Chinese AI Firm DeepSeek Trains Model for Just $294,000, Disrupting Cost Expectations
A Hangzhou-based AI company has demonstrated that high-performance models can be trained at a fraction of the expected cost. DeepSeek's $294,000 training run—executed on 512 Nvidia H800 chips—challenges the prevailing assumption that foundational AI models require nine-figure investments.
The disclosure, published in Nature, arrives as US firms like OpenAI cite training costs exceeding $100 million. DeepSeek's January product launch already triggered tech sector volatility by undercutting established players. Founder Liang Wenfeng maintains a low profile, but the cost differential signals China's growing competitiveness in constrained hardware environments.