South Korean AI chip startup FuriosaAI, founded in 2017 by CEO June Paik, is developing inference processors designed to challenge Nvidia's GPU dominance by prioritising power efficiency and architectural innovation. Its current chip, RNGD, uses a proprietary Tensor Contraction Processor architecture and operates at 180 watts — compared to 600 or more for comparable GPUs — and has been adopted by enterprise customers including LG AI Research. In an interview, Paik argued that the AI data centers of 2036 will not be GPU-dominated, predicting a shift toward diverse, task-specific silicon driven by the unsustainable power demands of existing GPU architectures.

“The AI data centers of 2036 won’t be filled with GPUs”: FuriosaAI’s CEO on the future of silicon - TechRadar