Moreover, there could be a bias from the AI training side, as the LLM can only be trained based on input that is acceptable to the Chinese government. Content on the internet which is critical of the Chinese Communist Party will surely not be included in the training of Deepseek, which limits its reliability and usefulness in particular tasks. Finally, concerns have been raised about the security of user data, with Australia already banning Deepseek on Government devices due to the “unacceptable risk” it poses to national security. Regulators in Ireland, France and South Korea have also launched investigations into how user’s data is stored, revealing a lack of trust in the international community for the technology’s security, and thus a potential for lower business engagement with the technology. 

Where is the EU in the AI race? 

The EU has long been seen as a ‘regulatory power’, where the protection of consumer rights and maintenance of ‘European standards’ forms part of the EU’s core identity and mission. European citizens have typically endorsed this level of regulation as a reflection of the European values of freedom and civil liberties, in contrast to the relatively low consumer protection law in rival markets such as the U.S. However, critics have cast this regulatory landscape as a critical hindrance to competition and innovation, particularly in the tech sector. 

Until recently the main legislative response of the EU to AI was the 2024 AI Act, which was quite cutting edge as the first major law passed on governing artificial intelligence anywhere in the world. The legislation categorised applications of AI to three risk categories with corresponding legal requirements, in an effort to protect European citizens from unacceptable risks. However, this legislation focused quite clearly on the EU’s role as ‘adjudicator’ rather than ‘innovator’ when it comes to global tech developments. In fact, the AI Act was criticised on this basis by member states such as France who were early AI innovators, with domestic policy such as Macron’s 2018 National AI Strategy focusing on investment rather than regulation. Many would connect this favourable regulatory landscape to the fact that France’s Mistral AI has emerged as Europe’s champion LLM,  whilst the European market as a whole struggles to compete with markets in the U.S and China. 

However, there has been a notable shift in the EU’s approach to AI in the past six months, beginning with the release of the Draghi Report on European Competitiveness in September 2024. In this report requested by the European Commission, former European Central Bank President Mario Draghi outlines that Europe needs to rethink its growth strategy in order to remain wealthy and prosperous, with the first key challenge being innovation. He suggests an economic paradigm shift, including minimising regulatory and financial barriers to entry for business and increasing investment – both public and private. 

As a result of this paradigm shift that is underway in the EU, commission President Ursula von der Leyen announced the release of a new ‘EU Competitiveness Compass’ in January, which is based on the recommendations of the Draghi report. The very first pillar of the Compass is ‘closing the innovation gap’, which proposes ‘AI gigafactories’ and ‘Apply AI’ initiatives which will push AI development and adoption in key sectors, such as advanced materials, quantum, biotech, robotics and space technologies. Moreover, the InvestAI initiative will mobilise €200 billion of investment in AI. The compass also outlines an ‘EU Startup and Scale Up’ strategy to address obstacles for startups in the EU, including simplification of regulation and reduction of the costs of failure. This new framework marks a significant departure from prior approaches to AI from within the EU which focused primarily on regulation, as the commission recognises the potential of AI to be “a force for good and for growth” in the EU. 

Implications of Deepseek’s R1 for the EU

The development of Deepseek’s R1 model can be seen as an advantage to Europe in the AI race, in the context of the current paradigm shift that the EU is undergoing. Firstly, the biggest ‘losers’ in the fallout of Deepseek’s release of R1 have unequivocally been the tech firms who have already invested billions into training and developing large language models. These are primarily firms in the U.S such as Nvidia, Apple, Amazon Web Services, Microsoft, Google, etc, who are the EU’s major competition in AI technologies. Although these firms have had a ‘head start’ on AI development as compared to European firms, they have also (as per the conventional wisdom prior to January 27th) invested extreme amounts of money into training their models. European startups, on the other hand, have not had access to these levels of capital investment, yet they will also not suffer from this sunk cost in the post-Deepseek AI landscape. Indeed, AI developments in the EU going forward under the Competitiveness Compass measures will benefit from the lessons of Deepseek’s technology, and likely be able to train their AI models at a much reduced cost. 

Moreover, due to the fact that Deepseek caused Nvidia to suffer the greatest single-day loss of value for any public company in history, it is clear that the tech developments made by Deepseek have generated anxiety for U.S investors. This uncertainty, paired with the ability of Deepseek to develop a similar product using much less money has caused U.S investors to be hesitant to continue to invest the billions of dollars that tech companies had been telling them was necessary to train sophisticated large-language models. This wavering faith in U.S markets and discussions about curtailing investments once again benefits European startups, especially in the context of the EU finally clearing away doubts about AI and beginning to invest heavily in it. 

Further, Deepseek’s ability to develop its R1 model on such low costs may also benefit investor confidence within Europe, as it has broken down the extreme barriers to entry that were previously associated with AI. Deepseek has demonstrated to private investors within Europe that the AI race is not over, and European startups can still compete on the world stage without requiring the level of investment that is available in the U.S. 

Finally, the limitations associated with Deepseek mean that for all its innovations, it is not a competitor to European AI startups. This is because above all, the EU is positioning European AI as credible, reliable, and human centred – a trustworthy alternative to the U.S – and in this domain, Deepseek cannot compete. Indeed, given the data protection concerns already expressed across the world, Deepseek’s ban on government devices in Australia and the input-bias of the technology due to Chinese censorship laws, Deepseek does not offer a credible alternative to U.S firms when it comes to business, government or security applications. 

Conclusion

Deepseek’s R1 model has disrupted the global AI landscape, challenging U.S. dominance and dealing a significant blow to major American tech firms. However, its low-cost innovation presents a unique opportunity for the EU. Unlike U.S. firms, European companies have less to lose from Deepseek’s emergence but much to gain from its technological breakthroughs, particularly in reducing the cost of training sophisticated LLMs. At the same time, Deepseek’s limitations—such as censorship laws and data security concerns—prevent it from competing with the EU’s niche in credible, reliable, and ethical AI. As the EU undergoes a paradigm shift toward fostering AI innovation, these developments may help Europe close the gap with the U.S. and solidify its role as a key player in the global AI race, proving that the competition for AI leadership is far from over.