The AI chatbot reportedly topped OpenAI’s ChatGPT as most-downloaded free app on Apple’s app store last week.
Chinese artificial intelligence (AI) app DeepSeek said on Monday that it would temporarily limit user registrations due to “large-scale malicious attacks” targeting its services.
DeepSeek reported a “major outage” affecting its application programming interface (API) and user logins on Monday.
Registered users could still log in to the new AI platform as usual, according to its status webpage.
The Chinese startup, founded in 2023, did not provide details about the attacks or specify when its regular registration process will resume.
The company launched its open-source reasoning model DeepSeek-R1 last week, with performance on par with OpenAI’s reasoning model called o1, according to its statement.
The company said that its AI model was developed at a fraction of the cost of rival OpenAI’s models, using NVIDIA’s less-advanced H800 chips.
DeepSeek AI chatbot reportedly topped OpenAI’s ChatGPT as the most downloaded free app on Apple’s app store last week. Shares of U.S. chip companies declined following the news, with Nvidia’s shares dropping about 17 percent on Jan. 27 and Advanced Micro Devices falling more than 6 percent.
Nvidia issued a statement following the decline in its shares, saying that “DeepSeek’s work illustrates how new models can be created using that technique, leveraging widely-available models and compute that is fully export control compliant.”
The U.S. microchip export controls were designed to freeze China’s development of supercomputers used to develop nuclear weapons and artificial intelligence systems.
One of DeepSeek’s research papers stated that it had used about 2,000 of Nvidia’s H800 chips to train its AI models, much less than the 16,000 chips typically used by other companies.
Bernstein analysts on Monday highlighted in a research note that DeepSeek’s total training costs for its V3 model were unknown but were much higher than the $5.58 million the startup said was used for computing power. The analysts also said the training costs of the R1 model were not disclosed.
Some developers are skeptical about DeepSeek’s claim.
Scale AI founder and CEO Alexandr Wang told CNBC on Jan. 23, without providing evidence, that it is possible that DeepSeek has about 50,000 Nvidia H100 chips.
