The DeepSeek shock: the Chinese model that wiped $593 billion off Nvidia in a day
In January 2025 a little-known Chinese lab released an AI model it said matched OpenAI's best at a fraction of the cost. Nvidia suffered the biggest one-day loss in market history. The cost claim was narrower than it looked, the US has since banned DeepSeek from federal systems, and the company has since raised billions at a valuation of 500 billion yuan.

HONG KONG On 20 January 2025 DeepSeek, a Chinese AI lab, released a reasoning model called R1, free to download under an open licence 1. The company claimed R1’s performance was “on par with OpenAI-o1” on maths, coding and reasoning, and priced access to it far below its US rivals 12.
A week later, on Monday 27 January, investors took fright. Nvidia, whose chips power most AI systems, fell almost 17% to close at $118.42. Reuters, citing LSEG data, put the loss in market value at about $592.7 billion, the biggest one-day fall for any company on Wall Street 2. The Philadelphia semiconductor index dropped 9.2% 2. “It could mean less demand for chips, less need for a massive build-out of power production to fuel the models,” said Brian Jacobsen, chief economist at Annex Wealth Management 2. President Donald Trump called DeepSeek a “wake-up call” for US industry, and said a cheaper way to build AI could be a positive for the US 3.
What it really cost
The panic rested on a number. DeepSeek’s technical report for its V3 model, on which R1 was built, said training needed 2.788 million hours on Nvidia’s H800 chips 4, which DeepSeek valued at about $5.6 million at rental prices. But the paper said that covered only the final “official training” run, and excluded the costs of “prior research and ablation experiments on architectures, algorithms, or data” 5. The research firm SemiAnalysis estimated that DeepSeek’s spending on hardware was “well higher than $500M over the company history” 5. The headline figure was real, but it measured something much narrower than the cost of building the model.
Washington’s response
The US government moved on several fronts. In April 2025 it told Nvidia that sales of its H20 chip, designed to comply with earlier export rules, would need a licence; Nvidia said it would take about $5.5 billion in charges 6. A House select committee report that month alleged that DeepSeek sent US user data to China, censored answers on sensitive topics and used “material unlawfully obtained from U.S. AI models” 7. China’s embassy in Washington has called the theft allegations baseless 8. A US government evaluation in September 2025 found US models outperformed DeepSeek’s on most benchmarks; “The report is clear that American AI dominates, with DeepSeek trailing far behind,” said Commerce Secretary Howard Lutnick 9.
Questions about how DeepSeek obtained its chips have not been settled. In February 2026 an unnamed senior US official told Reuters that DeepSeek’s next model had been trained on Nvidia’s top Blackwell chips, which are barred from export to China 10. No charge or official finding has named DeepSeek. The same month the US AI company Anthropic said DeepSeek was among three Chinese labs that had used fraudulent accounts to “distil” its Claude models, accounting for more than 150,000 exchanges 11. These are allegations; they have not been tested in court.
In October 2026 the Department of Homeland Security ordered federal civilian agencies to remove DeepSeek products, including third-party products built on its technology 1213. China’s foreign ministry said Washington was “overstretching the concept of national security and politicizing technology and trade issues” 12.
How it unfolded
-
DeepSeek publishes its V3 technical report.
-
DeepSeek releases R1.
-
Key moment. Nvidia loses about $593bn in market value in a day.
-
US requires licences for H20 exports; House committee report on DeepSeek.
-
Nvidia becomes the first $5 trillion company.
-
US official alleges DeepSeek trained on banned Blackwell chips; Anthropic alleges distillation.
-
DeepSeek previews V4, adapted for Huawei chips.
-
Key moment. DHS order removing DeepSeek from federal systems published.
-
Reuters reports a $15bn DeepSeek funding round.
What came next
The sell-off did not last. Nvidia became the first company worth $5 trillion in October 2025, less than a year later 14. DeepSeek kept releasing models; in April 2026 it previewed V4, which Reuters reported had been adapted for chips made by Huawei, though DeepSeek did not say which processors it used for training 158. Chinese rivals Zhipu AI and MiniMax listed in Hong Kong in January 2026 16. In October 2026 Reuters reported, citing sources, that DeepSeek had raised about 100 billion yuan ($15 billion) in a funding round that valued it at 500 billion yuan before the new money 17.
Why it matters
The DeepSeek shock showed how fragile the market’s assumptions about AI were: a model release from a little-known Chinese lab erased almost $600 billion from one company in a day. It also showed how quickly AI became a matter of national security, from export controls to bans on government devices. The cost claim turned out to be narrower than the headlines; the competitive pressure from Chinese open models did not.
Sources and references
- Official documentDeepSeek, “DeepSeek-R1 Release”, 20 January 2025. Link
- ArticleDawn (Reuters), “Chinese DeepSeek sparks global AI sell-off, Nvidia loses about $593bn of value”, 28 January 2025. Link
- ArticleFortune, “Trump calls DeepSeek a ‘wake-up call’ for U.S. tech”, 28 January 2025. Link
- ReportDeepSeek-AI / arXiv, “DeepSeek-V3 Technical Report”, 27 December 2024. Link
- ArticleCNBC (via NBC Chicago), “DeepSeek’s hardware spend could be as high as $500 million, new report estimates”, 31 January 2025. Link
- ArticleRTE (Reuters), “Nvidia faces $5.5 billion charge as US restricts chip sales to China”, 16 April 2025. Link
- Official documentHouse Select Committee on the CCP, “Moolenaar, Krishnamoorthi Unveil Explosive Report on Chinese AI Firm DeepSeek”, 16 April 2025. Link
- ArticleProfit / Pakistan Today (Reuters), “China’s AI darling DeepSeek previews new model adapted for Huawei chip technology”, 24 April 2026. Link
- Official documentNIST / US Department of Commerce, “CAISI Evaluation of DeepSeek AI Models Finds Shortcomings and Risks”, 30 September 2025. Link
- ArticleARY News (Reuters), “China’s DeepSeek trained AI model on Nvidia’s best chip despite US ban, official says”, 24 February 2026. Link
- Official documentAnthropic, “Detecting and preventing distillation attacks”, 23 February 2026. Link
- ArticleTechCrunch, “US orders federal agencies to remove China’s DeepSeek AI from their systems”, 20 October 2026. Link
- Official documentCongressional Research Service (via EveryCRSReport), “Removing PRC Artificial Intelligence Models from Federal Agencies (IF13336)”, 27 October 2026. Link
- ArticleDawn (AFP), “AI chip giant Nvidia becomes world’s first $5 trillion firm”, 30 October 2025. Link
- Official documentDeepSeek, “DeepSeek V4 Preview Release”, 24 April 2026. Link
- ArticleThe Star (AFP), “Startups go public in litmus test for Chinese AI”, 8 January 2026. Link
- ArticleARY News (Reuters), “DeepSeek wrapping up $15 billion fundraising to fuel AI race, sources say”, 22 October 2026. Link
Sources and references are provided for transparency and further reading. Allegations and claims are attributed to their respective sources. Triple Frontiers distinguishes reported allegations from independently established facts.
The Frontier Brief
The week’s essential international reporting and new investigations, every Friday.



