Developments in the AI Arms Race—Does The Emergence of DeepSeek Mean That OpenAI is in Deep Water?
By Nate Garcia; Photo Credit: Greg Baker/Agence France-Presse — Getty Images
Last week, “scrappy Chinese A.I. start-up” DeepSeek released a statement about their language model—DeepSeek-R1—which uses “a modest number of second-rate A.I. chips” to create a result “to match the performance of leading American A.I. models at a fraction of the cost.”[1] Additionally, DeepSeek claimed it could train its language model for a mere $5.6 million.[2] For reference, OpenAI spent over $60 million training ChatGPT-4.[3] Effectively, DeepSeek created a language model that seems to be at least as efficient and effective as OpenAI’s ChatGPT while using less expensive computer chips and much less infrastructure.[4] As a result of this “breakthrough,” A.I.-adjacent stocks, such as chip-manufacturing titan Nvidia, plummeted.[5] The market questioned whether Nvidia’s advanced A.I. chips—which account for a large percentage of its total revenue—are as necessary to create and operate advanced A.I. language models as we once thought, causing investors to dump Nvidia stock at an unprecedented level.[6] From share prices just before DeepSeek’s infamous press release, Nvidia lost a staggering $600 billion in its market capitalization this week in a matter of hours.[7] American tech executives, politicians, and social commentators are left questioning what this means for American A.I. companies like OpenAI and the American economy more broadly in the coming age of A.I.[8]
The industry has reacted to DeepSeek’s breakthrough in different ways.[9] Initially, there were a lot of skeptics who suggested perhaps DeepSeek fabricated the numbers in its report or perhaps they were using first-rate chips that they unlawfully appropriated.[10] OpenAI seems to be concerned that Chinese companies, like DeepSeek, have “inappropriately distilled” ChatGPT’s code.[11] As of a few days ago, OpenAI is investigating whether DeepSeek was able to distill (or wrongfully appropriate) ChatGPT’s famously protected intellectual property.[12] Interestingly, DeepSeek is allowing its model to be entirely open-source. Therefore, if DeepSeek was able to distill ChatGPT, DeepSeek’s open-source code would also reveal protected intellectual property belonging to OpenAI.[13] As this story developing day by day, we will have to wait and see what Open AI’s investigation reveals before we can validate or dismiss these concerns.
Others are using DeepSeek’s efficiency to criticize American tech companies’ inefficient spending habits on unnecessary computing infrastructure.[14] In the first few days of President Trump’s second term in office, he has revealed he plans to devote $500 billion to make the United States at the forefront of the artificial intelligence arms race of the 2020s.[15] However, if Chinese companies can create superior language models at a fraction of the cost, it would appear the United States is years behind.[16] For that reason, many are calling DeepSeek’s report this century’s “Sputnik moment” in the A.I. arms race between the United States and China.[17]
Whether or not DeepSeek’s report was 100% factually accurate or whether they have distilled OpenAI’s intellectual property, one thing is clear—the United States needs to think critically about the money it spends on computing infrastructure to support and perpetuate A.I. language models.
Nate Garcia is a 2L at Vanderbilt. He graduated from the University of Colorado Boulder in 2023. After graduating from law school, he plans to move to Dallas, Texas to practice corporate law.
[1] Kevin Roose, Why DeepSeek Could Change What Silicon Valley Believes About A.I., New York Times (Jan. 28, 2025), https://www.nytimes.com/2025/01/28/technology/china-deepseek-ai-silicon-valley.html.
[2] See Ryan Browne & Dylan Butts, DeepSeek’s AI Claims Have Shaken the World—But Not Everyone’s Convinced, CNBC (Jan. 30, 2025), https://www.cnbc.com/2025/01/30/chinas-deepseek-has-some-big-ai-claims-not-all-experts-are-convinced-.html.
[3] See Iliya Valchanov, How Much Did it Cost to Train GPT-$? Let’s Break it Down?, Team GPT (Jul. 12, 2024) https://team-gpt.com/blog/how-much-did-it-cost-to-train-gpt-4/.
[4] See id.
[5] Evrim Ağaci, DeepSeek Shakes AI Stocks, Nvidia Loses $600 Billion Valuation, The Pinnacle Gazette (Feb. 1, 2025), https://evrimagaci.org/tpg/deepseek-shakes-ai-stocks-nvidia-loses-600-billion-valuation-173918.
[6] See id.
[7] See id.
[8] See id.
[9] See Roose, supra note 1.
[10] See id.
[11] See Julia Shapero, OpenAI Investigating Whether DeepSeek Improperly Obtained Data, The Hill (Jan. 29, 2025), https://thehill.com/policy/technology/5113470-openai-deepseek-data-theft/.
[12] See id.
[13] See id.
[14] See Valchalov, supra note 3.
[15] See Paul Smith-Goodson et al., The Stargate Project: Trump Touts $500 Billion Bid for AI Dominance, Forbes (Jan. 30, 2025), https://www.forbes.com/sites/moorinsights/2025/01/30/the-stargate-project-trump-touts-500-billion-bid-for-ai-dominance/.
[16] See id.
[17] Id.