Samsung's decision to co-lead a €200 million (approximately $231 million) funding round for Dutch startup Euclyd, while simultaneously manufacturing its custom AI inference chip, signals that the world's largest memory suppliers are no longer betting exclusively on Nvidia's GPU-plus-memory model for AI data centers. The investment represents one of Europe's largest semiconductor raises and marks a turning point in how the industry approaches the fundamental challenge of powering artificial intelligence at scale.

Why Is Power Efficiency Becoming the Real Battleground in AI?

The reason Euclyd's technology matters comes down to a physics problem that has plagued AI data centers for years. Nvidia's latest Vera Rubin chips, which entered full production in 2026, pack 288 gigabytes of high-bandwidth memory and deliver approximately 22 terabytes per second of memory bandwidth per GPU. That sounds impressive until you look at the power bill. A single rack housing 72 Vera Rubin chips draws between 190 and 230 kilowatts of power. The culprit is not the computing itself, but the movement of data. In inference workloads, where trained models answer user queries, moving data between memory and processors can consume 60 to 80 percent of a system's entire power budget.

Euclyd's CRAFTWERK chip takes a fundamentally different approach. Instead of pairing a large compute processor with separate memory stacks, the system integrates 16,384 custom processors directly with what the company calls Ultra Bandwidth Memory (UBM), a custom memory architecture built specifically for this chip rather than adapted from a standard module. The entire system fits in a palm-sized package measuring roughly 100 by 100 millimeters. The goal is simple but radical: reduce the physical distance data must travel, and therefore the energy required to move it.

What Do Euclyd's Claimed Specifications Actually Mean for Data Centers?

Euclyd projects that its CRAFTWERK STATION CWS 32, a rack-scale system housing 32 chips, would deliver aggregate bandwidth of 8,000 terabytes per second, compute performance exceeding one exaflop (one quintillion floating-point operations per second), 32 terabytes of on-system memory, and a power draw of approximately 125 kilowatts. Running Meta's Llama 4 Maverick language model, the company claims the system could generate 7.68 million tokens per second. The headline figure, however, is the efficiency claim: roughly 100 times greater power efficiency per generated token than Nvidia's Vera Rubin chips on comparable large language model workloads.

It is important to note that every one of these numbers comes from Euclyd's own modeling and projections. Independent testing at commercial scale has not yet validated these claims. As one industry observer noted, "a term sheet for a chip company is not the same thing as a shipped device, a qualified manufacturing flow or a customer-validated benchmark".

Why Is Samsung Playing Both Investor and Manufacturer?

The most significant detail in this funding round is that Samsung is simultaneously a co-lead investor in Euclyd's Series A and the company manufacturing CRAFTWERK's first silicon in South Korea. In the semiconductor industry, this is unusual. Manufacturing partners typically hold enormous leverage over a startup's intellectual property, production schedule, and unit economics, which creates a conflict of interest with being a financial investor. Samsung's willingness to wear both hats signals something deeper than confidence in the team.

Samsung is one of only three suppliers, alongside SK Hynix and Micron, certified by Nvidia to produce HBM4 (High Bandwidth Memory, fourth generation) for the Vera Rubin platform. By simultaneously backing a non-HBM inference architecture with its own capital and manufacturing capability, Samsung is hedging against the possibility that the AI inference market does not converge entirely on Nvidia's GPU-plus-HBM model. If Euclyd's UBM architecture works at scale, Samsung would be positioned as the sole supplier of both the memory technology and the manufacturing process for a serious Nvidia alternative.

"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," said Bernardo Kastrup, Euclyd's founder and CEO.

Bernardo Kastrup, Founder and CEO at Euclyd

How to Understand the Broader Implications of This Funding Round

• European AI Sovereignty: The Scaleup Europe Fund, formally established by the European Commission on August 4, 2026, with €5 billion in target capital, made Euclyd its second major investment in its opening weeks and its first in AI silicon specifically. This represents institutional commitment to reducing European dependence on US chip manufacturers.
• Leadership Credibility: Peter Wennink, who ran chip-equipment giant ASML for eleven years and oversaw its ascent into the most strategically significant company in global semiconductor manufacturing, joined Euclyd's board as Chairman. His involvement signals that this is not a speculative venture but a serious attempt to reshape AI infrastructure.
• Investor Diversity: The round was co-led by Samsung, Somerset Capital Partners, the EU's Scaleup Europe Fund managed by EQT, and Innovation Industries. This mix of corporate, government, and private capital suggests confidence that spans multiple constituencies and geographies.

Bernardo Kastrup's path to founding Euclyd is itself noteworthy. He holds two PhDs, one in computer engineering with specialization in reconfigurable computing and AI, and another in philosophy focusing on ontology and philosophy of mind. Before founding Euclyd, he co-founded Silicon Hive, a parallel processor company that Intel acquired in 2011, and worked at ASML as a technology strategist. He began developing CRAFTWERK's architecture in his home attic in Eindhoven, working late nights after his day job.

"AI is becoming a foundation of economic growth, scientific discovery, and national competitiveness, but its potential will remain constrained unless we fundamentally change the infrastructure beneath it," Kastrup stated in the Series A announcement.

Bernardo Kastrup, Founder and CEO at Euclyd

The real test for Euclyd will come when CRAFTWERK moves from modeling and projections into production silicon and real-world customer validation. The company claims to have solved one of the most fundamental constraints in AI infrastructure, but the semiconductor industry has learned to be skeptical of efficiency claims until they survive contact with actual workloads. Samsung's dual role as investor and manufacturer suggests the company is confident enough to bet its own capital and manufacturing capacity on the outcome. Whether that confidence is justified will become clear as CRAFTWERK chips begin shipping to customers and independent benchmarks emerge.