China's DeepSeek that 'shocked' America and US technology companies reveals actual cost of training AI model

Chinese AI developer DeepSeek has revealed the cost of training its R1 model, a figure that is significantly lower than those reported by its US rivals. After launch, the DeepSeek R1 model stood out for its cost-efficient development and its open-source nature, offering unlimited free usage to a wide audience.

Its emergence as a direct competitor to models from industry leaders such as OpenAI, Google, and Anthropic has brought new scrutiny to the financial investments required for AI development.According to a report by Reuters, DeepSeek revealed that the company spent a significantly low amount—just $294,000—to train its R1 model. This figure was published in a peer-reviewed article in the academic journal Nature.The company’s disclosure may reignite a debate in the tech industry over why US giants are spending hundreds of millions of dollars to train large language models (LLMs).

Why DeepSeek and its model made waves in US tech industry

DeepSeek's release of what it claimed were lower-cost AI systems in January caused investors to sell off tech stocks, as they worried the new models could threaten the dominance of companies like Nvidia. The update, which is the first cost estimate the Hangzhou-based company has released for its R1 model, appeared in a paper co-authored by its founder, Liang Wenfeng. The paper specified that the reasoning-focused model was trained using 512 Nvidia H800 chips.

By comparison, OpenAI CEO Sam Altman said in 2023 that "foundational model training" had cost "much more" than $100 million, though his company has not provided detailed figures, Reuters reported.Some of DeepSeek's previous claims about its development costs and technology have been questioned by US companies and officials. The company’s use of H800 chips comes after the US in October 2022 made it illegal for Nvidia to export its more powerful H100 and A100 AI chips to China.In a supplementary document to the Nature article, DeepSeek acknowledged for the first time that it does own A100 chips and used them in the “preparatory stages of development” for R1. US officials claimed in June that DeepSeek had access to “large volumes” of H100 chips procured after export controls were in place, a claim Nvidia has denied, stating that DeepSeek used lawfully acquired H800 chips.