#ai-model-training

#ai-model-training

Chinese AI firm DeepSeek revealed it spent only $294,000 training its R1 model far below the hundreds of millions claimed by U.S. rivals. Using 512 Nvidia H800 accelerators, the company trained R1 in just 80 hours. The release of R1 earlier this year rattled tech markets, even denting Nvidia's valuation. DeepSeek also acknowledged limited use of A100s and defended model distillation, stressing it makes AI more accessible despite U.S. accusations of copying OpenAI's work.