The Chinese AI startup DeepSeek, which recently made headlines for its allegedly cost-efficient AI model development, may not be as revolutionary as initially portrayed. According to a new report from industry analyst firm SemiAnalysis, the company has invested approximately $1.6 billion in hardware and operates a massive fleet of 50,000 Nvidia Hopper GPUs.
DeepSeek previously claimed that its R1 AI model, which reportedly rivals OpenAI's capabilities, required only $6 million and 2,048 GPUs to train. However, SemiAnalysis's findings paint a different picture of the company's actual infrastructure and investments.
The report details that DeepSeek's GPU fleet includes 10,000 H800s and 10,000 H100s, with additional H20 units. These resources are spread across multiple locations and support various operations, including AI training, research, and financial modeling. Beyond the $1.6 billion hardware investment, the company has reportedly spent around $944 million on operating costs.
DeepSeek emerged from High-Flyer, a Chinese hedge fund with early investments in AI technology. While operating independently since 2023, the company has maintained its self-funded status, allowing for quick decision-making and operational flexibility.
One notable aspect of DeepSeek's operations is its in-house data center management, setting it apart from other AI startups that typically rely on cloud providers. This independence enables direct control over AI model optimization and experimentation.
The company's recruitment strategy focuses exclusively on mainland Chinese talent, particularly from institutions like Peking University and Zhejiang University. SemiAnalysis reports that some AI researchers at DeepSeek earn over $1.3 million annually, exceeding compensation at other Chinese AI firms.
While DeepSeek's $6 million training cost claim generated considerable attention, SemiAnalysis clarifies that this figure only represents the GPU time for pre-training. It excludes substantial costs related to research, model refinement, data processing, and infrastructure maintenance. The total investment in AI development since the company's inception exceeds $500 million.
These findings suggest that DeepSeek's achievements, while impressive, align more closely with traditional industry expectations about the substantial investments required for competitive AI development.