The geopolitics of artificial intelligence after DeepSeek

The sudden emergence of DeepSeek, a relatively unknown Chinese artificial intelligence start-up, has led to a massive correction in the stratospherically high valuations of the United States tech giants involved in AI. DeepSeek claims it can do what AI leader OpenAI can do – and more – with a much smaller investment and without access to the most advanced computer chips, which are restricted by US export controls. DeepSeek’s model seems to run at much lower cost and consumes much less energy than its American peers.

The clear implication is that US hegemony in AI is no longer guaranteed. With the right talent, similar results can be obtained with much less money. Furthermore, US export controls to contain China technologically appear ineffective.

In principle, the development is positive for the world. It suggests that the European Union, so far a follower in generative AI, could potentially find itself with a homegrown AI platform. This has seemed impossible so far because of the sheer amount of investment needed to develop AI models.

However, there are also less positive aspects. Technically, DeepSeek hardly compares with US AI platforms since it mainly optimises existing models, rather than developing new ones that could compete with those from the US. Model optimisation is important and welcome but does not eliminate the need to create new models. In other words, while DeepSeek has been able to reduce computing costs massively and opens the door to efficient architectures to reduce performance gaps between smaller and larger models, it does not fundamentally break the ‘scaling law’ according to which larger models deliver better results. Huge financial resources continue to matter.

DeepSeek is also not fully open-access. Its training data, fine-tuning methodologies and parts of its architecture remain undisclosed, although it is more open than US AI platforms. This is important considering that DeepSeek, as any Chinese AI company, must comply with China’s national security rules. China’s AI regulations require any AI output from a Chinese AI platform to avoid criticism of the Chinese political regime and to be in line with Chinese propaganda. Whether Western governments will accept such censorship within their jurisdictions remains an open question for DeepSeek.

The EU AI Act, for example, does not cover censorship directly, which is good news for DeepSeek. But AI systems deployed in the EU must be transparent and accountable and must respect human rights, including freedom of expression and political speech – a potential challenge for DeepSeek.

A more immediate challenge is data protection and in particular, the EU general data protection regulation. Italy blocked DeepSeek on 30 January on data-transfer grounds (DeepSeek acknowledges it stores most data in China), and Belgium and Ireland have started investigations. This raises concerns about data sovereignty and potential government access, which could limit DeepSeek’s usability in the EU.

The biggest risk to DeepSeek, however, is geopolitical. President Trump’s comments on how DeepSeek may be a wake-up call for US tech companies signal that AI will be at the forefront of the US-China strategic competition for decades to come. Heightened competition over AI leadership is related not only to its commercial use, but also to military use, from cyber warfare to unmanned weapons.

Both the US and China seem set to put even more financial resources into AI, while also further limiting access to this technology. From the US side, the most likely outcome will be a doubling down of AI-related export controls and the drying up of any remaining AI cooperation between the US and China.

While recognising the positive aspects arising from the commoditisation of AI after DeepSeek’s success, the EU should realise that even greater technological competition between the US and China for AI dominance will have consequences for Europe. Most immediately, there is likely to be a split into two AI worlds as a consequence of tighter export controls, sharply reduced scientific cooperation and regulation.

This is bad news for Europe as it unlikely to be able to operate in the two ecosystems, reducing the potential efficiency gains of AI advances. The censorship and data transfer risks of DeepSeek must be traded off against the US ecosystem under Trump, which may not bring gains to the EU in terms of scientific cooperation or technology transfer, as US allies are increasingly treated as non-allies. 

This is an output of China Horizons, Bruegel's contribution in the project Dealing with a resurgent China (DWARC). This project has received funding from the European Union’s HORIZON Research and Innovation Actions under grant agreement No. 101061700.