Bold claim: Europe may finally own the future of AI compute, but the path there clashes with hard financial realities.
In Eindhoven’s AI Innovation Center at the High Tech Campus, Bernardo Kastrup speaks with the calm, seasoned voice of someone who has already worn many hats: computer scientist, philosopher, ASML strategist, founder. Now, perhaps reluctantly, he stands as the face of Europe’s most ambitious quest to reinvent AI computing from the ground up.
Euclyd, the company he launched just over a year ago, has quickly become more than a startup: it’s a phenomenon. Backed by industry heavyweights—including former ASML CEO Peter Wennink, legendary Intel engineer Federico Faggin, Silicon Hive’s Atul Sinha, and Elastic founder Steven Schuurman—Euclyd burst out of stealth earlier this year and instantly dominated conversations. It wasn’t merely the ambition; it was the tone: unapologetically confident, deeply technical, and globally ambitious—Europe speaking with Silicon Valley swagger.
“Yes, you can call it bragging,” Kastrup chuckles. “It’s tongue-in-cheek, but confidence matters. Why should Americans have a monopoly on ambition?” Yet beneath the humor lies a deeper frustration that sparked Euclyd’s creation.
“I looked around and asked: Who allowed this to happen?”
When generative AI surged into public consciousness in 2023, Kastrup began surveying Europe’s landscape. World-class researchers, top-tier chip equipment, and a sophisticated industrial base were present, yet no serious effort existed to design the next generation of silicon to power future data-center AI.
“I remember searching for someone to blame,” he recalls. “Why is nothing happening in Europe? How did we let this happen?” The awkward truth hit quickly: there was no scapegoat. “Who could do something about it? People like me. People I know. So yes, there was a genuine sense of responsibility.”
That sense of duty propelled months of quiet work. Euclyd chose secrecy until there was solid proof the idea could work—proof he and his colleagues could trust. “When money comes from a venture capitalist and the venture fails, that’s risk inherent to the business,” he notes. “But losing friends’ money—that’s a different kind of hurt.”
For months the small team—many veterans from earlier chip ventures—tackled a fundamentally new AI inference architecture. No GPU legacy, no repurposed gaming hardware, no shortcuts. The system was designed from the gate level up. Initial design work even happened in Kastrup’s attic, where a personal simulator and nights spent sketching microarchitectures replaced the traditional whiteboard of ideas.
When the test chip finally appeared and Samsung agreed to manufacture it, Euclyd stepped into the spotlight.
A European chip with world-class ambitions
Euclyd’s promise is immense: ultra-low-energy AI inference that could be orders of magnitude more efficient than today’s Nvidia data-center chips, achieved not by magic but by sound architectural choices.
“Nvidia built for gaming,” Kastrup argues. “Large language models came along at the right time, but treating a neural network as a video game with a global variable space is one of the worst paths to efficiency.” Euclyd’s architecture takes a different route: no reuse of off-the-shelf IP, no generic bus architectures. Everything is specialized, deeply pipelined, and designed for neural inference first. The test chip features 64 processors; the full product will scale to 16,384.
In a field where most players optimize in small increments, Euclyd is attempting a bold reset.
The dream, the reality, and the money
Yet beneath the technical bravado lies a tension few engineers want to admit: money. Euclyd is pursuing a major funding round, and with it comes the reality that the company might no longer be “purely European.”
“The original dream was to do AI in Europe,” Kastrup says softly. “To let Europe compete in this space. But when a business is funded with other people’s money, fiduciary duties demand business-minded decisions that may diverge from the idealistic dream.” It isn’t hypothetical. It’s happening in real time. “We are raising money,” he confirms. “This is a situation we may confront—or have already confronted. That’s part of the game.”
He pauses. “I can dream, yes. I can dream as big as I want. But cannot single-handedly guarantee the dream will come true or stay true. The world is not entirely in my control.”
That candid acknowledgment—rare in Europe’s tech scene, which often clings to idealism—sets the tone. Kastrup wants Euclyd to be Europe’s Nvidia, built on European engineering within the Brainport ecosystem that shaped him. He also wants the company to survive the global race for compute, a race Europe has joined late and unprepared for.
If outside investors come in, and refusing them would jeopardize the company’s competitiveness, what should a founder do?
“I will do the best I can within the boundaries of what is allowed or required by the position,” he says. “But cannot control everything.”
A dream worth fighting for
Whether Euclyd remains European in ownership is uncertain; whether its technology becomes global seems increasingly likely. Whether Europe will finally have its own AI-compute champion may depend not only on Kastrup’s engineering brilliance but also on Europe’s willingness to move at global speed.
Still, the dream persists. “Europe must play in this field,” he asserts. “It’s a wish I share.”
With that, Bernardo Kastrup—philosopher, engineer, founder—returns to his office, where a small test chip, no larger than a fingernail, awaits. The architecture inside could reshape AI’s global energy footprint. The question remains: will Europe be the stage where that future unfolds, or will the world rise first and pull the dream elsewhere?
