Euclyd Lands €200M From Samsung for Inference Chips

A Two-Year-Old Chip Startup Raises Series A Money at Scale

Euclyd, an Eindhoven-based AI chip company founded in 2024, has closed a Series A of more than €200 million — about $231 million. CEO Bernardo Kastrup told CNBC on September 14, 2026 that Samsung co-led the round alongside Somerset Capital Partners, the Scaleup Europe Fund managed by EQT, and Innovation Industries. Tech Funding News reports that EIFO, imec.xpand, the Brabant Development Agency and Quadri also participated, and that former ASML chief executive Peter Wennink has joined as chairman of the board.

For a company with no chips in customers' hands, that is a large number. It is also a legible one: capital is flowing hard toward anything that promises cheaper AI inference.

• Round size: more than €200 million, roughly $231 million, co-led by Samsung
• Prior expectations: Euclyd told CNBC in April it was seeking at least €100 million — the final round more than doubled that
• Product timeline: Kastrup says physical chip systems begin rolling out in 2028, with a target of thousands of enterprise customers by 2030
• Business model: selling rack systems for self-hosted inference, plus licensing the architecture as IP

What Is Euclyd Actually Building?

The thesis is that GPUs are the wrong shape for inference. Training and inference are different workloads: a GPU spends a large share of its energy budget shuttling data between memory and compute cores, which is a reasonable trade when you are doing enormous matrix multiplications for weeks, and a wasteful one when you are answering a prompt.

Euclyd's chip, called CRAFTWERK, packs 16,384 custom processors designed to operate directly on data held in memory rather than fetching it into separate compute cores. A rack-scale configuration, CRAFTWERK Station, combines 32 of those chips and targets one exaflop of compute by 2028.

The company claims up to 100 times the power efficiency of Nvidia's Vera Rubin generation, modelled on Meta's Llama 4 Maverick. Be clear about what that figure is: a company-supplied projection from simulation, not a measurement from a live deployment, and not something anyone outside Euclyd has verified. Treat it as a design target rather than a result.

Why Does Samsung's Involvement Matter More Than the Money?

This is the part investors should actually weigh. Kastrup was direct about it: Samsung is one of the largest memory manufacturers on earth, with deep systems engineering and an established supply chain. For a chip architecture whose entire premise is a different relationship between memory and compute, having a memory giant as a strategic investor is worth more than the cheque.

Getting a novel architecture from simulation to silicon is where most chip startups die, and the failure mode is usually manufacturing and supply chain rather than design. Samsung's semiconductor innovation group framed its participation around infrastructure efficiency being the defining constraint of the next AI phase — which is, conveniently, also an argument for selling more high-bandwidth memory.

The Competitive Field Is Getting Crowded

Euclyd is not alone in going after Nvidia's inference business, and the field has been well funded. Axelera AI, also in Eindhoven, raised more than $250 million in February for edge inference. Etched's valuation climbed from $5 billion in June to $21 billion by August, a round we covered when Etched raised $700 million. Meanwhile OpenAI, Google, AWS and Meta are all building in-house silicon.

What distinguishes Euclyd is scope: rather than picking edge inference or pure IP licensing, it is attempting the full stack — chip, memory architecture and the datacenter systems around them. That is the most expensive path and the slowest to validate.

For anyone tracking the semiconductor supply chain as an investment theme, the signal here is less about Euclyd specifically and more about where money is moving. Inference, not training, is where the operating cost of AI now lives, and that is drawing capital toward memory-centric architectures and the equipment that makes them — a thread we also followed through ASML's High-NA EUV roadmap. Euclyd says it is in talks with four prospective customers. The next 18 months, and first silicon, will settle whether the thesis holds. More market coverage in our stock trading section.

This article is news and analysis, not investment advice.

Sources: CNBC — September 14, 2026; Tech Funding News — September 2026.