According to a recent report from an independent research firm, the actual operating costs of Chinese tech company DeepSeek are far higher than initially claimed — by a lot.
DeepSeek said its expenses were under $6 million and required only 2,048 GPUs, with its R1 model rivaling OpenAI’s o1. However, industry research firm SemiAnalysis disputes this, estimating the company has invested over $1 billion in server infrastructure.
DeepSeek operates a massive computing network with roughly 50,000 Hopper GPUs, including 10,000 H800 units, 10,000 H100 units, and additional H20 chip purchases. These resources support AI training, research, and financial modeling across multiple locations.
SemiAnalysis estimates the company’s total server investment at $1.6 billion, with operating costs around $944 million — far greater than the initially reported $6 million.
GAI President and bestselling author Peter Schweizer made similar claims during a recent episode of The Drill Down. “What’s interesting to me is the timing of this announcement from this Chinese company of little note occurs right in the midst of a couple of things,” Schweizer said.
“Number one: the announcement of Stargate. The United States is going to spend $500 billion dollars, and then China comes out and says, ‘Oh, we just did it for six million,” Schweizer said.
“The reason to be skeptical is China has a long history of lying about all kinds of things. They lied, of course, about the COVID virus. Anybody who trades stocks or invests in China will tell you that you cannot really rely on financial data from that country because they make stuff up all the time. The government statistics on the growth and the gross national product — they lie about that.”
Watch the clip below. Full episode on DeepSeek above.
