Frontier Model Trained for $294k

DeepSeek reports its R1 reasoning model cost just $294,000 to train using 512 Nvidia H800 GPUs, according to a peer‑reviewed Nature article. That’s orders of magnitude below public figures mentioned by U.S. labs for foundational training. If accurate, the barrier to training competitive models is falling fast. — Lower training costs could broaden who can build powerful AI, reshaping competition, export‑control strategy, and safety governance.