In its report, DeepSeek stated that the total training cost amounted to $5.576 million, primarily due to the rental price of Nvidia's graphics processing units. The company also cautioned that this figure only reflects "formal training" and does not include costs related to prior research or experiments on architecture, algorithms, and data. This immediately attracted investor attention and caused global technology company stocks to plummet by $1 trillion on January 27th alone.
DeepSeek's operating costs are much higher than they claim?
Now, a report from semiconductor research and consulting firm SemiAnalysis offers a different perspective on DeepSeek's development costs. The company estimates DeepSeek's hardware spending to be significantly higher than previously stated, while also highlighting substantial research and development (R&D) costs and total cost of ownership.
SemiAnalysis states that creating "synthetic data" for training models will require "significant computational effort." Furthermore, companies will also need to test and develop new architectures, collect and clean data, and pay employee salaries.
Does DeepSeek need 50,000 GPUs instead of just over 2,000?
According to calculations, experts say DeepSeek operates a large-scale computing system comprising approximately 50,000 Hopper GPUs, with 10,000 H800 GPUs and 10,000 more powerful H100 GPUs, along with additional H20 GPUs. This contradicts DeepSeek's previous claim of using only 2,048 Nvidia GPUs. The total capital cost for the servers is estimated at around $1.6 billion, while operating costs amount to $944 million.
SemiAnalysis believes that DeepSeek has used 50,000 Nvidia GPUs.
DeepSeek was founded in 2023 by Liang Wenfeng, co-founder of High-Flyer, an AI-focused hedge fund in China. The startup evolved from the fund's AI research unit in April 2023, aiming to develop large-scale language models and achieve general artificial intelligence (AGI).
Interest in DeepSeek increased when the company released R1, a reasoning model that competes with OpenAI's o1, but is offered as open source, allowing any AI developer to use it. However, like many other Chinese chatbots, DeepSeek also has limitations on certain topics.
OpenAI CEO Sam Altman praised DeepSeek's model, emphasizing that it is "clearly a great model." However, he also stated that there is evidence suggesting DeepSeek collected OpenAI's data to operate using a "distillation" method.
Source: https://thanhnien.vn/chi-phi-phat-trien-deepseek-bi-tang-boc-185250203151508438.htm
Comment (0)