Chinese AI startup DeepSeek has become a focal point in the AI industry after claiming it developed its groundbreaking models on a modest $6 million budget. However, experts now suggest the true cost of development far exceeds this figure, pointing to significant investments in earlier stages of the process.
DeepSeek, whose AI assistant recently surpassed OpenAI’s ChatGPT on Apple’s App Store in the U.S., reported spending $6 million on the final training run of its V3 model using Nvidia’s H800 chips. While the company’s achievement has drawn widespread praise for its efficiency and innovation, industry leaders argue that the actual development costs—factoring in research, design, and initial chip usage—could have reached billions of dollars.
The $6 million figure refers only to the final training phase. The earlier stages, which involve determining model architecture and conducting experimental runs, likely required a much larger investment, said an executive at a leading AI lab who spoke on condition of anonymity to Reuters.
DeepSeek’s training process reportedly utilized 2,048 H800 chips, designed to comply with U.S. export controls. Experts argue that these chips, while less powerful than their unrestricted counterparts, were still deployed in sufficient numbers to incur substantial costs during the development cycle.
DeepSeek’s rise has not only triggered excitement but also raised competition concerns within the U.S. tech sector. Industry giants like OpenAI and Nvidia have acknowledged the startup’s innovation, with OpenAI CEO Sam Altman praising its cost-effective model. However, insiders caution that DeepSeek’s success should be viewed in the context of its broader, potentially higher-cost development efforts.
As DeepSeek continues to disrupt the AI landscape, analysts predict that the startup’s claims of affordability will face increasing scrutiny, especially as global competition in AI intensifies.
Chinese AI startup DeepSeek has become a focal point in the AI industry after claiming it developed its groundbreaking models on a modest $6 million budget. However, experts now suggest the true cost of development far exceeds this figure, pointing to significant investments in earlier stages of the process.
DeepSeek, whose AI assistant recently surpassed OpenAI’s ChatGPT on Apple’s App Store in the U.S., reported spending $6 million on the final training run of its V3 model using Nvidia’s H800 chips. While the company’s achievement has drawn widespread praise for its efficiency and innovation, industry leaders argue that the actual development costs—factoring in research, design, and initial chip usage—could have reached billions of dollars.
“The $6 million figure refers only to the final training phase. The earlier stages, which involve determining model architecture and conducting experimental runs, likely required a much larger investment,” said an executive at a leading AI lab who spoke on condition of anonymity.
DeepSeek’s training process reportedly utilized 2,048 H800 chips, designed to comply with U.S. export controls. Experts argue that these chips, while less powerful than their unrestricted counterparts, were still deployed in sufficient numbers to incur substantial costs during the development cycle.
DeepSeek’s rise has not only triggered excitement but also raised competition concerns within the U.S. tech sector. Industry giants like OpenAI and Nvidia have acknowledged the startup’s innovation, with OpenAI CEO Sam Altman praising its cost-effective model. However, insiders caution that DeepSeek’s success should be viewed in the context of its broader, potentially higher-cost development efforts.
As DeepSeek continues to disrupt the AI landscape, analysts predict that the startup’s claims of affordability will face increasing scrutiny, especially as global competition in AI intensifies.