Confirmed in a primary source 15 Sept 2026 Chips

A Dutch chip company raised more than 200 million euros, and Samsung helped lead the round.

EUCLYD, founded in 2024 on the High Tech Campus in Eindhoven, announced a Series A of more than 200 million euros, reported in dollars at 231 million, co-led by Samsung, Somerset Capital Partners, the EQT-managed Scaleup Europe Fund and Innovation Industries, with Peter Wennink, who ran ASML from 2013 until April 2024, joining as chairman. The caveat is that nothing has shipped: the release names two products, Craftwerk and the Craftwerk Station, but publishes no transistor count, no memory bandwidth, no power figure, no tokens per second, no tape-out date, no customer and no valuation.

Until now

The credible attack on Nvidia came from American hyperscalers designing their own accelerators and from a handful of United States startups; Europe's chip story was lithography and manufacturing, not compute.

Now

Treat European inference silicon as a funded category with Asian memory money and ASML's former chief executive attached, and watch what it publishes rather than what it raises.

The round is large for a company that has not shipped anything. EUCLYD is two years old, sits on the High Tech Campus in Eindhoven, and was founded by Bernardo Kastrup and Atul Sinha. It says it is building two things: Craftwerk, a chip it calls agentic AI silicon, and the Craftwerk Station, a rack-scale system it calls an ultra-low-power exascale AI factory. The argument underneath both is that the bottleneck in running large models is not raw arithmetic but moving data between the processor and its memory, and that designing the two together beats buying them separately. That argument is not new and not obviously wrong. What is new is who paid for it. Samsung, which makes the memory the argument is about, co-led. So did Somerset Capital Partners, the Scaleup Europe Fund that EQT manages, and Innovation Industries. Denmark's state export fund, imec's venture arm, the province of North Brabant and Quadri came in behind them. Peter Wennink, who ran ASML for eleven years and left in April 2024, took the chairman's seat.

Set it next to the rest of the week and the shape is clearer. On September 8 Mistral closed 3 billion euros with Samsung among the investors. On September 14 Cornelis Networks raised about 205 million dollars to attack the network between chips, with Qualcomm beside it. On September 15 EUCLYD raised more than 200 million euros to attack the chip itself. Three rounds, eight days, three different layers of the same stack, and the common thread is money that is not Nvidia's and not American. Against that, the release is unusually empty where it matters: no transistor count, no bandwidth, no watts, no tokens per second, no tape-out date, no named customer, no valuation. A press release with a former ASML chief executive in it and no numbers in it is a statement about confidence, not about silicon. Treat European inference silicon as a funded category with Asian memory money and ASML's former chief executive attached, and watch what it publishes rather than what it raises.

How it got here

Sep 08
Mistral AI closes 3 billion euros, the largest equity round a European technology company has done, with Samsung among the investors. Carded in Edition 016.
Sep 14
Cornelis Networks announces about 205 million dollars led by IAG Capital Partners for its network fabric, with Qualcomm alongside it. The Philadelphia semiconductor index has its worst day since July. Both carded in Edition 022.
Sep 15
EUCLYD announces more than 200 million euros, co-led by Samsung, Somerset Capital Partners, the Scaleup Europe Fund and Innovation Industries, and names Peter Wennink chairman.
Sep 15
The ten-year United States Treasury yield closes at 5.006 percent, its highest since 2007, the day before a Federal Reserve decision.
Who has to react
Nvidia, whose inference margin is the target of all three rounds this week; Samsung's memory division, which now holds equity in both a European model company and a European chip company; ASML, whose former chief executive has taken a chair at a customer of its customers; the Dutch and Belgian public investors who put state money into an unshipped design; EIFO in Copenhagen, now exposed to Dutch silicon; every European official who has argued that the continent cannot build compute; and the founders of the eight or nine other companies pitching processor-memory co-design, whose next meeting just got easier or much harder.

The evidence

ObservedTuesday September 15, 2026, dateline Eindhoven, Netherlands. EUCLYD announced a Series A of over 200 million euros, carried on PR Newswire and on the company's own press page, and reported in dollar terms at 231 million by DatacenterDynamics, Quartz and Yahoo Finance and at about 230 million by CNBC.

The whole observation, as recorded

Co-leads, in the order the release lists them: Samsung; Somerset Capital Partners; the Scaleup Europe Fund, managed by EQT; and Innovation Industries. Also participating: EIFO, the Export and Investment Fund of Denmark; imec.xpand, the venture arm of the Belgian research institute imec; the Brabant Development Agency, the development company of the Dutch province that contains Eindhoven; and Quadri. Peter Wennink, president and chief executive of ASML from 2013 to April 2024, joins as chairman. The company was founded in 2024 by Bernardo Kastrup, chief executive, and Atul Sinha, and sits on the High Tech Campus in Eindhoven, a short drive from ASML's headquarters in Veldhoven. Two products are named. Craftwerk, which the company calls agentic AI silicon. And the Craftwerk Station, which it calls an ultra-low-power exascale AI factory, described as programmable ASIC compute with processor and memory designed together rather than bought separately. Quotes: Kastrup, that 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; Wennink, that Europe possesses world-class capabilities across semiconductors, advanced manufacturing and systems engineering; Dede Goldschmidt, a Samsung senior vice president, that the next phase of AI will be defined not only by model innovation but also by the efficiency and scalability of its infrastructure. Stated use of funds: expand the engineering team, accelerate the silicon and systems roadmaps, strengthen partnerships, and prepare for commercial deployment across enterprise, sovereign and hyperscale markets. No headcount was given. The surrounding week: Cornelis Networks announced about 205 million dollars on September 14 for the network layer with Qualcomm alongside, carded in Edition 022; Mistral closed 3 billion euros on September 8 with Samsung among the investors, carded in Edition 016, which puts Samsung equity into two European AI companies inside eight days; and the Scaleup Europe Fund also led Tandem Health's 86.49 million euro round announced September 14. On the same day the ten-year United States Treasury yield closed at 5.006 percent, its highest since 2007.

Established

EUCLYD announced a Series A of more than 200 million euros on September 15, 2026, with Samsung, Somerset Capital Partners, the Scaleup Europe Fund and Innovation Industries as co-leads.
Peter Wennink, chief executive of ASML from 2013 to April 2024, has taken the chairman's role at EUCLYD.
Samsung has now taken equity positions in two European AI companies within eight days, Mistral on September 8 and EUCLYD on September 15.
Public money from Denmark, Belgium and the Dutch province of North Brabant is in the round.

Not established

That EUCLYD has a working chip. No transistor count, bandwidth, power figure, throughput number or tape-out date has been published, and no independent party has measured anything.
That Nvidia's inference position is weakening. Three funded attempts in a week is evidence about capital, not about market share, and Nvidia's share of inference silicon has not been shown to move.
That Samsung has committed to manufacture or supply anything. The release names an equity investment and a quote, not a foundry slot, a memory agreement or a volume.
That Europe can build competitive AI compute. One Series A in Eindhoven does not establish a supply chain, and the design will be fabricated somewhere that is not named.

Why it weighs this much

Signal strength

CostHigh

CommitmentMedium

ReversibilityHard

PersistenceYears

Cost is high and unambiguous: more than 200 million euros left investors' accounts, and Samsung's participation is corporate money with a strategic bill attached, not a fund ticking a box. Commitment is medium rather than high because a Series A funds an attempt, not an outcome, and none of the investors has promised a second cheque. Reversibility is hard in the only sense that matters here, which is that the money is spent and cannot be recalled, though the plan it funds can change completely. Persistence is years because silicon is slow: the earliest honest test of this round is a tape-out and then a customer, and that is a 2027 and 2028 conversation. We are not scoring this as a turning point. It is a well-funded attempt, and the bars say so.

What to watch

strengthens EUCLYD publishes a measured performance or efficiency figure for Craftwerk against a named Nvidia part, or announces a tape-out or a named customer, by June 30, 2027.
weakens No tape-out, no independent measurement and no named customer is published by EUCLYD, Cornelis Networks or Mistral's silicon partners by June 30, 2027, and no further round above 100 million euros closes in European inference silicon.

The words

Series A: The first large round of venture money a company raises, usually the point where it moves from a prototype to building the real thing.
Inference: Running a trained AI model to answer a question, as opposed to training it in the first place. Most of the electricity and most of the future spending sits here.
Tape-out: The moment a chip design is finished and sent to a factory to be made. Until a company tapes out, a chip is a drawing.
ASIC: A chip built for one job rather than for anything. It is faster and cheaper per task than a general-purpose processor, and useless if the job changes.

Your call

Which explanation do you believe?

A call is final. Other people's calls show after yours.