Stockholm startup FirstQFM uses AI foundation models to run quantum hardware

[LordJonathan92]

FirstQFM, a Stockholm company building AI foundation models designed to help operate quantum hardware. Rather than pitching quantum computers as rivals to classical ones, it treats them as specialised parts inside larger systems guided by AI. The company was founded in 2025 and works with Chalmers University of Technology. Adam Wesolowski, a research scientist at the company, is the main voice in the article

The AI models do several jobs. They predict calibration drift in quantum systems before errors build up, suggest corrective pulse sequences for superconducting and photonic hardware, and forecast which qubits will perform best and which pairs are causing trouble. They can also break big problems into smaller pieces suited to quantum processing. One result they highlight is a compiler that removes around 30 percent of circuit depth, which cuts the chance of errors

Wesolowski frames the question in an interesting way. Instead of asking whether a quantum computer beats a classical computer, he asks whether a system containing a quantum computer can beat the best system without one. He also suggests practical quantum advantage might show up first on an energy bill or a hardware budget, rather than as a faster time on a stopwatch

The company raised €1.2 million in pre-seed funding in October 2025, led by BSV Ventures with Almi Invest, Further than Capital and Luminar Ventures also involved. That is small by quantum standards, but software companies need much less capital than hardware builders. Using AI to squeeze more out of existing machines seems like a sensible niche

Calibration is one of the most tedious parts of running a quantum computer, so automating it could save a lot of time. Is AI the thing that finally makes quantum hardware practical, or just another layer of hype on top of hype?

Learned Software Now Operates Quantum Hardware, Says FirstQFMquantumzeitgeist.com

[Amy80]

Cutting circuit depth by 30 percent is a big deal on noisy hardware. Every gate you remove is one less chance for an error. That improvement could matter more than adding a few extra qubits

[Josh93]

Calibration drift is a nightmare on current machines. Qubits change behaviour over hours, and recalibrating takes the machine offline. Predicting drift before it causes errors would keep systems running longer. That is a practical win that operators would pay for straight away

It is less glamorous than advantage claims. But it is the kind of thing that makes a technology usable. Boring improvements add up

[Wolfhound44]

AI plus quantum in one pitch is peak 2026 buzzword bingo. Investors must have loved it

[Doomsday15]

Sweden has a strong quantum scene thanks to Chalmers and the Wallenberg funding. Nice to see start ups coming out of it

A pre-seed round of €1.2 million is modest, but software needs less. The next round will tell us more