Eindhoven is where ASML was built, and it is where the next generation of European chip companies is now taking shape. On September 15, one of them, Euclyd, announced a Series A round of more than 200 million euros, about $231 million, led in part by Samsung.
The co-investors read like a map of European deep-tech capital. Somerset Capital Partners, EQT’s Scaleup Europe Fund, and Innovation Industries joined Samsung as co-leads, with Denmark’s EIFO and imec.xpand among the followers. The same day, the company named Peter Wennink, the former president and chief executive of ASML, as chairman of its board.
Wennink’s arrival is the signal that matters as much as the money. A man who spent decades at the top of the world’s most important chip-equipment company does not take a chairman’s seat at a two-year-old startup casually. His name tells customers and suppliers that Euclyd is a serious bet, not an experiment.
Euclyd was founded in 2024 and is walking away from the GPU. Its products target inference, the everyday work of running AI models, rather than the training that has consumed most of the industry’s chips and most of its power. The thesis is that the next bottleneck in AI is not making models smarter but running them cheaply.
The company’s lineup includes a chip it calls craftwerk, which it describes as the world’s first agentic AI chip, and an exascale AI factory system built around minimum power draw. The names are ambitious, and so is the timeline: physical racks in 2028, and service for more than a thousand enterprise customers by 2030.
Chief Executive Bernardo Kastrup framed Samsung’s role in plain terms. Samsung can help with more than money, he said. Memory, engineering, and supply chain are where the real value sits, and those are exactly the places a small Dutch startup cannot build on its own.
The Samsung connection is the most interesting detail in the round. A Korean chip giant taking a lead position in a Dutch inference-chip startup is a bet that the post-GPU world will be diverse, and that the winners will come from places other than the established graphics-card makers.
Analysts said the deal is part of a broader European push to build chips for the next phase of AI. Europe has little presence in training chips, a market now dominated by a single American company. Inference, by contrast, is still open, and European startups are moving early.
The capital is substantial for a Series A, and it reflects how expensive chip design has become. A company that wants to tape out advanced silicon and stand up a product line needs hundreds of millions before it has sold a single unit. The investors are buying the design team and the roadmap, not revenue.
The risk is familiar to anyone who has watched hardware startups. Getting from a chip design to physical racks, from 2024 founding to 2028 delivery, is a long road full of manufacturing delays, yield problems, and the simple difficulty of winning enterprise trust. Wennink’s presence is meant to shorten that road.
The ambition, delivering an AI factory built for minimum power, speaks to the industry’s newest anxiety. Data centers are running into power limits, and chips that do the same work with less electricity are becoming a strategic asset rather than a technical preference.
For Samsung, the stake is also strategic. The company wants to be inside the next generation of AI infrastructure wherever it is built, and a position in a European inference player extends its reach beyond the memory and foundry business it already dominates.
Wennink’s career gives the appointment its weight. He led ASML through the years when its extreme-ultraviolet lithography machines became the single most important tool in chipmaking, and he knows what it takes to scale a hardware company from a Dutch town to a global supplier. His presence is a bet that Euclyd can follow a similar arc.
The inference market Euclyd is chasing is distinct from the training market that made Nvidia’s fortune. Training happens once, at enormous scale; inference happens constantly, everywhere a model is used, and it is where most of the long-term power and cost will concentrate. A chip built for inference is aimed at the larger of the two markets.
The European context adds a policy tailwind. The continent has watched its chipmaking capacity erode for decades and has committed public money to rebuilding it. A homegrown inference-chip company, backed by a Korean giant and chaired by an ASML veteran, is exactly the kind of project that policy was written to support.
The 2028 target will be the test. If the racks arrive on time and the enterprise customers follow, Euclyd will have proved that Europe can build a piece of the AI stack it currently only consumes.
