Background
On January 20, 2025, Chinese AI startup DeepSeek released its latest model, DeepSeek-R1. Boasting impressive performance on various benchmarks and advanced capabilities in reasoning tasks, this model challenged OpenAI’s o1 model, which sent shockwaves through the U.S. stock market and rattled the American tech and finance industries.
After the release of ChatGPT in 2022, Chinese tech firms scrambled to create their own AI chatbots. Founded by Liang Wenfeng and funded solely by Liang’s hedge fund High-Flyer, DeepSeek emerged in May 2023 amidst this rush. By November 2023, DeepSeek released its first open-source model, DeepSeek Coder. DeepSeek-V2, which launched in May 2024, gained significant attention for its strong performance at a low cost, subsequently causing a price war within the Chinese AI market. Other market participants, such as ByteDance, Tencent, Baidu, and Alibaba were forced to lower their prices to remain competitive.
DeepSeek’s most recent models, however, brought disruptive competition directly to American shores. DeepSeek-V3 and DeepSeek-R1 challenged OpenAI’s o1 model while requiring significantly fewer resources. Specifically, the most recent model’s training required less than $6 million worth of computing power from Nvidia H800 chips. In comparison, Meta just last week said it would spend upward of $65 billion on AI development this year, and OpenAI said the AI industry would need trillions of dollars in investment to support the development of chips needed to power the data centers that run the industry’s models. Therefore, DeepSeek’s recent model and published paper revealed that it could achieve what American tech companies spent billions of dollars and numerous years to do in a fraction of the time and cost. On top of this, DeepSeek achieved the feat with Nvidia H800 GPU (in comparison to export-restricted and powerful H100) with a reduced chip-to-chip transfer rate and slower training speed. These findings consequently sent American markets into a frenzy.
In response to DeepSeek accomplishing the near-unimaginable, OpenAI launched an investigation into whether DeepSeek used its proprietary AI models without authorization. While DeepSeek denies any wrongdoing, asserting that its models are independently trained, OpenAI claims that DeepSeek trained its chatbot by repeatedly querying ChatGPT, a process known as distillation. Additionally, OpenAI is investigating whether DeepSeek obtained API access to interact with GPT-4 at scale, which if proven, would constitute a violation of OpenAI’s terms of service. The White House chimed in on the conflict. Trump administration’s AI and crypto czar, David Sacks, stated that DeepSeek’s distilling amounted to intellectual property theft.
Legal Implications
While OpenAI has not expressly stated whether it plans to pursue legal action, a spokesperson for the company stated that the company plans to take “aggressive, proactive countermeasures to protect [its] technology.”
IP Infringement
However, with OpenAI on the heels of its own copyright infringement lawsuit filed by the New York Times, legal experts concur that a legal challenge against DeepSeek would be an uphill battle “now that the content-appropriation shoe is on the other foot.” AI-generated content and models exist in a legal gray area when it comes to IP protection, leaving OpenAI with the difficult hurdle of proving that DeepSeek’s model directly replicated OpenAI’s proprietary model. The primary issue lies within whether AI outputs qualify as creative expression, which would grant the outputs with copyright protections. However, OpenAI argued in the New York Times case that training AI is an allowable “fair use” exception to copyright protections. Therefore, for OpenAI to have any chance against DeepSeek on the IP infringement claim, OpenAI has to abandon its position in the New York Times suit on the “fair use exception,” convince the court of the copyrightability of AI output, and then finally provide credible evidence of DeepSeek’s improper use of OpenAI’s models. Therefore, OpenAI’s potential IP infringement claim would be unlikely to occur or to succeed.
Breach of Contract
Given the uphill battle in IP infringement, however, OpenAI may be able to bring forth a breach of contract claim. The OpenAI terms of use expressly forbid using its outputs to develop models that compete with OpenAI. If OpenAI can obtain evidence that DeepSeek did train its model with OpenAI’s output, it would have grounds for a lawsuit. Even so, the OpenAI terms of use require that most claims be resolved through arbitration, with a slim exception for lawsuits intended to stop the unauthorized use or abuse of the services or IP infringement. Therefore, to have its day in court on the breach of contract claim, OpenAI has to necessarily revive its IP infringement claim.
Apart from the merits of its claims, OpenAI would have to litigate in an ever-present legal gray area with the uphill battle of establishing precedent at the intersection of AI and law. To date, no AI-model creator has even tried to enforce its terms of service with monetary penalties or injunctive relief. Even if OpenAI manages to win a judgment in the United States, for any relief such as enjoining DeepSeek or requiring it to pay monetary damages, OpenAI would need to take its claim to the Chinese legal system, where DeepSeek is based.
Thus, the combination of sparse precedent and geographical complications leads to the conclusion that a lawsuit, and a successful outcome at that, would be extremely rare and difficult to accomplish for OpenAI.
Current Climate
Just nine days after DeepSeek’s V3 and R1 models ravaged U.S. markets, cybersecurity researchers at Wiz Research revealed that the company had suffered a significant data leak, leaving over one million sensitive records exposed. The records were publicly accessible to anyone with an internet connection, and researchers found that the leak was caused by an oversight of cloud storage on its system. While investigations are underway, discourse regarding the privacy concerns of AI has skyrocketed. Because DeepSeek’s servers store sensitive user data in China, the data could be subject to China’s cybersecurity laws, influencing future AI export regulations between the United States and China.
Business Implications
It remains clear that DeepSeek’s emergence has already impacted AI-related stocks, and its full effects are yet to be known. Companies such as Nvidia and Broadcom saw significant stock value drops, and investors are questioning whether AI startups are overvalued. As for existing AI companies, DeepSeek’s open-source model threatens OpenAI’s subscription-based model, opening the door for discussions about changing its business model, lowering prices, or adjusting licensing policies.
The conflict between DeepSeek and OpenAI underscores the growing business, legal, and geopolitical tensions in the AI industry. OpenAI’s allegations of unauthorized distillation and potential API misuse highlight the challenges of enforcing intellectual property rights in an era where AI models can be replicated or fine-tuned using publicly available techniques. Meanwhile, DeepSeek’s independent development and open-source accessibility raise important questions about the boundaries of fair use in AI training and the future of the industry. Regardless of legal action, this dispute will shape AI regulation, IP law, and global competition in emerging technologies.
*The views expressed in this article do not represent the views of Santa Clara University.
Comments