A new report by SemiAnalysis has estimated that DeepSeek’s hardware spend is “well higher than $500M”, significantly more than its training costs of almost $5.6 million. The model has been gaining attention for its capabilities in large language models and artificial general intelligence, surpassing OpenAI’s ChatGPT in the US market.
DeepSeek’s AI Assistant took the top spot as the most-downloaded free app in the US on Apple’s App Store, dethroning ChatGPT. However, experts and analysts have raised concerns about the model’s limitations, particularly when discussing sensitive topics such as Chinese leader Xi Jinping’s policies.
Despite these concerns, OpenAI CEO Sam Altman has praised DeepSeek’s model, saying it is “clearly a great model”. The report from SemiAnalysis adds context to DeepSeek’s expenses, highlighting significant R&D costs and total cost of ownership. The company’s unique approach in achieving this level of cost and capabilities first is seen as impressive by experts.
DeepSeek was founded in 2023 by Liang Wenfeng, co-founder of High-Flyer, a quantitative hedge fund focused on AI. The startup has grown rapidly, with its AI model being open-source, allowing any AI developer to use it. However, the company’s relationship with OpenAI remains uncertain, with allegations that DeepSeek improperly harvested OpenAI data.
The report from SemiAnalysis also notes that Anthropic’s Claude 3.5 Sonnet costs significantly more to train, at “tens of millions of dollars”, highlighting the significant investment required to run these models. As experts weigh in on DeepSeek’s capabilities, it remains to be seen how this will impact the AI market, predicted to reach $1 trillion in revenue within a decade.
Source: https://www.cnbc.com/2025/01/31/deepseeks-hardware-spend-could-be-as-high-as-500-million-report.html