Euclyd, an Eindhoven startup founded by a former ASML director, is seeking between €100 million and €200 million to build a purpose-built AI inference chip it claims is 100 times more power-efficient than Nvidia’s latest hardware.
Its advisory and investor board includes Peter Wennink, who led ASML for over a decade, and Federico Faggin, the physicist who designed the Intel 4004, the world’s first commercial microprocessor, in 1971.
The raise would rank among the largest early-stage rounds for a European AI chip company, arriving as the continent races to build homegrown alternatives to Nvidia’s dominance in AI infrastructure.
The man who designed the world’s first commercial microprocessor is backing a startup working out of Eindhoven. So is the former CEO who turned ASML into Europe’s most valuable technology company. That combination of institutional weight and semiconductor pedigree is one reason Euclyd, a Dutch AI chip startup founded less than two years ago, is already in serious funding conversations.
According to Bloomberg, it is seeking up to €200 million in a Series A round. The founder, Bernardo Kastrup, told CNBC in April that the figure he is targeting is at least €100 million.
However the final number lands, the people standing behind it matter as much as the money.
From an attic in Eindhoven to the front line of the inference race
Kastrup built Euclyd’s early concepts from his home attic, working late into the night after finishing his day job. A computer scientist and philosopher who spent years at Philips Research, Silicon Hive, which waslater acquired by Intel, and ASML, where he focused on product strategy. He watched the generative AI surge of 2023 and saw a gap that no European company was moving to fill: next-generation silicon designed specifically for AI inference, the process by which trained AI models generate responses and make decisions in real time.
Kastrup launched Euclyd in 2024 with three co-founders: Gerard Egelmeers, who leads processor firmware co-design; Ingolf Held, who oversees product; and Harm Peters, who heads silicon design engineering. Together they bring decades of experience in processor architecture and semiconductor development.
The advisory and investment board they assembled is arguably even more striking.
Peter Wennink, who spent more than a decade as ASML’s chief executive, helping make it the most valuable technology company in Europe before retiring in 2024, is both an adviser and an investor. So is Federico Faggin, an Italian-American physicist who designed the Intel 4004 in 1971, the world’s first commercial microprocessor and the chip that made the modern computing industry possible. Rounding out the board is Steven Schuurman, founder of the enterprise search company Elastic.
Why inference and why now
For years, the most bankable bet in AI hardware was training, building and refining the enormous foundation models that power ChatGPT, Gemini, and their peers. Nvidia won that race decisively. Its GPUs, originally built for video games, turned out to be well-suited to the parallel computation training requires, and the company became the world’s most valuable company on the back of that demand.
But the next infrastructure bottleneck is different. Running trained models at scale is known as inference, and it has rapidly become one of the most expensive, energy-intensive problems in enterprise technology. GPUs are not ideal for this job, as they spend enormous energy shuttling data back and forth between memory and compute cores, a design that made sense for gaming but wastes power during inference.
Euclyd works by eliminating that data movement altogether. Its platform, packs 16,384 custom processors that operate directly on data stored in memory, rather than moving it back and forth. The startup claims this delivers up to 100 times greater power efficiency for inference workloads compared to Nvidia’s latest Vera Rubin chips, based on performance modelled against Meta’s Llama 4 Maverick, though the figures are projections and have not been validated at commercial scale. A rack-scale system combining 32 of these units, targeting one exaflop of compute, is expected by 2028. The company is in discussions with four potential customers, two of which it hopes to begin supplying in 2027.
A crowded race, but a different fight
Euclyd is not alone in this pursuit. In the first half of 2026, investors funnelled more than $250 million into Dutch peer Axelera AI, which is targeting AI inference at the edge, and $220 million into UK-based Olix, which is developing photonics processors. Globally, Etched raised $500 million and Cerebras is in talks for up to $10 billion, as enterprises and hyperscalers hunt for alternatives to Nvidia’s supply-constrained GPUs.
What distinguishes Euclyd is not just its architectural approach but the specific ambition behind it. Kastrup has argued publicly that AI sovereignty cannot be achieved without control over hardware. ASML gave Europe dominance in semiconductor manufacturing equipment. Euclyd is the most credible attempt yet to extend that dominance into the chips themselves.
Whether that thesis is right will take years to prove. The €100 million to €200 million now being raised is intended to take the company from prototype to initial customer deliveries, a transition that will test every element of the founding team’s semiconductor experience.
The chip industry has a long history of bold architectural bets that never reached production scale.