Euclyd, the ASML alum's chip startup, eyes €100M to take on Nvidia in inference
What's the deal? Euclyd, an Eindhoven-based AI chip startup founded in 2024, is in talks to raise at least €100M, according to CNBC. The company previously closed a seed round of under €10M and is now seeking capital to scale its technology and begin serving its first customers. Talks are ongoing and terms could change.
Euclyd was founded by Bernardo KastrupDealroom has a profile for this one. Try Dealroom →, a computer scientist, philosopher, and former director at ASML, who identified a gap in European efforts to develop inference-specific silicon when generative AI surged in 2023.
The company's product, CRAFTWERK, is a chip system designed from the ground up for AI inference — processing data across multiple locations simultaneously rather than shuttling it between memory and compute cores, as conventional GPUs do. Euclyd claims this delivers 100 times greater power efficiency for inference than Nvidia's latest Vera Rubin chips, though the figure has not been validated through large-scale commercial deployment.
Why now? AI inference — the act of running a trained model to generate outputs — is fast becoming the dominant AI workload as deployment scales globally. Existing GPUs were designed primarily for gaming and adapted for AI training; they are energy-intensive and increasingly expensive for inference at scale. The case for purpose-built inference silicon is strengthening as the cost of running frontier models becomes a central concern for enterprises and AI labs alike.
The timing is also geopolitical. Europe has no significant domestic AI chip industry, and the continent's dependence on Nvidia and US cloud infrastructure has become a strategic concern. Euclyd is explicitly positioning itself as a European answer to that gap.
What could go wrong? Euclyd's 100x efficiency claim is unvalidated at commercial scale — a significant caveat for a company still at the prototype stage. Custom silicon development is notoriously expensive and time-consuming, and the gap between a working prototype and volume production has derailed many well-funded chip startups before it.
The company currently has no deployed customers, with its first deliveries not expected until 2027. A multi-chiplet version of CRAFTWERK isn't expected until 2028. That long runway creates significant execution risk, particularly in a market moving at AI's pace.
The signal: Euclyd's advisory board is unusually credible for a company of its age — former ASML CEO Peter WenninkDealroom has a profile for this one. Try Dealroom →, Federico FagginDealroom has a profile for this one. Try Dealroom → (who designed the world's first commercial microprocessor at Intel), and Elastic founder Steven SchuurmanDealroom has a profile for this one. Try Dealroom →. That pedigree reflects a broader pattern: serious semiconductor veterans are increasingly lending their names and networks to European AI chip challengers, signalling genuine conviction that the Nvidia-dominated status quo is vulnerable.
Europe's AI sovereignty ambitions are creating a receptive funding environment for companies like Euclyd. Whether a startup founded a little over a year ago in an Eindhoven attic can actually deliver is an entirely different question.
Sources:
NL Times
CNBC
Tech Funding News
TechBuzz
Tech in Asia
Image credit:
Euclyd
J.V.
