Barriers as Incentives? AI Export Controls and US-Led Cooperation Against China

The new US regulations on the export of devices for the development of artificial intelligence technology pose a short-term challenge to transatlantic unity, especially in the face of threats from China and promotion of the DeepSeek AI model. In the long term, however, these restrictions may serve as an additional incentive for the affected countries to further develop their AI industries and enhance security measures.

On January 13, the US Department of Commerce announced new export control regulations for advanced computing chips and closed AI models. Titled the “Framework for Artificial Intelligence Diffusion,” the regulations focus on the spread of AI technologies and the requirements for US companies to obtain export permits. According to analysts from the Carnegie Endowment for International Peace, the regulations aim to establish new export protocols and security procedures for the US AI technology market, from which the most advanced global solutions emerge. Issued in the final days of the Biden administration, the document builds upon restrictions introduced in September 2022 and October 2023, with the new measures set to take effect 130 days after their announcement, depending on the decision of the Trump administration.

The New Categorization of AI Export Destinations

The most significant aspect of the new regulations is the introduction of three categories of countries to which American companies, primarily NVIDIA, can sell their products. The first group includes 18 major US allies, including the Five Eyes nations, South Korea, Japan, Taiwan, Belgium, and the Netherlands. There are no restrictions on sales to these countries, though the US companies, including technology giants like Google, Microsoft, and Amazon, must adhere to strict security rules, such as retaining half of their computing power within the US and limiting their usage to no more than 7 percent in countries classified under the second group.

The second group includes most countries in the world, including NATO and EU member states in Central and Eastern Europe (CEE), such as Austria, Poland, and the Baltic states. These countries face an export limit of 50,000 chips, which may be doubled by 2027 if they sign a separate AI security agreement with the US. Individual companies from these countries may also apply for an increased limit, contingent on meeting chip-related security conditions and submitting a declaration that their use of US-produced AI tools does not violate human rights (e.g., through social surveillance). The third group consists of US adversaries, including China and Russia, to which AI chip exports are strictly prohibited.

Many second-group countries, particularly in CEE, viewed the restrictions as a sign of distrust from the Biden administration. The European Union shares this perspective and has expressed plans to raise the issue with the Trump administration, emphasizing the restrictions’ symbolic significance in the context of NATO’s deterrence on the eastern flank.

However, uncertainty remains over whether the Trump administration will uphold these restrictions. Given its pro-American and anti-regulatory stance, it is likely that the regulations will remain in place, despite potential attempts to influence the categorization of countries. This is because the regulations align with Trump’s rhetoric of US hegemony, which is based on demanding concessions and leveraging tariffs as a political tool. The January 22 announcement of a $500 billion investment by OpenAI, Oracle, and SoftBank, attended by Trump himself, underscores the growing importance of AI in the US-China rivalry.

Security Concerns and the DeepSeek Model

Without delving into the technical complexities of building generative AI models like ChatGPT and DeepSeek, and their related needs for chips, it is important to note that a key factor in the Biden administration’s classification of countries was the security of these technologies. There were concerns that the previous export control system was not stringent enough, allowing Chinese entities to access restricted chips.

The release of the DeepSeek model in January – officially based on Chinese processors and a relatively modest budget, but most likely utilizing NVIDIA chips that have been subject to restrictions since 2019 – provided further justification for the new restrictions. The open-source nature of DeepSeek has also led some experts to suggest that countries from the second group may leverage it to bypass American restrictions and develop their own AI models.

Transatlantic Challenges and NATO’s AI Strategy

These developments raise important questions about transatlantic relations. Will the Trade and Technology Council return and serve as an effective forum for discussing these restrictions? How will NATO balance AI security with deterrence against China and Russia?

Further regulations like those introduced by Biden could impact obligations of NATO allies and the Alliance’s broader strategy in Europe. In the short term, such measures may further strain transatlantic unity, which is already under pressure. In the long term, they present an opportunity for the second-group countries to take steps to meet the US security demands while expanding their AI ​​sectors. Moreover, the EU has emphasized the need for discussions on the inconsistency of the US approach toward its member states. Afterall, coordinating policies in this area could not only reinforce transatlantic relations during challenging times but also provide a competitive edge against Chinese AI models like DeepSeek and similar efforts by companies such as Alibaba.

For NATO, AI security is not a top priority, at least when compared to issues such as Baltic Sea security, critical infrastructure protection, or deterrence along the eastern flank. However, strengthening the security of American chips and maximizing their production capacity could enhance US-EU cooperation against China. And while AI may not be its primary focus, NATO has already incorporated AI considerations into its 2021 and 2024 strategies, with the latter including updates on generative AI.

The updated strategy states: “AI is becoming a general-purpose technology, with different risks, actors and levels of complexity between narrow AI, designed to perform specific tasks, and emerging AI technologies, such as frontier or foundation AI models, which are multi-purpose and capable of performing complex tasks.” Recognizing AI as a transformative technology with varying risks and applications, the strategy aims to accelerate the safe and responsible adoption of AI within NATO while enhancing interoperability and safeguarding against threats from adversarial use of these technologies.

The US export restrictions, however, appear to contradict the goals outlined in the updated strategy. On one hand, NATO seeks to accelerate the expansion and interoperability of AI systems, including the supply of ready-made models, which the US regulations hinder. On the other hand, the restrictions align with NATO’s goals of securing and monitoring AI technologies – an area where the US remains dominant, especially in the field of generative AI chips. Indeed, the updated strategy highlights this issue: “In order to promote NATO’s AI readiness, strategic foresight is needed, including a wide range of proactive activities, from anticipatory governance to alternative scenario planning based on participatory and responsible approaches.”

The Future of US AI Policies

Further context for the US decision to restrict AI chip exports to certain European allies is provided by the NATO Innovation Fund, established in 2023. Supported by 24 NATO countries (excluding the US), the fund invests in companies developing emerging, cutting-edge technologies, including AI. In June 2024, the fund confirmed investments in the British firm Fractile, which specializes in the development of large language models (LLMs) similar to ChatGPT and DeepSeek. Notably, the UK is among the countries exempt from the export restrictions under Biden’s regulations.

It is expected that Trump’s approach will be similar to Biden’s in that it will focus on the development of generative AI based on American technologies, prioritizing US businesses and competencies while taking a transactional approach to cooperation with NATO partners and allies. Even if some adjustments are made to the list of restricted countries, the broader emphasis on US dominance in AI technology is unlikely to change. In this context, China’s DeepSeek model – while not necessarily cheaper but more accessible due to its open-source nature – could serve both as an incentive and a bargaining chip in negotiations between the US and developing countries, as well as within transatlantic cooperation and the EU’s ambitions to develop its own generative AI.