South Korean artificial intelligence chip startup FuriosaAI said Wednesday it will supply more than 8,800 of its RNGD inference accelerators for a new artificial intelligence data center being developed in Stockholm.
The 15-megawatt project is being developed with US AI infrastructure company I/ONX HPC and data center developer and operator Velox. A 2 MW first phase is scheduled to begin operations in early 2027, with another 8 MW to be added later that year. The site will eventually be expanded to 15 MW.
"Power efficiency, total cost of ownership and sustainability have become key competitive factors for AI data centers," FuriosaAI CEO Baek Jun-ho said. "This project marks an important step in deploying our AI inference infrastructure at a large-scale European data center."
FuriosaAI said it will initially provide 1,800 RNGD (Renegade) chips, with total deployment expected to exceed 8,800 units as the facility expands.
The chips will be installed in I/ONX HPC’s Symphony SixtyFour high-density AI servers and operate alongside graphics processing units in a heterogeneous computing system.
RNGD accelerators will handle large language model and agentic AI inference workloads, while GPUs will be used for model training and other computing tasks, the startup said. The system is designed to improve both performance and power efficiency across the data center.
FuriosaAI will provide the chips and software stack, while I/ONX HPC will handle system integration and the AI platform. Velox will develop and operate the facility.
The data center will provide AI inference services to companies and cloud service providers across Europe, according to FuriosaAI.
The project follows the launch of an RNGD-based NPU-as-a-service offering by Samsung SDS and the development of an AI inference hub using FuriosaAI chips at an Equinix data center in Lisbon.
herim@heraldcorp.com