Nvidia’s AI chip empire has another challenger.
Dutch startup Euclyd has raised more than €200 million ($231 million) in Series A funding to develop AI chips designed specifically for inference, the stage where trained models process requests and generate responses.
Samsung co-led the round with Somerset Capital Partners, EQT’s Scaleup Europe Fund and Innovation Industries. Additional investors included EIFO, imec.xpand, the Brabant Development Agency and Quadri, according to Euclyd.
The Eindhoven-based company was founded in 2024 by Bernardo Kastrup and Atul Sinha at the High Tech Campus Eindhoven. Its technology takes a different approach from the GPUs that dominate AI computing, combining processor and memory design to reduce the energy spent moving data.
“AI is becoming a foundation of economic growth, scientific discovery, and national competitiveness, but its potential will remain constrained unless we fundamentally change the infrastructure beneath it,” Kastrup said in Euclyd’s announcement.
Samsung brings more than money
Samsung’s involvement could be particularly significant because Euclyd’s strategy depends heavily on rethinking the relationship between computing and memory.
“Samsung can help us in more ways than money,” Kastrup told CNBC. “They are one of the biggest memory manufacturers in the world. They do a lot of engineering, they know a lot about systems, they know the supply chain, they have a huge network.”
Euclyd plans to generate revenue by selling hardware and rack-scale systems to organizations that want self-hosted AI inference, while also licensing its technology to companies developing their own chips.
The technology still has something to prove
At the center of Euclyd’s roadmap is CRAFTWERK, which uses 16,384 custom processors designed to work directly with the company’s memory architecture. Its planned CRAFTWERK Station system combines 32 chips and is aimed at exascale-class AI computing.
Euclyd says its architecture could deliver up to 100x the power efficiency per generated token of Nvidia’s Vera Rubin systems on comparable workloads. However, that figure is based on company modeling rather than commercial-scale testing. That distinction matters as Euclyd has not yet proven its systems in large commercial deployments, with physical chip systems targeted to begin rolling out in 2028.
A full-stack approach
The company’s unusual proposition is that it is not simply trying to build another AI accelerator. It is developing the processor, memory architecture and surrounding data center systems together.
That could give Euclyd more control over efficiency, but it also makes the engineering challenge much larger. A compelling benchmark on paper will not be enough if production, software compatibility or customer deployment proves difficult. Former ASML President and CEO Peter Wennink has joined Euclyd as chairman, bringing semiconductor industry experience as the company moves toward commercialization.
“The next phase of AI will be defined not only by model innovation, but also by the efficiency and scalability of its infrastructure,” Samsung’s Dede Goldschmidt said. Euclyd plans to use the new funding to expand its engineering team, accelerate silicon and systems development, and prepare for enterprise, sovereign and hyperscale customers.
Why channel partners should pay attention
Euclyd is still years away from broadly commercializing its rack-scale systems, so partners do not need to rethink their AI infrastructure portfolios today. But its strategy highlights an area worth watching: specialized infrastructure built specifically for AI inference.
For solution providers, that could eventually create more architectural choices for private and self-hosted AI deployments. Euclyd is designing its processors, memory architecture, and data center systems together, with the goal of lowering power consumption and cost per generated token.
Samsung’s involvement also gives partners another reason to watch the company as it moves toward commercialization. If Euclyd can turn its modeled efficiency gains into production hardware, it could eventually become another option for organizations evaluating large-scale inference infrastructure. For now, however, those performance claims remain company projections, and the CRAFTWERK Station is targeted for commercial availability in 2028.
More news: See how Samsung’s next-generation AI memory designs could reshape AI infrastructure for channel partners.