Quantum Computing, Computer Science
Microsoft Majorana 2: A 1,000x More Reliable Quantum Chip Powered by Agentic AI
Microsoft has unveiled Majorana 2, a new topological quantum chip whose qubits hold their information for roughly 20 seconds on average — a roughly thousand-fold improvement over the company's previous device. It is the most credible demonstration yet of Microsoft's bet that the way to scale quantum computing is not to bolt error correction on top of fragile bits, but to build error-resistant qubits directly into the physics of the hardware.
The qubit fragility problem
Every quantum computer faces the same obstacle: a qubit is exquisitely sensitive. Interaction with stray heat, magnetic fields, or even a single photon can flip its state and corrupt the calculation. Conventional quantum processors fight this with layers of error-correction software, which consume most of the available qubits just to keep a handful of "logical" qubits stable. Microsoft's approach is different. Its topological qubits encode information in the global arrangement of exotic quasiparticles called Majorana modes, spread across a material. Because the information is stored non-locally, a local disturbance cannot easily read or destroy it — the hardware itself becomes inherently tolerant of noise.
What Majorana 2 changed
Majorana 2 introduces an entirely new materials stack. Microsoft replaced the aluminum used in the first-generation Majorana 1 chip with lead, and paired it with an indium-arsenide and indium-arsenide-antimonide active region. In these devices the team recorded a mean "parity" lifetime of about 20 seconds for the four-qubit array — and individual runs that held stable for close to a full minute. For comparison, the prior device's qubits lasted only milliseconds before parity flips scrambled the stored information. A 20-second coherence window is not just longer; it is long enough to run real operations before the qubit loses its state.
AI redesigned the chip
The materials breakthrough was not reached by trial and error alone. Microsoft used its Microsoft Discovery platform — agentic AI tools designed specifically for materials science — to analyse experimental data and optimise the chip's fabrication. The AI system explored combinations of material interfaces and processing steps that would be impractical for human teams to test by hand, accelerating a path that normally takes years. The result is a chip whose hardware is, in Microsoft's accounting, 1,000 times more reliable than its predecessor.
The roadmap shrinks
With Majorana 2, Microsoft has halved the timeline it set in February 2025. The company now expects a scalable quantum computer — one large enough to solve commercially useful problems — by 2029, and a system relevant to real-world applications within this decade. Outside physicists remain appropriately cautious: a four-qubit array is a proof of concept, not a competitor to the largest machines, and scaling topological qubits to thousands of units has not yet been demonstrated. But the jump from milliseconds to tens of seconds is the kind of step-change that changes whether a roadmap feels plausible.
Knowledge takeaway: Majorana 2 is Microsoft's topological quantum chip with qubits lasting ~20 seconds on average — about 1,000 times more reliable than Majorana 1; its reliability comes from an inherent, non-local way of storing information rather than from software error correction; agentic AI redesigned the lead-based materials stack, and Microsoft now targets a scalable quantum computer by 2029.