South Korean startup FuriosaAI has deployed its RNGD AI accelerators at Equinix's LS2 datacentre in Lisbon, marking its first entry into the European market. The RNGD chip uses a 5nm Tensor Contraction Processor design, delivering 512 teraFLOPS of FP8 compute within a 180W thermal envelope — roughly one-third the power draw of Nvidia's nearest comparable card. Eight units form a 3kW air-cooled server that fits standard racks without liquid cooling. The European push targets enterprises seeking efficient, locally sourced AI compute amid rising energy costs and growing desire to reduce dependence on American silicon. FuriosaAI, which has raised over $250M, is also developing a third-generation chip with Broadcom targeting trillion-parameter models using HBM4 memory.
