The writer is a professor at Vrije Universiteit Brussels and head of the geopolitics and technology programme at the Centre for Security, Diplomacy and Strategy

There has been a proliferation of export control measures introduced in recent years, either to sanction states that have violated international rules (like Russia after its invasion of Ukraine) or deny them access to critical technologies. 

The US is leading the way in the latter. Since 2022, the Biden administration has limited the export of dual-use technology and artificial intelligence (AI) semiconductors to China. On December 2, it upped the ante by tightening those controls and extending them to three types of software. 

Export controls are not new. In the US they date back to the Trading with the Enemy Act of 1917. However, their objectives have remained the same: to protect national security while minimising economic losses from market distortions. 

What is new about export controls is that they now take place in a transformative technological phase, marked by rapid advances in AI. There will be great economic and military benefits for those who innovate and control AI. 

However, before other governments follow the US example by restricting the export of software and other technologies underpinning AI models, they must bear in mind three crucial factors: AI’s private owners, international network and risk of damaging the virtuous links which drive AI innovation.  

Private owners

AI innovation is characterised by advances in software, particularly in neural networks, which allow AI algorithms to identify, reuse and improve the accuracy of the key features of data. That software sits on a huge hardware infrastructure, such as graphics processing units and cloud computing. 

The entire AI infrastructure, both software and hardware, is driven by the private sector, and in particular by big tech. These corporations control the world’s largest clusters of AI processors, best algorithms and most advanced computers. This is a major difference from previous technological revolutions, which were mainly driven by the public sector (think nuclear energy). 

The private sector being in the lead and owning this critical technology has fundamental implications for export control. It means the government must strike agreements with big tech firms to implement their policies. This is further complicated by AI being at the centre of geopolitical rivalry, giving its corporate owners a great deal of leverage over governments. 

More on US trade and investment policies:

• Europe faces up to Trump tariffs impact
• Foreign firms caught by US outbound FDI restrictions
• US Rust Belt: China-backed battery plants become flashpoint amid polarised FDI sentiment

International infrastructure

The second factor governments must bear in mind is that although the AI market is concentrated around a few big tech firms, innovation in AI is open and diffuse. It relies on continuous feedback loops between big tech, software developers and infrastructure providers. Software development is now driven by open-source groups such as the Linux Foundation or GitHub, with developers based in multiple locations. The physical infrastructure supporting AI is also underpinned by a complex global network of participants.

This international infrastructure requires careful co-ordination of export control between different states. The case of semiconductors is emblematic: US export controls to prevent China’s access to the most advanced microchips would have been ineffective without co-ordination with the Netherlands (where ASML is based) and Japan (for memory chips). 

Co-ordination on AI among allies is possible, but its prospects are hindered by geopolitical fragmentation which could deter some governments from co-operating. Further, all nations have an interest in AI innovation, and some may exploit others’ export controls to fill the gap. It wouldn’t be the first time this has happened. In the 1980s, the US imposed export controls on satellites and space technologies, which hurt US companies and allowed industries in other regions — including Europe — to gain market share.

Virtuous links

Finally, export controls in the AI age pose compliance challenges. Controlling software is particularly complex because, unlike hardware (think border controls on semiconductors), it is intangible and difficult to inspect. In addition, AI systems feed off a vast amount of data in open-source libraries, which are typically located in multiple jurisdictions, making them difficult to capture with export controls. 

The will to control software could therefore damage the virtuous links between big tech, software developers and infrastructure providers, which drive AI innovation. There are new challenges with hardware, too. For example, it is no longer necessary to physically own the technology to reap the benefits. A company or country can use cloud computing to train AI systems without having to own all the chips and data centres needed to run them. 

Export controls in the age of AI require co-ordination with the private sector and like-minded countries, while avoiding negative impacts on innovation. Governments must consider these three fundamental principles when introducing new policies, and embrace new frameworks and creative solutions to uphold them. 

Do you want more FDI stories delivered directly to your inbox? Subscribe to our newsletters.