DeepSeek, the Chinese AI model that has shocked the tech industry and disrupted Wall Street with its significantly low cost and good performance, has "some real innovation" and is "all good news," said Satya Nadella, Microsoft's CEO.
Nadella made the comment during the company's quarterly earnings call on Thursday, when he said that AI is in no way different from normal computing developments. Microsoft has begun offering the DeepSeek-R1 model for inference, an AI model that reveals its "thought process" so users can check its results. This model was available to users of Microsoft's cloud platform the same day.
The accusations, despite the use of
Microsoft, a major investor in DeepSeek's U.S. competitor, OpenAI, is also investigating whether a Chinese AI developer obtained OpenAI's output data in an unauthorized manner.
OpenAI has informed the media that it has evidence that DeepSeek has been using the OpenAI service to train its AI models, behavior that is not in accordance with OpenAI's terms of service.
However, neither OpenAI nor Microsoft provided evidence.
While the investigation continues, a Trump administration appointee has already accused DeepSeek of using "stolen" US technology.
Howard Lutnick, the nominee for Commerce secretary, said Wednesday in a meeting with U.S. senators that the administration will address the issue.
Meanwhile, David Sacks, President Trump's AI adviser, said there is "substantive evidence" that DeepSeek "gained knowledge" from OpenAI models.
"Distillation will violate most terms of service, yet it's ironic - or even hypocritical - that Big Tech is criticizing it," technology investor and Cornell University lecturer Lutz Finger said Wednesday. "Training ChatGPT on Forbes or New York Times content also violated their terms of service."
What is "distillation"?
Distillation is the process by which a new AI model repeatedly queries a larger model and learns from its outputs.
This process is described in DeepSeek's public research papers, but the researchers said they used it in a different way. According to the company, they used the DeepSeek-R1 model for inference as a "larger model" that "taught" other models, such as Alibaba Qwen and Meta Llama, to also become capable of inference.
The distilled models and the original R1 have been released for free download, allowing people with less powerful computers or even smartphones to run the models offline with full control, which is impossible for ChatGPT users because the model behind the OpenAI service is hidden even from paying users.
Social media posts have indicated that DeepSeek sometimes identifies itself as ChatGPT, which could be evidence of data theft. But like all other AI models, DeepSeek doesn't always tell the truth. A previous version of Google's Gemini chatbot identified itself as Baidu's Ernie bot when queried in Chinese, but Baidu has never accused Google of stealing data.
English 
Czech 
Spanish 
French 
Italian 
German 
Russian 
Chinese