Invest5 publishers3 min readPublished
Samsung co-leads EUCLYD's EUR 200m Series A against a 2028 hardware date
Samsung has co-led a EUR 200m round for Euclyd, an Eindhoven chip company founded in 2024, and the chief executive says the Korean group's memory manufacturing and supply-chain reach matter as much as its cash.
The Investor · Invest desk
What happened
- Samsung co-led Euclyd's 200-million-euro Series A alongside Somerset Capital Partners, Innovation Industries and the EQT-managed Scaleup Europe Fund, a round CNBC reported at $231 million.
- The company, founded in 2024 by Bernardo Kastrup and Atul Sinha in Eindhoven, is designing an inference chip system whose processor and memory architecture differs from a GPU's.
- Kastrup told CNBC that Euclyd aims to begin rolling out physical chip systems in 2028 and to be serving thousands of enterprise customers by 2030.
Compiled by The InvestorSomething wrong?How this is made
Why it matters
- constraint With first hardware due in 2028, the enterprise revenue line cannot be tested for years, so the follow-on round will be priced on engineering milestones and whatever licensing has been signed.
- exposure The Scaleup Europe Fund is carrying a position worth about 4% of its 5-billion-euro target in a company with no shipped product, while the same sum is six hundredths of a percent of EQT's book.
- precedent Selling chip IP into a market where Google, AWS, Meta and OpenAI already run in-house silicon programmes means the licensing line competes with the customers' own design teams.
- decision An enterprise that wants self-hosted inference has nothing from Euclyd to evaluate before 2028, so for now this round is an allocator's decision.
Bernardo Kastrup was specific about what Samsung adds beyond money. "Samsung can help us in more ways than money," he told CNBC [6], and then gave his reasons: "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" [7]. Euclyd's own description of its platform is programmable ASIC compute, processor-memory co-design and system-level optimisation [14].
Launched in 2024, aiming to begin rolling out physical systems in 2028 [3][8], the company has four years between founding and first hardware, and EUR 200m [1] spread across that gap averages about EUR 50m a year [3]. Euclyd says the financing will expand its engineering organisation, accelerate the silicon and systems roadmap, strengthen ecosystem partnerships and prepare it for deployment across enterprise, sovereign and hyperscale markets [22]. The stated target of thousands of enterprise customers sits two years past that first rollout [4].
EQT manages EUR 341bn in total assets [19], and its Scaleup Europe Fund is targeting EUR 5bn of pooled public and private capital for deeptech, AI and life sciences [18]. The whole Euclyd round is about 4% of that fund's target and roughly six hundredths of one percent of EQT's assets under management [1][2]. The co-leads did not disclose how the EUR 200m divides between them.
Two revenue lines are planned: racks and hardware sold to enterprises that want secure self-hosted inference, and intellectual property sold to companies building their own chips [10]. The second line can produce cash before any rack ships, and its buyers are the firms that have already decided to design in-house. OpenAI said in August that its first chip, the Jalapeño, had "industry-leading speed and efficiency" [11], and Google, AWS and Meta are each developing their own AI silicon [12].
The product claims in the announcement are the company's: craftwerk billed as the world's first agentic AI silicon, and craftwerk station CWS as the world's lowest-power exascale AI factory [13]. CNBC reported that Euclyd's systems have yet to be proven at scale in commercial deployments [9], and the announcement carries no benchmark figure.
What EUR 200m buys here, in my view, is a team and a supply relationship. Peter Wennink, formerly president and chief executive of ASML, chairs the board [17], and Samsung's Dede Goldschmidt said Euclyd "combines an accomplished team with a differentiated vision addressing constraints in AI datacenters" [16].
If a licensee signs in 2027, the 2028 date stops governing the cash and this becomes an IP business with a hardware option attached. If memory allocation is the scarce input in 2028, the thing worth owning here is the Samsung relationship, and the architecture is what secured it.
What would falsify the first reading is a paid enterprise deployment or a named licensee before 2028. Absent either, the next price for this equity gets set on engineering milestones, with the first rollout still dated 2028 [8].
What to watch
- A named licensee or a paid enterprise deployment before 2028 would move Euclyd's revenue off the hardware date.
- Whether Samsung's involvement extends from equity to a disclosed memory supply or manufacturing agreement.
- The split of the 200 million euros between the four co-leads, and the price at which the next round is struck.