FuriosaAI unveiled its NXT RNGD Server, a turnkey solution for AI inference targeting data center deployment. Built around the company’s RNGD accelerators, the system aims to deliver high performance on major AI workloads without requiring exotic infrastructure. It ships with the Furiosa SDK and LLM runtime preinstalled, designed for immediate use upon installation.

The server operates on just 3 kW, fitting within the power and cooling limits of most modern data centers. This efficiency allows enterprises to scale AI inside existing facilities rather than undertaking costly retrofits. By optimizing for standard PCIe interconnects, Furiosa eliminates the need for proprietary fabrics, making deployment more accessible for organizations with established infrastructure.

Furiosa claims the NXT server offers 3.5x better efficiency compared to Nvidia's H100 GPUs. LG AI Research has already adopted RNGD accelerators for its EXAONE models, reporting strong token generation speeds. With orders starting for January 2026 delivery, Furiosa is positioning itself as a viable alternative in the competitive AI hardware market.