Microsoft gives DARPA on-site access to its Majorana 2 quantum chip

Share
Microsoft gives DARPA on-site access to its Majorana 2 quantum chip

Microsoft's quantum bet picked up a federal referee on Tuesday, and the referee's job is to try to prove the company wrong.

Microsoft said it will give DARPA direct, on-site access to its latest topological quantum hardware — including qubits built on its Majorana 2 chip — at a new 15,000-square-foot quantum research center in Maryland. The center sits in the University of Maryland's Discovery District near Washington, inside a building the state acquired and renovated with support from Governor Wes Moore's Capital of Quantum Initiative; Microsoft says it designed the space with UMD and the initiative over the past year. It holds dedicated areas for Microsoft's own research, room for DARPA's topological system and future prototypes, labs for partner companies, and a training laboratory. The building is the easy part. The arrangement inside it is the news.

DARPA's access is the substance. Microsoft is one of the first companies to reach the final stage of the agency's Underexplored Systems for Utility-Scale Quantum Computing program, part of its broader Quantum Benchmarking Initiative, which is trying to establish whether any competing approach can produce a commercially useful machine by 2033. The evaluation pulls in the Air Force Research Laboratory, Johns Hopkins University's Applied Physics Laboratory and national laboratories including Los Alamos, Oak Ridge, Lawrence Berkeley and Lawrence Livermore. DARPA's test and evaluation team works across the whole stack — hardware, controls, software, applications — which is precisely the kind of scrutiny a company cannot supply for itself.

That matters because Microsoft's claims have been the contested part of its quantum story for years. The company says Majorana 2's qubits are 1,000 times more reliable than those in its previous processor, and that the improvement halved its timeline to a scalable quantum computer — now expected in 2029. Handing the hardware to an agency whose mandate is to falsify claims is the strongest evidence Microsoft can offer without publishing the underlying data. The quieter claim in Tuesday's post is about how the work gets done: Microsoft says AI has compressed its own design cycle so far that architectures which once took a year or more to analyze are now compared in about a week. Faster exploration is a real advantage, and it also raises the cost of being wrong, because more candidate designs get committed to than the experimental infrastructure can absorb.

The center also anchors a workforce play. A hardware makerspace with AMD, Bluefors, Intel, IQM, Fermilab, Riverlane and Quantum Motion is meant to put students, researchers and career-changers on real equipment rather than theory, and Microsoft is starting an annual workshop series on measurement-based quantum computing alongside its first funded quantum postdoctoral positions at UMD.


An AI agent sent to hunt open-source bounties for three days earned $0, and the write-up is the most useful field report on agent economics published this month. The agent did not fail at the work so much as find the market structurally closed: funded issues sit behind maintainer allowlists that are invisible from outside, the queue of solvers has exploded while the number of funded issues has not, payouts run roughly $50 to $250, and one contributor's first fix landed in a codebase whose test suite never exercised the broken case. The supply side is now largely agents competing with other agents, which is why maintainers — who did not sign up to referee a dozen bots — stop answering.

The instructive part is what the agent did with the remaining time: it audited its own orchestrator and fixed the failures it found, reporting numbers that went against its own interest when its scanner turned out to be reporting partial results as complete. An audit you can trust is worth more than a $50 bounty, which is a strange conclusion to reach about software that earned nothing.

What to watch: whether DARPA's evaluation produces anything public before 2033, and whether Microsoft's AI-compressed design cycle shows up as hardware rather than as a faster slide deck.

If an agency can test a company's quantum claims on site, should that be the standard for every frontier hardware promise — or does it only work when the buyer is the government? Tell us in the comments.

Sources: Reuters · Microsoft Quantum — Microsoft's Quantum Research Center Opens in Maryland · The Quantum Insider · HackerNoon