The notion that Chinese DeepSeek spent less than $ 6 million to develop her artificial intelligence system is “exaggerated and a little misleading,” according to Demis Hassabis, Google Deepmind’s head.
Last month, DeepSeek shocked the world to and an AI model using much less money than US rivals, such as Deepmind and Openai. Hassabis, who directs Google’s AI unit, told the Bloomberg Television That Deepseek “seems to have only reported the cost of the final training round, which is a fraction of the total cost.”
The executive also ruled out the idea that DeepSeek’s emergence alters AI development economy. “We don’t see any new miraculous technology,” Hassabis said on Monday (10) in Paris at the Artificial Intelligence Action Summit. “Deepseek is not a point outside the efficiency curve.”
The Chinese startup reported having spent $ 5.6 million on computing costs to train its model using older Nvidia chips. Several researchers questioned these allegations. .
A Bloomberg News Previously reported that a DeepSeek -linked group obtained OpenAi data using a process known as “distillation”, in which one AI model stack outputs from another for training purposes.
Hassabis said DeepSeek seems to “have been based on some western models to distill” without offering details.
Continues after advertising
A DeepSeek representative did not immediately respond to a request for comment.
Google owner, Alphabet is one of several silicon valley giants that kept great spending plans after DeepSeek caused a riot in public markets. Last week, Alphabet reported plans to $ 75 billion in capital expenses in 2025. This money will go to its cloud computing division and services such as Gemini, an AI model that Google is integrating with research and other products .
“Gemini is more efficient than DeepSeek in terms of performance or performance cost,” said Hassabis. “We just don’t talk about it much.”
Continues after advertising
© 2025 Bloomberg L.P.
