Samsung has backed Dutch artificial intelligence chip startup Euclyd in a $231 million funding round, adding another major semiconductor company to a growing wave of investment aimed at challenging Nvidia’s dominance of the AI computing market.
Euclyd, founded in 2024, raised 200 million euros in its Series A round from Somerset Capital Partners, EQT’s Scaleup Europe Fund and Innovation Industries, which co-led the financing alongside Samsung, the company’s chief executive Bernardo Kastrup told CNBC.
The startup is developing an AI computing system built around an architecture different from the graphics processing units that have made Nvidia the dominant supplier of advanced AI chips. Euclyd is initially focusing on inference, the process of running trained AI models to generate responses and perform tasks.
Register for the next Tekedia Mini-MBA.
Register for Tekedia AI in Business Masterclass.
Join Tekedia Capital Syndicate and co-invest in great global startups.
Nvidia became the world’s most valuable company after its GPUs, initially developed primarily for gaming and graphics, became the industry standard for training and running sophisticated AI models. The company has established a near monopoly in the highest-end AI accelerator market, creating an enormous commercial opportunity for competitors seeking to offer alternatives on price, energy consumption or specialized performance.
The competitive landscape is already widening.
Hyperscalers including Google, Amazon Web Services and Meta are developing their own processors for AI workloads, while OpenAI announced in August that its first AI chip, Jalapeño, delivered what it described as “industry-leading speed and efficiency.”
For startups such as Euclyd, the opportunity is not necessarily to replicate Nvidia’s GPU architecture but to question whether the same computing approach will remain optimal as AI workloads evolve.
“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 said.
Euclyd is betting that inference presents an especially attractive opening.
Training a frontier AI model can require enormous amounts of computing power, but once a model is deployed, the number of times it is queried can quickly become much larger. As companies integrate AI into customer service, software development, enterprise search and other applications, the cost and energy required to generate those responses can become a significant part of operating expenses.
That creates room for specialized hardware if it can deliver lower power consumption and lower costs without sacrificing performance.
Euclyd said its silicon systems for foundation models are designed to reduce the energy requirements and costs of AI data-center infrastructure. The company has not yet demonstrated its technology through commercial deployments at scale, leaving a significant gap between its architectural claims and what customers will ultimately pay for.
Samsung’s Bigger Role Beyond the Investment
Samsung’s involvement could nevertheless give Euclyd an advantage that goes beyond capital.
Kastrup said the South Korean technology giant can contribute memory expertise, engineering capabilities and supply-chain relationships as Euclyd moves from chip design toward physical systems.
“Samsung can help us in more ways than money,” Kastrup said. “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.”
Samsung’s involvement is crucial because AI accelerators are not standalone products. Their performance depends heavily on the interaction between processing, memory bandwidth, packaging, networking, and the broader data-center architecture.
The recent AI-chip boom has exposed bottlenecks well beyond processor design, particularly in advanced memory. High-bandwidth memory has become one of the most important components in AI infrastructure, while packaging and manufacturing capacity can determine how quickly promising chip designs reach customers.
Samsung’s position across memory and semiconductor manufacturing therefore potentially gives Euclyd access to capabilities that a young chip designer would struggle to build independently.
The investment also shows that the AI-chip race is becoming less concentrated around the question of who can produce the fastest accelerator.
For years, Nvidia’s advantage was reinforced by the combination of its GPUs, software ecosystem and close relationships with cloud providers and AI developers. Competing directly on raw computing performance is therefore extremely difficult.
A startup with a different architecture can instead target a narrower problem, such as inference efficiency, where customers may care less about peak performance and more about the cost of running millions or billions of AI queries.
Euclyd is pursuing two business models. It plans to sell hardware and complete physical rack systems to enterprises that want to run AI inference on their own infrastructure, particularly where security and control make self-hosting attractive. It also plans to license its intellectual property to companies that want to develop their own processors using Euclyd’s existing technology.
The second model could potentially give the company a way to scale beyond the number of physical systems it can manufacture and deploy itself. But it also introduces a different competitive challenge: customers buying intellectual property will need to believe that Euclyd’s architecture offers enough of an advantage to justify developing around it rather than using established chip platforms.
Euclyd expects to begin rolling out its physical chip systems in 2028 and aims to serve thousands of enterprise customers by 2030, according to Kastrup. Those targets are ambitious given that the company is only two years old and has yet to demonstrate its systems in large-scale commercial deployments.
The funding nevertheless shows how investors are becoming more willing to finance alternatives to Nvidia at a time when demand for AI computing remains exceptionally strong. The opportunity is attracting both startups and the largest technology companies, with hyperscalers developing internal chips to reduce dependence on external suppliers and improve control over the economics of their AI infrastructure.
Nvidia’s position has been built not only on chip performance but also on software, developer adoption and scale. Any challenger must therefore demonstrate more than an attractive processor design. It must prove that customers can deploy the technology reliably, that developers can build for it efficiently and that the total cost of ownership is materially better.
Euclyd’s focus on inference could give it a narrower path into the market. If AI usage continues expanding, inference could become a larger and more persistent infrastructure expense than the current training boom suggests. Lower energy consumption and cheaper computation would then become commercial advantages rather than simply technical specifications.
Samsung’s investment gives that bet additional semiconductor credibility, especially because memory and system engineering are becoming as important to AI infrastructure as the accelerator itself. But the real test will come in 2028 and beyond, when Euclyd’s systems are expected to reach customers. Until then, the company remains an ambitious challenger in a market where Nvidia’s technological and ecosystem advantages are substantial.
While the funding signals investor confidence that the AI-chip market will support multiple architectures, it does not yet prove that Euclyd can displace Nvidia. The more immediate bet is that the next phase of AI computing will reward companies that can make inference cheaper, more energy-efficient, and easier to deploy, rather than simply making another faster GPU.