The analyst estimated Depseek’s real cost at $ 500 million

“Depseek spent more than $ 500 million in graphic processors in the history of the company. Although the learning process [нейросети] It was very effective, required significant experiments and tests to achieve the desired results, ”said Patel.

According to the analyst, Depseek and the main investor of the company, the Chinese Main Fund High-Flyer, have access to tens of thousands of Nvidia graphic processors, which were used to teach AI models that preceded the last Deepseek-R1 and Deepseek – V3.

Deepseek launched the V3 model in December, saying that only $ 5.6 million were spent on development, and a little more than 2 thousand Nvidia chips were used to study, which the company could obtain without violating US export restrictions. Last week, Depseek presented its new R1 “thought” model, which compares with OPENAI’s O1 model.

The appearance of Deepseek led to a massive sale of western companies, which caused investors to doubt the convenience of multimillion -dollar investments in the creation of advanced AI models. As a result, the combined state of the five hundred richest people in the world decreased by $ 108 billion, and Nvidia lost almost $ 600 billion of capitalization.

The president of the United States, Donald Trump, said he considers the advance of Depseek “positive, because now he is not obliged to spend such large amounts” on the creation of neuronal networks.