NPUs Threaten GPU Inference Hegemony

Furiosa’s RNGD NPU is entering mass production and claims similar inference performance to advanced Nvidia GPUs at much lower energy use; large tech firms (Meta, OpenAI, LG) are already testing or courting the startup. If true at scale, NPUs could drive a shift in who supplies inference compute, change datacenter energy profiles, and alter bargaining power in the AI stack. — A credible move from GPUs to energy‑efficient, specialized NPUs would lower deployment costs, reshape supply chains and vendor power, and force new industrial, antitrust and energy policy responses.