Microsoft's Majorana 2 Chip Is 1,000 Times More Reliable and Halves Quantum Roadmap

Microsoft just unveiled Majorana 2, its second-generation topological quantum chip, and the headline number is hard to ignore: qubits that are 1,000 times more reliable than those in its predecessor, Majorana 1, which launched just over a year ago. That jump is big enough that Microsoft is now cutting its quantum computing roadmap in half, targeting a commercially viable, scalable quantum computer by 2029.

The other half of the story is how they got there. The chip was developed with significant help from Microsoft Discovery, the company's agentic AI platform for scientific R&D, which is also launching into general availability alongside this announcement. It's a rare case where an AI system directly contributed to building the hardware that may one day run more powerful AI.

A new state of matter as the foundation

To understand why Majorana 2 is different, you need to understand what topological qubits actually are. Most quantum computers, like those from IBM or Google, store quantum information in the fragile physical state of a single particle. Noise from the environment constantly disrupts that state, causing errors.

Microsoft's topological approach instead stores quantum information in the physical shape of a material rather than in the state of a single particle. Topological qubits theoretically offer inherent protection against environmental noise, which is the primary source of errors in competing approaches. Think of it like encoding a message in the knot of a rope rather than in the position of a single thread: the knot survives minor disturbances that would destroy the thread's position.

The chip's qubits are built from structures called tetrons, which are pairs of superconducting nanowires designed to host Majorana zero modes at their endpoints. Information is stored in the parity of electrons occupying the topoconductor wire, and quantum operations are performed by measuring that parity rather than by directly controlling the quantum state. This measurement-based approach produces digital outputs and, in theory, reduces sensitivity to the analog errors that affect superconducting gate-based qubits.

The materials swap that changed everything

To create Majorana 2, the Microsoft Quantum team improved Majorana 1's material stack to create a more stable topological phase. Majorana 2 replaces Majorana 1's superconductor, aluminum, with lead, and also updates the semiconductor active region to a combination of indium arsenide and indium arsenide antimonide. This change in materials results in significant increases in performance, reflected in the improved robustness of the topological phase.

The key technical improvements break down as follows:

• Superconducting gap: Lead provides a superconducting gap of ~1,300 microelectronvolts vs. ~300 for aluminum, making it much harder for environmental disturbances to destabilize the topological phase.
• Topological gap doubled: The topological gap, which protects the topological qubits from environmental noise and errors, is more than double that of the previous quantum processor.
• Qubit lifetime: Majorana 2 is 1,000 times more reliable than its predecessor, Majorana 1, as its qubits can survive for an average of 20 seconds, rather than milliseconds, with some instances lasting as long as one minute.
• Qubit count: Where the first iteration of the Majorana chip only scaled to eight qubits, Majorana 2 has scaled to 12 qubits.
• Operation speed: Gate operations run on the microsecond scale, while qubit lifetimes are measured in seconds, giving a large margin for error correction.

Where AI actually helped build the chip

This is where the story gets genuinely novel. By applying recent advances in agentic AI specially designed to speed the scientific process and accelerate collaboration, Microsoft's quantum team is overcoming key barriers in reliability, speed and size that have limited the application of quantum computing to real-life scenarios.

The quantum team used Microsoft Discovery's AI agents in four concrete ways during development:

1. Automating measurements: Setting up a topological state requires tuning hundreds of parameters, and measurement cycles previously took weeks per iteration. AI agents cut this cycle time by orders of magnitude by running voltage adjustments in parallel, something no human operator can do linearly.
2. Synthesizing two decades of data: The project has accumulated nearly 20 years of experimental data across incompatible formats and research silos. The Microsoft Quantum team has overcome many of these challenges with the assistance of AI. Agents were able to surface correlations invisible to any single researcher.
3. Optimizing fabrication: Critical parts of the chip are designed atom by atom. AI simulations identified the right concentrations of dopant materials before physical experiments, reducing costly trial-and-error cycles.
4. Catching noise sources: An AI agent combining physics, device, and institutional knowledge detected an uncalibrated temperature sensor in the fabrication process that was introducing noise no human had spotted.

"Agentic AI has permeated almost everything we do," said Chetan Nayak, Microsoft technical fellow. "The agents can really accelerate things as much or as little as you want. It can be as little as pulling information together and summarizing it, or it can go further down the road of synthesizing it more or generating an interesting hypothesis."

Microsoft Discovery goes GA

Microsoft Discovery is now generally available. The platform for Frontier R&D lets customers deploy AI agent teams, guided by human expertise, to speed up scientific discovery. The platform includes a Discovery Engine for research and reasoning workflows, plus enterprise-grade security and governance controls.

The new Microsoft Discovery app provides a local version of the platform's core capabilities that individuals can download for free and use with a GitHub Copilot account. Early customers span life sciences, chemicals and materials, energy, and manufacturing. In one benchmark case, researchers at Microsoft used Microsoft Discovery to detect a novel coolant prototype with promising properties for immersion cooling in data centers in under 200 hours, rather than months or years with traditional methods.

The skeptics haven't gone away

Microsoft's topological quantum program has a complicated history with the scientific community, and Majorana 2 has not resolved that tension. Microsoft's topological approach has been the most controversial in the field. The company's 2018 claim to have observed Majorana zero modes was retracted after independent scrutiny. Majorana 1, introduced in 2025, re-established credibility with peer-reviewed results.

With Majorana 2, the criticism continues. Henry Legg, a theoretical physicist at the University of St Andrews