Samsung Bets on New AI Chip Rival to Cut Data Center Costs

Samsung has joined a $231 million funding round for Euclyd, a Dutch company developing an alternative to the graphics processing units that currently power most artificial intelligence systems. The investment signals a growing appetite for hardware that promises to lower the energy costs and infrastructure burdens associated with running large-scale AI models.

Nvidia has held a near-monopoly on high-end AI chips for years, but major technology firms are increasingly seeking independence from this single supplier. By backing Euclyd, which was founded in 2024, Samsung is positioning itself at the forefront of a shift toward specialized silicon designed specifically for inference, the process of using trained AI models.

Challenging the GPU monopoly

Euclyd’s approach differs from traditional graphics cards by using a distinct architecture tailored for inference tasks. While Nvidia’s chips were originally designed for gaming and later repurposed for AI, Euclyd is building systems from the ground up for this specific workload. The company argues that this fundamental change in infrastructure is necessary to unlock the full potential of AI for economic growth and scientific discovery.

This move aligns with trends at other major players. OpenAI recently announced its first custom chip, and giants like Google, Amazon, and Meta are all developing proprietary processors. The common goal is to reduce reliance on external suppliers and optimize performance for specific AI tasks, a strategy that Euclyd’s CEO Bernardo Kastrup describes as essential for reducing the constrained potential of current infrastructure.

Why Samsung invested beyond money

For Samsung, the investment in Euclyd is about more than just financial gain. As one of the world's largest memory manufacturers, Samsung possesses deep expertise in systems engineering and supply chain management. Kastrup noted that Samsung’s industrial network and technical knowledge provide critical support that goes far beyond the capital itself, helping the startup navigate the complex landscape of hardware production.

The path to commercial rollout

Despite the significant funding from Somerset Capital Partners, EQT, and others, Euclyd is still in the early stages of proving its technology at scale. The company plans to begin rolling out its physical chip systems in 2028, with a target of serving thousands of enterprise customers by 2030. Their business model includes selling hardware to companies seeking secure, self-hosted AI, as well as licensing their intellectual property to other manufacturers.

The main challenge remains whether these new systems can deliver on their promise of reduced energy needs in real-world data centers. If successful, Euclyd could help alleviate the massive power consumption issues currently plaguing the AI industry, but that proof of performance is still years away. As reported by GN technics/ai (en-US), the stakes are high, with the entire future efficiency of AI infrastructure hanging on these new architectural bets.