Samsung Backs Dutch AI Chip Startup Euclyd in $230 Million Round to Challenge Nvidia

Samsung has co-led a €200 million (about $230 million) Series A round for Euclyd, a Dutch startup building AI inference chips meant to give enterprises an alternative to Nvidia's GPUs (Korea Herald, TechRadar).

What Euclyd is building

Founded in 2024 by Bernardo Kastrup and Atul Sinha at the High Tech Campus Eindhoven, Euclyd is designing a complete chip system for inference, running trained AI models to produce answers, rather than training new ones. It moves away from the GPU design that made Nvidia the world's most valuable company, building both the processor and memory architecture around energy efficiency and cost per token instead of general-purpose flexibility. The pitch is that this cuts the energy spent moving data between memory and compute, a cost that grows as data centres approach their power and budget limits.

Detail	Euclyd Series A
Size	€200 million (about $230 million to $231 million)
Co-leads	Samsung, Somerset Capital Partners, EQT-managed Scaleup Europe Fund, Innovation Industries
Other investors	EIFO, imec.xpand, Brabant Development Agency, Quadri
Chairman	Peter Wennink, former ASML CEO
First shipments	2028
Target	Thousands of corporate customers by 2030

Euclyd plans two revenue streams: selling hardware and rack systems directly to companies that want secure, self-hosted inference, and licensing its intellectual property to other chip designers. The money will expand the engineering team and speed up its silicon and systems roadmap (Channel Insider).

Why Samsung matters more than the money

Kastrup has said Samsung "can help us in more ways than money", pointing to its memory manufacturing scale, engineering knowledge and supply chain. Samsung is one of the largest memory makers in the world, and memory bandwidth and packaging increasingly decide how well AI processors handle large workloads, which makes it a strategic partner rather than only an investor.

The deal also fits Samsung's widening ties across AI chips. It is reported to be working with OpenAI on next-generation chip designs and is expected to supply memory for the Stargate data centre project. Bringing in Wennink as chairman adds semiconductor credibility well beyond a typical Series A.

Part of a much bigger pattern

Euclyd is one of a fast-growing field of Nvidia challengers. OpenAI has unveiled a custom inference chip built with Broadcom, Anthropic has reportedly held talks with Samsung about chips of its own, and Meta, Amazon and Google have run custom silicon programmes for years. The shared logic: purpose-built chips rarely match Nvidia's best GPUs across every workload, but for narrow, high-volume jobs like inference at scale, a dedicated chip can be much cheaper to run than GPUs bought at Nvidia's margins. Training frontier models, where Nvidia's hardware and CUDA software are hardest to dislodge, is a different matter, but less of the industry's future spending growth has to flow to Nvidia.

What to watch next

Euclyd's power and cost claims are unproven in commercial deployment. Whether it meets its 2028 shipping target, and whether its efficiency holds up under real enterprise workloads, will decide if it dents Nvidia's lead.

Related on Gadgets365

• Nvidia's Jensen Huang expects chip sales to double next year
• Mistral raises €3 billion in Europe's largest tech funding round
• Samsung posts a record 107.4 trillion won quarterly profit