FuriosaAI introduces its Nvidia competitor chips to Europe.

      A South Korean chip startup aims to offer Europe a cooler and more affordable alternative to Nvidia. Its first "renegade" accelerators have just been activated in a Lisbon data center.

      FuriosaAI has launched its RNGD AI accelerators in Europe. The company announced on Tuesday that it is installing RNGD servers at Equinix's LS2 data center in Lisbon. The name is pronounced "renegade."

      This deployment in Lisbon builds on the presence that Furiosa already has in the city, where it operates a compiler-centric R&D lab and a new flagship office. The timing is strategic, coinciding with the RAISE Summit in Paris, and it arrives as European companies seek efficient AI computing options close to home.

      The efficiency proposition

      RNGD is designed for power efficiency rather than peak speed. Each accelerator utilizes a 5nm Tensor Contraction Processor architecture, providing 512 teraFLOPS of FP8 performance while adhering to a strict 180-watt thermal profile. Eight of these units combine to form the NXT RNGD Server, a 3kW system. Furiosa markets it as a compact, air-cooled inference engine that fits into standard racks, with no need for liquid cooling or modifications.

      The efficiency gap is a key selling point. The Register highlights that Nvidia's closest competitor, the RTX Pro 6000, provides double the memory and processing power but consumes over three times the energy. “We enable enterprises to conduct inference in a sustainable and reliable manner,” stated co-founder and CEO June Paik.

      Why Europe, why now

      The initiative is as much about gaining recognition as it is about generating sales. Europe is eager to establish its own AI infrastructure and reduce dependency on American silicon. Meanwhile, the region's chip industry is concerned about its future. A component that consumes less power and fits into existing racks is more appealing amid rising energy costs.

      Furiosa is also planning for the future, collaborating with Broadcom on a third-generation accelerator designed for frontier models with one trillion or more parameters. This chip will utilize faster HBM4 memory for hyperscale inference. RNGD is already in mass production, manufactured using TSMC's process with SK hynix memory. Furiosa reports it has raised over $250 million to date.

      Why it matters

      While Furiosa is unlikely to displace Nvidia, as its products are smaller and slower, this is not the company's goal. Certain European customers simply desire a chip that is cost-effective, operates quietly, and is sourced outside the United States. Furiosa joins a competitive landscape of Nvidia challengers, all striving to demonstrate that there is room for more than one player in AI computing.