Samsung leads 200 million euro round for Euclyd AI inference silicon

The massive investment in Euclyd signals a shift toward specialized hardware as the industry hits the physical limits of general-purpose GPUs for large-scale model execution. By prioritizing memory-bandwidth and cost per token over raw arithmetic throughput, Euclyd addresses the primary economic bottlenecks facing hyperscale and enterprise AI deployments. This move connects to the broader streaming and media ecosystem as providers seek more efficient ways to run agentic AI and recommendation engines at scale. Success depends on whether the craftwerk architecture can deliver measurable gains in watts per token during production silicon testing. Watch for independent workload validation and software stack maturity as the company approaches its 2028 hardware launch.

Euclyd's craftwerk architecture enters a market where several established and emerging players are already shipping or sampling purpose-built inference silicon. In August 2026, Samsung announced its Mach-2 inference accelerator targeting large language model workloads with HBM4 memory integration, positioning the Korean giant's own silicon alongside its investment in Euclyd. The dual-track approach suggests Samsung is hedging across multiple inference architectures rather than committing to a single design. Meanwhile, Groq reported in July 2026 that its LPU inference chips had been deployed across three hyperscale data center customers, processing more than 500 tokens per second per chip on Llama-class models, establishing a performance benchmark that Euclyd's craftwerk station CWS will need to match or exceed when it reaches commercial availability. The funding round also lands amid intensifying regulatory and policy pressure on AI compute supply chains in Europe. In June 2026, the European Commission published its AI Compute Action Plan, allocating €4 billion toward sovereign inference infrastructure and prioritizing European-designed silicon to reduce dependence on US-based GPU vendors. Euclyd's Eindhoven base and its backing from the Scaleup Europe Fund and Innovation Industries align directly with that policy framework. Peter Wennink, former ASML CEO and Euclyd advisory board member, told Financial Times in September 2026 that Europe's lithography advantage gives the continent a structural edge in bringing novel chip architectures to volume production, though he cautioned that design-to-fab timelines remain the primary risk for pre-silicon startups. For streaming and media buyers evaluating inference hardware for recommendation engines, content moderation, and agentic AI workflows, the competitive landscape extends beyond Euclyd. Netflix AI production tools utilized across 300 titles to cut costs demonstrate how major platforms are already integrating AI into their pipelines, creating sustained demand for cost-efficient inference at the edge and in the cloud. That demand profile is exactly what Euclyd's craftwerk architecture targets: high-throughput, low-power token generation for always-on workloads rather than burst training jobs. The company's 2028 commercialization timeline means streaming platforms planning 2027 infrastructure refreshes will likely evaluate AWS agentic video intelligence, Groq, Cerebras, and Samsung's own Mach-2 before craftwerk silicon is available for independent benchmarking.