Euclyd lays out a dual hardware-and-licensing strategy after its $231 million Samsung-backed Series A.
The AMW Read
A known Nvidia-alternative entrant details its business model on an already-covered sub-$500M round, so novelty stays incremental while chip-architecture specificity keeps significance at segment level.
Euclyd lays out a dual hardware-and-licensing strategy after its $231 million Samsung-backed Series A.
Euclyd, a Netherlands-based chip startup founded in 2024, detailed plans for its €200 million-plus ($231 million) Series A, co-led by Somerset Capital Partners, the Scaleup Europe Fund and Innovation Industries, with Samsung also participating. CEO Bernardo Kastrup said the funding supports two lines of business: selling complete AI inference rack systems directly to enterprises for on-premises deployment, and licensing Euclyd's combined processor-and-memory architecture to other chipmakers building their own AI hardware. The company does not expect to ship commercial hardware until 2028 and aims to serve thousands of enterprise customers by 2030.
Nvidia's GPU dominance in AI inference has drawn a wave of challengers attacking the cost structure from the memory-bandwidth side rather than raw compute alone. Samsung's involvement is notable less for the capital than for the memory engineering, supply-chain access and systems know-how Kastrup cited — a dependency built into Euclyd's approach, since it designs processing and memory together rather than sourcing them separately. The IP-licensing line is effectively a hedge: if direct enterprise hardware sales underperform against an entrenched CUDA ecosystem, Euclyd can still monetize the architecture through other chipmakers. Per the AI Market Watch index, which tracks the AI Chips/Semiconductors category across roughly 5,000 companies (coverage, not a census), Euclyd's round adds to a widening list of Nvidia alternatives that also includes OpenAI's newly disclosed Jalapeño chip and in-house silicon efforts at Google, AWS and Meta.
For enterprise buyers, the 2028 shipment target and lack of at-scale deployment mean Euclyd is a multi-year bet rather than a near-term Nvidia substitute; on-prem inference planning today still has to assume GPU supply chains. For investors, the hardware-plus-licensing structure is worth tracking — if the processor-memory design proves out, the licensing business could scale without the capital intensity of building and shipping full rack systems.
