently announced by rival IBM, which has pledged $10 billion toward the effort. Google, Amazon, and several Chinese programmes are also competing in what has become a global race to develop quantum systems capable of solving problems in medicine, chemistry, and cybersecurity that would take conventional computers thousands of years.
Microsoft's quantum approach relies on so-called topological qubits, based on exotic quasiparticles known as Majoranas — first theorised in the 1930s by Italian physicist Ettore Majorana but not demonstrated to exist until Microsoft claimed to have observed them. That claim has drawn persistent scepticism from independent physicists, who argue the company has not released sufficient data for its results to be independently verified. Henry Legg, a lecturer in quantum physics at the University of St Andrews in Scotland, noted that switching to lead as a material does not address the underlying scientific requirement that results be reproducible. Microsoft says trade secrets prevent full public disclosure, but it has shared data confidentially with DARPA — the US Defense Advanced Research Projects Agency — which is independently assessing several competing quantum approaches. Alongside its hardware announcement, Microsoft also spun out a separate company to manufacture quantum chips for external customers, backed by the US government.
On the artificial intelligence front, Microsoft unveiled MAI-Thinking-1, its first in-house "reasoning" model — AI systems that break problems into steps before responding, similar to offerings from OpenAI, Google, and Anthropic. The company says the model was built from scratch without relying on rivals' outputs to train it, though it arrives roughly 18 months behind the field's leaders. Microsoft also introduced Microsoft Scout, an "always-on" autonomous assistant for scheduling and communications, along with new models for image generation, audio transcription, and coding, as well as an Nvidia-powered mini-PC capable of running AI offline and a range of Android-based home devices controlled by voice.
The dual announcements reflect Microsoft's broader strategic ambition: to build technological self-sufficiency across both AI and quantum computing. Chief executive Satya Nadella has repeatedly warned against over-dependence on any single partner — a lesson he draws from IBM's experience of backing Microsoft's rise in the 1980s only to be eclipsed by it. If the company's 2029 quantum timeline holds, it could open the door to accelerating solutions to long-intractable problems, from eliminating microplastics to developing better agricultural fertilisers. For now, however, the chip has just 12 qubits, while a commercially viable machine would require millions — leaving substantial engineering challenges ahead.