Elon Musk, CEO of Tesla, raised doubts about the Chinese AI startup DeepSeek, which claimed to have developed an artificial intelligence (AI) model at a low cost.
On the 27th (local time), Musk shared a post on X (formerly Twitter) stating that DeepSeek holds more of NVIDIA's 'H100' than previously revealed.
The post included a statement from Alexander Wang, CEO of ScaleAI, noting that 'DeepSeek has applied approximately 50,000 NVIDIA H100 chips, but they cannot comment on it due to U.S. export controls.' A video of Wang's recent interview with CNBC was also included.
In response, Musk commented, 'Obviously.'
Previously, DeepSeek claimed to have developed the open-source AI model 'DeepSeek-R1' in two months for an expense of less than $6 million. It stated that it used over 2,000 of NVIDIA's low-cost chips, 'H800.' The H800 is a product made for export to China to evade U.S. government export regulations, with lower performance and much cheaper price.
Musk also commented on a post that critically analyzed DeepSeek's announcement regarding AI model development costs, saying, 'Interesting analysis. The best I have seen so far.'
Gavin Baker, chief investment officer of Atreides Management, wrote on X, 'According to DeepSeek's technical document, the $6 million does not include 'expenses related to previous research and experiments on the architecture, algorithms, and data,' which suggests that if they had already spent hundreds of millions on prior research in the lab and can access a much larger (chip) cluster, it means they could train an R1 quality model for just $6 million.' He added, 'DeepSeek clearly has more than the H800.'
Meanwhile, Musk co-founded OpenAI with Sam Altman, former CEO of OpenAI, but resigned from the board in 2018 and disposed of all his equity. After that, he founded his own AI startup, xAI, in 2023 and has continued to make substantial investments.