DeepSeek's approach, which involves building and training models for under $6 million, contrasts sharply with the billions spent by US companies on acquiring advanced Nvidia H100 chips. Due to US sanctions, China cannot import these flagship GPUs and instead uses the less advanced H800s. The success of DeepSeek highlights the potential for smaller players to compete with tech giants, potentially undermining US efforts to maintain a competitive edge in AI technology. Meanwhile, OpenAI's ChatGPT ranks second on the Apple chart, following a significant investment in AI infrastructure in collaboration with the US government.
Key takeaways:
- DeepSeek, a Chinese AI startup, released its flagship model R1 on January 20, surprising competitors in Silicon Valley and the AI world.
- DeepSeek's models, built with inferior Nvidia chips and less capital, matched or surpassed capabilities of models from OpenAI, Google, and Meta.
- DeepSeek's V3 model was built and trained for under $6 million using about 2,000 Nvidia H800 chips, while US companies invest heavily in more advanced H100 chips.
- The success of DeepSeek challenges US efforts to maintain a competitive edge in AI development with significantly fewer resources.