US AI chip developer Groq has opened its first European data center in Helsinki, Finland. This marks a step forward in the company’s international expansion. It also responds to the growing demand for fast and scalable AI inferencing in Europe.
The choice of Helsinki is strategic. The region has a reliable energy infrastructure, cool climatic conditions, and complies with European data management regulations. The Nordic region has long been attractive for data center construction due to its combination of environmentally friendly energy sources and favorable temperature conditions for cooling.
The new facility has been set up in collaboration with data center provider Equinix. This partnership will enable Groq to make its inferencing capacity available to customers in Europe, the United States, and the Middle East via the existing Equinix network. Thanks to this integration, organizations can run AI workloads with minimal latency and access to public, private, or sovereign infrastructure, depending on their needs.
Crucial role for LPU
Groq is known for its specialization in AI inferencing. This is the process whereby a pre-trained model analyzes new data and generates a result immediately. Unlike training AI models, which requires a lot of computing power and time, Groq focuses on fast and efficient execution of existing models. Central to this approach is the proprietary Language Processing Unit (LPU). This chip is optimized for inferencing and uses Tensor Streaming Processors (TSPs), which make the computing process more efficient and direct than traditional GPU-based solutions.
The LPU recently demonstrated impressive performance during a demonstration in which it controlled Meta’s Llama 2 model. In a speed test, the system performed significantly faster than ChatGPT, even though the LLM used was smaller than the GPT-4 Turbo model. The demonstration shows that Groq’s technology has the potential to simplify the hardware requirements for AI applications.
With a market value of $2.8 billion, Groq is positioning itself as a challenger to established names such as Nvidia.