Euclyd

We’re 49 days from AI Infra 2026, the most consequential AI event of the year where the vanguard of silicon, hyperscalers, and model developers converge to discuss where AI is heading. Custom silicon is at the epicenter of that story. The dominant grip GPUs have had on the market is weakening. Headlines have held one story after another of big funding rounds and teams rapidly bringing better solutions to market - delivering huge gains in inference efficiency and energy density to reduce the burden on operators. As a Media Partner for the Summit, I’m kicking off our pre-event coverage by spotlighting those that are setting the pace and democratizing AI in the process: Cerebras - A year ago, some still doubted a plate-sized accelerator was viable, let alone competitive. Now public, Cerebras has deals with OpenAI and AMD by driving compute and token delivery up, and latency down. I’ll bet the few remaining detractors will be quieter this year. Euclyd - Bernardo and his team are reimagining inference from the logic gate up, purpose-built for agentic AI. Their CRAFTWERK architecture delivers an astonishing 16K SIMD cores for 100x power efficiency, and having early investment from ASML’s former CEO is a clear signal of confidence in their approach. d-Matrix - Let's remember Sid and Sudeep's team pioneered Digital In-Memory Compute to make ultra-low latency, high-throughput LLM inference real. They’ve now built multiple products and have a full-stack solution in production for Enterprise AI. Never take your eye off the quiet builders. Positron AI - Their arc: stay lean and build fast. Mitesh and Thomas's team took 18 months to move from FPGAs to production-ready silicon. Then they inked a deal with Oracle, the first silicon company other than Nvidia or AMD to land a hyperscaler. AI Infra is a great platform to hear what's next. Rebellions - Sunghyun and his team are hitting home runs. From launching their flagship Rebel100 chip with 144GB HBM3e, to running SK Telecom’s LLM on a single server, they’re building a global sovereign AI powerhouse that's going toe-to-toe with hyperscale compute. A blockbuster IPO looks within reach. SambaNova - Their Reconfigurable Dataflow Architecture might have once felt overbuilt, but then model sizes blew up and Rodrigo's early vision now looks prophetic. A three-year Intel deal, $1B in new funding, and rapid moves to scale manufacturing signal Samba’s time is now. Tenstorrent - What haven’t I said about the criticality of open-source architecture and chiplets to give teams control over their silicon and democratize AI? Whether they're looking to exit or IPO remains to be seen but it's clear that these rebel builders are on a mission and I look forward to seeing more about Ascalon, Galaxy, and rest of their lineup. If you could pass the mic to one of these founders at AI Infra, who would it be and what’s the one question you’d want them to address? #semiconductorindustry #artificialintelligence #AIInfraSummit

Europe will not build a sovereign AI inference stack through isolated initiatives. The system is too complex. A useful inference platform is not one component. It is silicon, software, power, security, packaging, rack integration, cloud deployment, manufacturing access and industrial demand — all aligned around real workloads. From a system perspective, this is where Europe’s challenge becomes visible. We have strong capabilities in many layers. But if every actor optimizes only its own part, the result will not be a platform. It will be a collection of excellent fragments. A concrete pattern repeats across deep tech ecosystems: the critical question is not whether the technology works in isolation. It is whether the surrounding interfaces make it adoptable, scalable and investable. That is why the next step should not be another symbolic project. It should be a focused European AI inference alliance. Such an alliance would need a clear role for each layer: 🔹 Specialized AI accelerator IP and software 🔹 Semiconductor credibility, power efficiency and security 🔹 Packaging, manufacturing and system integration paths 🔹 Cloud, HPC and industrial anchor customers The real question is not who can build one more chip. The real question is who can orchestrate the stack Europe will need. Europe does not need another isolated initiative. Europe needs a strong AI inference alliance. #semiconductors #AI #alliances #deeptech #Europe

🎉 Meet the winners of the G&A Awards 2026. Today: Euclyd The explosive growth of AI has produced an uncomfortable truth: intelligence at scale demands ever more energy, memory and data-centre infrastructure. Euclyd is not choosing a small optimisation within existing systems, but a fundamentally new approach to AI inference. Read more here: https://lnkd.in/esCp2x_v

Our people are, by far, our greatest asset! #Euclyd _is_ its people.

Thanks for the trust, Gerard & Anton! It was a lot of fun!

Thanks for the confidence, Gerard & Anton! It was a lot of fun!

Europe should stop framing AI compute as a GPU race. The more strategic question is where Europe can build a differentiated position before the next dependency becomes permanent. Training gets the headlines. But inference is where AI becomes industrial reality. It is the layer that will run inside factories, vehicles, energy systems, telecom networks, healthcare environments, robotics, defence applications and public infrastructure. What often gets underestimated is that inference has a different strategic profile than training. Peak performance matters — but it is not the only metric. For many industrial and sovereign workloads, the decisive questions are: 🔹 How much energy does each inference consume? 🔹 How secure is the workload from chip to cloud? 🔹 Can the system be deployed reliably under European control? A useful example is industrial AI. The value is not created by running the largest model once. It is created by running many smaller, specialized workloads continuously, securely and efficiently across real production environments. That changes the architecture question. The real question is not whether Europe can build another GPU. The real question is whether Europe can build the trusted AI inference layer its industries will depend on. #semiconductors #AI #deeptech #industrialpolicy #Europe