Abstract:
Euclyd, a Dutch artificial intelligence chip startup, recently completed a round of Series A financing worth 200 million euros, or approximately US$230 million. The financing was co-led by Samsung, Somerset Capital Partners, Scaleup Europe Fund managed by EQT, and Innovation Industries, giving the young company founded in 2024 the financial support needed to further challenge Nvidia’s AI chip market position.
Euclyd did not choose to directly copy the development route of NVIDIA GPU, but tried to find different solutions from the chip architecture level. The company is developing a chip system specifically for AI inference tasks. Its design covers both processor and memory architecture, hoping to improve the efficiency of running artificial intelligence models in a way different from traditional GPUs.
The so-called AI inference refers to the process in which the trained artificial intelligence model performs tasks in practical applications. For example, users ask questions to chatbots, AI generates pictures or videos, and enterprises use large language models to process business requests, etc., all belong to inference workloads. As AI models are increasingly deployed, the computing power required for inference is growing rapidly and has gradually become an important battlefield in the AI chip market.
Nvidia currently has an absolute advantage in this field. Nvidia's GPU, originally designed for the gaming market, later became the core hardware for training and running AI models with its highly parallel computing architecture, and thus established a huge CUDA software ecosystem. Today, a large number of AI companies and cloud computing companies rely on NVIDIA GPUs to run artificial intelligence services. This is also an important reason why NVIDIA can become one of the most valuable technology companies in the world.
Euclyd hopes to use different structures to enter this market. The company not only plans to sell its own AI hardware and complete rack systems, but also plans to allow other companies to use Euclyd's technology to design their own chips by licensing intellectual property rights. In other words, the company’s business model is not limited to being a chip supplier, but also hopes to become an AI chip architecture and technology licensing provider.
Samsung’s participation in this financing is particularly noteworthy. Samsung itself has a complete industrial chain from semiconductor design and wafer manufacturing to memory chips and electronic products. Therefore, investing in an AI chip company trying to challenge Nvidia not only has financial investment significance, but may also provide a basis for future cooperation between the two parties in AI computing, storage and advanced semiconductor technology.
Euclyd has only been established for about two years, but it has already received such large-scale financing, which also reflects the high attention investors attach to the AI inference chip market. As generative AI gradually moves from the model training stage to the large-scale commercial deployment stage, the number of inference tasks that data centers need to handle continues to increase, and the energy efficiency, memory bandwidth and overall computing cost of the chip are becoming increasingly important.
For AI data center operators, continuing to purchase a large number of NVIDIA GPUs can obtain a mature software ecosystem and powerful computing performance, but it also means extremely high hardware, power and infrastructure costs. Therefore, more and more cloud service providers, large technology companies, and AI companies are beginning to look for alternatives to NVIDIA. This also creates opportunities for Euclyd, AMD and other AI accelerator startups to enter the market.
However, challenging Nvidia is not just as simple as making an AI chip with good performance. The real strength of Nvidia is not only the GPU hardware itself, but also the CUDA software platform, development tools, drivers and the developer ecosystem accumulated over many years. Any competitor hoping to truly replace Nvidia in large-scale data center environments will have to address a series of issues including software compatibility, developer migration costs, and long-term deployment risks for customers.
Euclyd is still in a relatively early stage of development, so the more important significance of this 200 million euro financing is to provide sufficient funds for the company to continue to develop chips, improve products, and establish a business ecosystem. Whether it can ultimately achieve significant performance and energy efficiency advantages in actual AI inference tasks with different processors and memory architectures will need to wait until its products enter the actual market to be verified.
As the demand for AI reasoning continues to grow, NVIDIA is facing more and more competitors from different directions. Euclyd chose to start with architectural innovation and memory design, while other companies may look for breakthroughs in dedicated accelerators, custom ASICs, optical computing, and advanced packaging. In the future, the AI chip market is likely to no longer be a pure GPU competition, but a comprehensive competition centered on computing architecture, memory, software ecosystem, energy consumption and overall system cost.
Comments