Deepseeek’s hardware expense could be as high as $ 500 million: Report

Chinese Deepsek was the largest technology topic this week, industry and many walls focused on a single number: $ 6 million.

Deepseek paper About the most recent model of artificial intelligence, the company said that his training costs were $ 5.576 million, based on the price of rent Nvidia’s Graphics processing units. Deepsees had a clean Covate, saying that numbers are only included in the “official training” model, and exclude the costs related to “architecture, algorithms or data” in preliminary research and ablation experiments.

At the beginning of the week, Deepsees took a precious place to get the AI ​​assistant free downloaded application AppleApp Store, Dethhroning Openai’s chat. Glogo Tech Stocks for sale, Nvidia and Moreomom disappear Combined $ 800 million in Monday.

A New semianalisia reportSemiconductor research and consulting firm, added more context to Deepseeek’s expenses. The company calculated that Deepseeek’s Hardware Expenditure “is greater than $ 500 million above the company’s history.” The creation of “synthetic data” training should be “a large number of computers”, “semianalisian wrote.

The report said Claude 3.5 Sonneta is “training millions of $ 10 million”, but he said he raised a million dollars in a million Gym and GoogleIndicates how much more money to run the model and business.

“It’s because they have to experiment with new architecture, collecting data and cleaning, paying workers and much more,” Semianalisians said.

Deepsees its own paper does not receive an estimate of its calculation costs. The company did not respond immediately to a request for comment.

“It is unique to clean it unique when they have achieved this level of cost and abilities,” Semianalyst wrote. The company has added that Deepseeek’s R1 “is a very good model” and “quickly capture the rush to the edge is rapidly impressive.”

Experts and analysts this week took Deepse’s quality, and it is impressive what impressive is what it examines in the US Chip-covered exports three times in China in three years. This worried that the US falls behind the main quotation of the market Planned $ 1 trillion in the entries of a decade.

Bernstein analysts wrote on Monday, “this weekend) (weekend) (weekend) (this weekend),” this is very interesting “is the death of the AI ​​complex infrastructure. As we know. ‘”

Deepse created Liang Wenfeng in 2023, the founder of High-flyer, a quantitative coating fund in AI. AI Startup grew in AI research unit in AI in AI 2023, to focus on major language models and gain a branch of general intelligence, equalizing or exceeding human mind, and that is followed by Openai and others are.

Deepseeke still has high aircraft and financed by Flyer, according to Jefferies’s analysts.

Deepsee’s roar around this month began receiving the steam, when he released the starting point, reasoning reasoned for the Openai O1 rivals. It is an open source and can be used by air developers.

Like other Chinese shaltters, Deepseek has limits on certain topics: When Chinese Leaders asked about Xi Jinping policies, such as DeEpsekek reported The user is left out from similar to similar questionnaire.

Openai CEO Sam Altman praised the pattern publicly, but the company also said that Secseeeches prove badly harvest Openai data to build their product.

In Washington, DC event, Openai Thursday Thursday, Altman said Deepse is “clearly a great model”.

“It’s a reminder of the competition level and to win the need for Al Democrat,” he said. “The level of interest in reasoning, he said he also points to the level of interest in the open source.”

See: Nvidia CEO Jensen Huang and Trump President meet at AI politics