FuriosaAI introduces its Nvidia-comparable chips in Europe.

      A South Korean chip startup is aiming to offer Europe a more efficient and affordable alternative to Nvidia. It has recently activated its “renegade” accelerators at a datacentre in Lisbon.

      FuriosaAI has launched its RNGD AI accelerators in Europe, announcing on Tuesday that it is setting up RNGD servers at Equinix’s LS2 datacentre in Lisbon. The name “RNGD” is pronounced as “renegade.”

      This deployment in Lisbon builds upon Furiosa's existing presence in the city, which includes a compiler-focused research and development lab and a new flagship office. The timing is strategic, coinciding with the RAISE Summit in Paris, as European businesses seek efficient AI computing solutions closer to home.

      The efficiency aspect is a key selling point. RNGD accelerators are designed for power efficiency rather than peak speed. Each accelerator employs a 5nm Tensor Contraction Processor design capable of delivering 512 teraFLOPS of FP8 performance while maintaining a strict 180-watt thermal profile. Eight of these accelerators make up the NXT RNGD Server, a 3kW system. Furiosa promotes this setup as a compact, air-cooled inference engine that fits into standard racks without the need for liquid cooling or modifications.

      The efficiency advantage is central to the pitch; The Register points out that Nvidia’s closest competitor, the RTX Pro 6000, offers double the memory and computing capability but consumes over three times the power. “We unlock the ability for enterprises to run inference sustainably and reliably,” stated June Paik, co-founder and CEO.

      The rationale for targeting Europe at this moment involves not just sales but also influence. Europe is eager to establish its own AI infrastructure and lessen its dependence on American chip technology. Additionally, the European chip sector is concerned about its future, making a power-efficient component that fits existing racks an appealing option, especially as energy costs rise.

      Furiosa is also planning for future developments; it is collaborating with Broadcom on a third-generation accelerator aimed at handling frontier models with a trillion parameters or more, utilizing faster HBM4 memory for large-scale inference. The RNGD is already in mass production, manufactured using TSMC’s processes and SK hynix memory. Furiosa has reported raising over $250 million to date.

      While Furiosa is unlikely to replace Nvidia, its chips are smaller and slower. The next generation of chips will also utilize HBM4 memory, which is just beginning to reach the market. However, that is not the primary objective. Some European clients are simply looking for a more affordable chip that operates quietly and can be sourced outside the U.S. Furiosa is entering a competitive arena of Nvidia challengers that are all aiming to demonstrate that there is space for multiple players in the AI computing market.