Quint Computing raises €9.1M as IBM links two 120-qubit processors

The newest money in European quantum hardware landed on 3 November. Quint Computing, a German startup founded in October 2025, raised €9.1 million in an initial round led by Vsquared, according to Tech.eu. Quantonation, Nucleus Capital, Faber, Vanagon Ventures and ZOHO.VC joined. Angel investors included Prof. Mikko Möttönen, founder of IQM and QMill. The company also holds a €2.2 million Pathfinder grant from the European Innovation Council, Tech.eu reported. That is public money for a company barely a year old. It shows how early the bets in this sector still are.

Quint's pitch is hardware-level error reduction. It is building a superconducting chip architecture meant to cut error rates before software correction gets involved. That problem has kept quantum machines out of industrial use.

The founders are quantum physicists Prof. Michael Hartmann and Prof. Christopher Eichler, plus economist Dr Jona Stinner. The team came out of Friedrich-Alexander-Universität Erlangen-Nürnberg and sits inside the Munich Quantum Valley ecosystem, per Tech.eu. Their bet is that error rates have to fall inside the chip itself, not only in the classical software stacked on top of it. They also bet that a superconducting design tuned for that job can reach industrial users sooner than architectures that lean on correction layers added later. Whether a seed-stage team can carry that from design to fabrication is the open question the round does not answer.

Two processors, one meter, four couplers

On 2 November, Semiconductor Engineering flagged an IBM Research paper, "Meter-scale quantum interconnects between large quantum processors," posted to arXiv as 2610.29986. The abstract describes four L-couplers operating between two 120-qubit IBM Nighthawk R1 superconducting processors. Modular designs need exactly this: long-distance couplers that hold coherence across meter-scale distances. They must also scale to multiple connections between any two modules.

Why does one meter matter? A single dilution refrigerator has a finite wiring budget. Error-corrected machines need many more qubits than any one chip can hold. If you can stitch modules together without collapsing the quantum state, you can grow a machine by adding modules instead of redesigning a chip. That is the argument. The paper demonstrates couplers; it does not demonstrate a fault-tolerant computer.

The commercial logic around this is moving faster than the physics. On 26 October, Intel said it had signed a definitive agreement to acquire Classiq, an Israeli quantum software company whose synthesis platform is agnostic to qubit modality. Intel's release frames the deal as combining Classiq's algorithm synthesis with its own silicon spin-qubit manufacturing. The deal adds Quantum Processing Units to a portfolio that already spans CPUs, GPUs and ASICs.

"Joining forces with Classiq transforms Intel's Quantum Computing Group into a full-stack quantum computing business," said Lip-Bu Tan, Intel's chief executive, in the company's announcement.

The transaction is subject to regulatory approval, Intel said. No price was disclosed in the release.

Error correction moves from papers to products

Error rates are the thread running through all of this. Quint sells a chip architecture aimed at them. IBM's interconnect work exists so that error-corrected modules can be combined. And on 26 October, arXiv posted "Magic state cultivation on a superconducting quantum processor" by Emma Rosenfeld, Craig Gidney and a long author list from the Google quantum team. The paper was first submitted in December 2025 and revised last week. Magic states are the raw material for the non-Clifford gates that fault-tolerant algorithms need. Producing them cheaply is one of the quieter bottlenecks in the field.

Not every milestone this week came from a chip fab. Researchers at Florida Atlantic University, led by Prof. Arslan Munir, published a comparison of quantum machine learning methods for lung cancer prediction under deliberately limited training data in Discover Quantum Science. HPCwire reported on 2 November that a D-Wave annealing machine running a quantum support vector machine reached 94.31% accuracy in simulation and 91.90% on quantum hardware.

The dataset deserves scrutiny. It started with 309 public records covering smoking history, coughing, wheezing, fatigue and chest pain. It was then balanced with synthetic examples to 540 samples. The added records were not new patients. Models trained on 108 samples and tested on 432 across five rounds. A 91.90% hardware figure on a small, synthetically augmented dataset is a research result, not a clinical one. The authors present it that way.

The money question, and the naming question

Capital is arriving in small cheques and large acquisitions at the same time. Quint's €9.1 million is a seed-sized round for a company that needs cryogenics, fabrication access and physicists. Intel is buying software to sit on top of hardware it already makes. The gap between those two positions is the shape of the current market.

Meanwhile the AI infrastructure boom that is paying for a lot of adjacent compute has a public relations problem. The Verge reported on 3 November that OpenAI used the term "compute campus" in a Grid Protectors Program announcement on Monday. "Data center" appeared twice in roughly 900 words. Tom's Hardware noted on 2 November that Axios had reported the rebranding plan ahead of the US midterms. It cited OpenAI chief global affairs officer Chris Lehane describing the midterms as "a bit of a movie trailer for what you're going to see in 2028."

"OpenAI is clearly responding to the unpopularity of data centers," Holly Elmore, founder and executive director of PauseAI US, told The Verge. "Rebranding them 'compute campuses' may fool listeners once or twice but it's not the name 'data center' that is unpopular." That quote is The Verge's reporting, not ours.

Two arXiv papers this week point at the economic layer underneath all this compute. One, submitted 29 October by Nastaran Okati, Ricardo Olmedo and Moritz Hardt, formalises computational arbitrage. In that setup, an intermediary buys inference from a portfolio of models and resells it. The paper proves the allocation problem is generally NP-hard, then shows it becomes tractable under a diminishing-returns condition. The other, submitted 30 October by Genliang Zhu and Chu Wang, sets out an authorization protocol for computer-use agents. Those agents act across graphical interfaces, command lines, APIs and Model Context Protocol servers. The protocol was tested over 3,464 traces on 1,732 scenarios.

Neither paper is a milestone in the sense the quantum crowd means. Both describe the plumbing that will decide whether any of this hardware earns its keep.

What to watch next is less glamorous than qubit counts. IBM's couplers need to work between more than two modules. Quint needs to show its architecture in silicon rather than in a funding announcement. Intel needs its regulators to agree. And the field needs error rates low enough that a 120-qubit processor linked to another one is a computer rather than an experiment.