Release of DeepSeek-R1
On 20 January 2025 the Hangzhou AI start-up DeepSeek released its reasoning model DeepSeek-R1 as open source under the MIT licence, claiming performance close to comparable OpenAI models; its earlier V3 technical report had put the full training cost at about $5.576 million. A week later its app topped Apple's US App Store, and on 27 January Nvidia's shares fell almost 17%, erasing a record $593 billion in market value, as President Donald Trump called DeepSeek a "wake-up call" for American industry. Italy, South Korea and others then restricted it on data-security grounds, and China's foreign ministry said Beijing had never asked companies to collect data illegally. In September 2025 the R1 paper was published in Nature after peer review.
Background
DeepSeek is an artificial intelligence start-up based in Hangzhou whose controlling shareholder, according to records cited by Reuters, is Liang Wenfeng, co-founder of the quantitative hedge fund High-Flyer. Under US export controls, Nvidia's H800 chips were barred from sale to China from 2023.
In December 2024 DeepSeek published a technical report on its DeepSeek-V3 model saying the full training run had taken only 2.788 million H800 GPU hours — about $5.576 million at an assumed rental price of $2 per GPU hour. The report noted that the figure excluded the costs of prior research and ablation experiments on architectures, algorithms and data.
The Release of R1
On 20 January 2025 DeepSeek released its reasoning model DeepSeek-R1, making the model weights open source under the MIT licence, which allows free distillation and commercial use, together with six smaller models distilled from R1; it said the 32-billion and 70-billion-parameter versions performed on a par with OpenAI's o1-mini. That afternoon Premier Li Qiang chaired a symposium of experts and entrepreneurs in Beijing to hear views on the draft government work report, at which Liang Wenfeng spoke.
Market Shock
Over the weekend of 25–26 January DeepSeek's app rose to the top of Apple's US App Store, overtaking ChatGPT. On 27 January investors dumped technology stocks on fears that the low-cost Chinese model threatened the dominance of American AI firms; Nvidia's shares fell almost 17%, wiping $593 billion off its market value, a record one-day loss for any company on Wall Street. The same day President Donald Trump told House Republicans at their policy retreat that DeepSeek's release "should be a wake-up call for our industries", while adding that it could be a positive development if the same results could be achieved for less money.
Restrictions and China's Response
Several countries then restricted DeepSeek on data-security grounds. Italy's data protection authority, citing unsatisfactory answers from the company, ordered it to stop processing Italian users' data and opened an investigation, and several South Korean government ministries blocked access. On 6 February Chinese foreign ministry spokesperson Guo Jiakun said the Chinese government protected data privacy and security in accordance with the law and "has never asked and will never ask enterprises or individuals to collect or store data illegally", and opposed overstretching the concept of national security and politicising economic and technological issues. On 17 February South Korea's Personal Information Protection Commission said new downloads of the app had been suspended after DeepSeek acknowledged failing to follow some of its data protection rules.
Peer-Reviewed Paper
On 17 September 2025 a peer-reviewed paper describing R1 was published in Nature, making it what is thought to be the first major large language model to undergo peer review. Its supplementary material disclosed for the first time that R1 cost the equivalent of about $294,000 to train, on top of about $6 million spent developing the base model it was built on; DeepSeek said R1 was trained mainly on H800 chips.
Narrative Comparison
| Source | Narrative |
|---|---|
| Chinese official position (Foreign Ministry spokesperson Guo Jiakun, 6 February 2025) | Regarding countries that have banned or restricted DeepSeek: the Chinese government attaches great importance to and protects data privacy and security in accordance with the law, and has never asked and will never ask enterprises or individuals to collect or store data illegally. China consistently opposes overstretching the concept of national security and politicising economic, trade and technological issues, and will firmly safeguard the legitimate rights and interests of Chinese companies. |
| US official response (Trump at the House Republican retreat, 27 January 2025) | The release of DeepSeek, AI from a Chinese company, should be a wake-up call for American industries that they need to be laser-focused on competing to win. It could also be a very positive development: instead of spending billions and billions, companies might spend less and hopefully come up with the same solution. |
| Overseas research analysis (Gregory C. Allen, CSIS, 2025) | Gregory C. Allen of the Center for Strategic and International Studies argued that DeepSeek's success partly reflected failures in the earlier implementation of US export controls, but that the controls could still play a critical role in America's AI strategy toward China — though at best they could slow and disrupt Chinese efforts, not stop them. The gap had narrowed significantly, he wrote, and it was unrealistic to expect a US lead of more than a year or two even with extremely aggressive export controls. |
Key Milestones
- DeepSeek-V3 technical report
The report said V3's full training took 2.788 million H800 GPU hours, about $5.576 million.
- DeepSeek-R1 released
R1 was released under the MIT licence; the same day Liang Wenfeng spoke at a symposium chaired by Li Qiang.
- Nvidia loses $593 billion
Tech stocks sold off and Nvidia fell almost 17%; Trump called DeepSeek a "wake-up call".
- China responds to restrictions
Guo Jiakun said the Chinese government had never asked companies to collect or store data illegally.
- South Korea suspends downloads
South Korea's data protection commission suspended new downloads of the app.
- R1 paper published in Nature
The peer-reviewed paper disclosed that R1 cost about $294,000 to train.
Last verified:
Cite this page
Chronicles of Modern China. (2026, November 2). Release of DeepSeek-R1. Retrieved November 2, 2026, from https://www.cmcarchive.org/events/deepseek-r1-release-2025
“Release of DeepSeek-R1.” Chronicles of Modern China. Last modified November 2, 2026. Accessed November 2, 2026. https://www.cmcarchive.org/events/deepseek-r1-release-2025.
Chronicles of Modern China. Release of DeepSeek-R1[EB/OL]. (2026-11-02)[2026-11-02]. https://www.cmcarchive.org/events/deepseek-r1-release-2025.