Sign up to save tools and stay up to date with the latest in AI
bg
bg
1

Were DeepSeek's Development Costs Much Higher Than Reported? - Slashdot

Feb 01, 2025 - slashdot.org
Nearly three years ago, Chinese AI engineers from DeepSeek's parent company unveiled a supercomputer built with 10,000 Nvidia A100 GPUs. This project cost $139 million, but U.S. restrictions later banned Nvidia from selling more A100s to China. Recently, DeepSeek's new chatbot caused a $1 trillion stock market panic, claiming it was trained on just $5.6 million in computing costs using 2,000 Nvidia H800 chips. However, experts argue that DeepSeek's parent company has spent over half a billion dollars on GPUs, with total expenditures reaching $1.3 billion, suggesting the project was much costlier than initially claimed.

Despite skepticism about DeepSeek's cost and energy-efficiency claims, the company's competitive pricing and model performance are causing concern among U.S. companies. Experts believe DeepSeek's strategy could disrupt the global market. China's supportive AI policies, including a 2017 blueprint to become a global AI leader by 2030, have fostered an environment for companies like DeepSeek to thrive. This has led to increased funding and support for AI development from both national and local governments.

Key takeaways:

  • DeepSeek's new chatbot caused significant disruption in the stock market, erasing $1 trillion in value, largely due to claims of low training costs and efficient use of Nvidia's H800 chips.
  • Despite DeepSeek's claims of spending only $5.6 million on training, experts suggest the actual costs are much higher, with total capital expenditures estimated at almost $1.3 billion.
  • DeepSeek's pricing strategy and model performance are competitive, posing a potential threat to U.S. AI companies and possibly disrupting the global market.
  • China's supportive AI policies and funding initiatives have created an environment that fosters the growth of companies like DeepSeek, aligning with its goal to become a global AI leader by 2030.
View Full Article

Comments (0)

Be the first to comment!