Photo: Google Gemini · AI-generated
At Build 2026, Microsoft introduced Majorana 2, the successor to last year’s Majorana 1 topological quantum chip. The headline number is striking: qubit lifetimes now exceed 20 seconds, up from 1–12 milliseconds in the previous generation — a roughly 1,000x improvement in stability. The change came from swapping the original aluminum-based material stack for one built on lead, a material already familiar from radiation shielding in hospitals and industrial settings.
Two details make this announcement different from the usual quantum computing news cycle.
AI is now part of the materials science. Microsoft says the new material stack was designed with help from Microsoft Discovery, its agentic AI platform for research workflows — automating measurements, optimising fabrication steps, and flagging defects. Whatever one thinks about the quantum claims themselves, this is a clear example of AI accelerating the pace of physical science, not just software.
External scrutiny is built in. Microsoft’s topological qubit programme is part of DARPA’s Quantum Benchmarking Initiative, which independently evaluates progress claims rather than relying on company-published results alone. Microsoft is one of two companies to reach the programme’s final phase. Based on these results, Microsoft has shortened its own roadmap, now targeting a scalable, utility-class quantum computer by 2029 — half the time previously projected.
The honest caveats. This field has a history of premature claims — including a 2018 paper Microsoft had to retract. The Majorana 2 results are published as a preprint; peer review and independent reproduction across multiple devices are still pending. Outside physicists have welcomed the data while noting it doesn’t yet demonstrate repeatable results at scale. The 2029 target should be read as an internal goal under external evaluation, not a delivered product.
Why this matters for us. Most of the problems we solve for clients in precision machinery, construction, and intralogistics are, at their core, hard optimisation problems: scheduling, routing, resource allocation, materials flow. These are exactly the class of problems where quantum computing is expected to eventually offer an advantage over classical approaches — but “eventually” is the operative word. Nothing changes in how we build software today.
What’s worth tracking is the trajectory, not the announcement. Quantum computing is following a path that looks familiar from AI’s own history: years of incremental, externally-scrutinised progress before a threshold is crossed and the technology becomes operationally relevant. The teams that benefit will be the ones who understood the fundamentals early — not the ones who reacted to the first headline.
For now, our focus stays where the near-term value is: on-device AI, agentic workflows, and cloud-edge architectures that are shipping today. Quantum is on the list of things we read carefully. It’s not yet on the roadmap.