DeepSeek: New hope for Europe and UK in the AI arms race?

The start of 2025 has seen major shifts in the global AI landscape. The launch of the reasoning model "DeepSeek R1" (R1) by a China-based start-up of the same name (DeepSeek) has attracted significant attention across technical, commercial, as well as political fronts. With its novel low-cost techniques and open-source distribution, R1 has challenged conventional wisdom around the limits of AI development and re-ignited ongoing conversations about the intensifying AI arms race between the US and China, as well as the competing trends between open and closed source AI. In light of these emerging dynamics, this article explores the implications of the recent developments around DeepSeek for both the UK and European Union.

DeepSeek's disruptive impact

On 20 January 2025, DeepSeek introduced R1, a new AI model that demonstrated performance on par with industry leaders. The launch triggered a sharp reaction in the markets at the time, with Chipmaker Nvidia losing $589 billion (17% of its value) in a single day, and the tech-weighted Nasdaq dropping by over 3% alone. The impact was partly due to DeepSeek’s claims that R1’s training costs were under $6 million, a fraction of the amount required to train comparable incumbent models, and that the model runs on a limited set of legacy microchips owing to US export restrictions on China. However, the accuracy of these claims has been questioned, with critics arguing out that the $6 million figure only covers a narrow portion of R1’s total training costs and excluded key research and development expenses.

Nevertheless, DeepSeek’s innovative use of more efficient training techniques, which require significantly less data points, and reduced compute requirements have been widely recognised. Additionally, the distribution of R1 via open source – with all its weights and source code available without charge for the global public to download and deploy – represents a disruption to an incumbent business model of pools of ‘closed source’ models and their subscription-based pricing. This shift could have the potential to alter the global dynamics of AI development.

Historically, the US’s financial power has enabled it to lead AI development. In 2024 alone, the US saw $80.8 billion in AI and machine learning venture capital deals, compared to $12.8 billion across Europe as a whole. One of President Trump's first major announcements upon returning to the White House was an unprecedented $500 billion investment package into AI infrastructure (Project Stargate). However, the release of R1 defies the prevailing view that creating an advanced AI model is reserved to those with the largest financial and technical resources (and in particular the most advanced next generation chips). What's more, China shows no signs of abating as evidenced by the launch of Manus AI by Chinese startup Butterfly Effect in early March, which some are calling the world's first general AI agent. In effect, it is understood that Manus can act autonomously and dictate its own actions once assigned a task, without needing further human prompting, thus offering a glimpse into what a truly general artificial intelligence could look like. The MIT Technology Review published an article on Manus which explored its capabilities.

These lower-cost development model, which build on existing open-source technology and share its innovations, could provide a valuable opportunity for "deployer" nations such as the UK and those within the EU to enhance their AI competitiveness in a global tech race so far dominated by the "developer" powerhouses of the US and China. This aligns with the UK government’s ambition, unveiled earlier this year in its AI Opportunities Action Plan, to develop sovereign AI.

With R1 priced 20 to 30 times cheaper than its competitors, it also opens doors for UK and EU based start-ups in the "application layer (i.e. companies that integrate AI models, such as large models, into specific use cases) to access advanced AI technology with significantly lower capital investments. R1 was quickly embraced by some European start-ups despite the data privacy concerns of national authorities. Data privacy authorities in France, Belgium, Croatia, Ireland, Germany, the Netherlands, and Italy have all launched investigations into or requested information from DeepSeek, with Italy going furthest by blocking access to the app.¹

Seizing the AI opportunity: Will the UK or EU lead the way?

While it is still early to assess the impact of DeepSeek on the UK or EU, the two jurisdictions have taken different approaches to AI that may shape how they capitalise on the opportunities presented by R1.

UK

According to Pitchbook's 2024 Venture Report, the UK is Europe’s largest AI ecosystem, it ranks fourth globally in the Global AI Index and is home to some of the world’s leading research universities. However, the House of Lords Communications and Digital Committee has recently highlighted challenges in scaling the UK’s homegrown tech companies, citing limited access to domestic growth capital and conservative attitudes towards investment and entrepreneurship as key barriers. The Committee warned that, without "bold policy reform", the UK risks becoming an "incubator economy" for other nations. In an effort to address these structural challenges, the Government has announced plans to ease restrictions on the use of surpluses in defined benefit schemes (read our briefing here) and to create pension "mega funds" by consolidating defined contribution schemes and pooling assets from Local Government Pension Schemes. The scale and pace of these reforms will become clearer when the Pensions Schemes Bill is introduced later this year.

In terms of AI regulation, the UK has adopted a light-touch, principles-based approach, opting for flexibility rather than prescriptive legislation. Under then Prime Minister Sunak, the previous Government led by the Conservatives established five cross-sector principles for existing regulators to interpret and apply within their respective domains. Although the King's Speech on July 17, 2024, proposed "binding measures" for the "most powerful [AI] models", Labour has so far maintained a "pro-innovation" stance, emphasising oversight by existing regulators. In its response to the AI Opportunities Action Plan (the Response), the Government committed to funding sector-specific regulators to scale up their AI capabilities and issuing strategic guidance to ensure safe AI innovation. Further evidence of the UK Government's cautious stance on AI regulation came recently when it declined to sign the statement issued after the Paris AI Conference, citing concerns about national security and the lack of specific enforcement mechanisms, favouring a more adaptable, industry-driven regulatory approach. Additionally, the UK has delayed introducing its AI Bill until at least summer 2025, partly to align with the US’s deregulatory stance and avoid discouraging AI investment.

EU

The EU, by contrast, enacted the EU Artificial Intelligence Act (AIA), one of the world's first comprehensive regulatory frameworks for AI. The AIA adopts a functional "risk-based" approach, imposing varying levels of obligations on AI system providers (broadly the entities that develop AI) and deployers (broadly the entities that use AI for professional activities) depending on the use of the AI. Certain AI systems, deemed to present unacceptable risks to health, safety and fundamental rights, are prohibited altogether.² This includes those used for real time biometric identification, social scoring and cognitive behaviour manipulation. The AIA categorises all other AI systems as high, limited or minimal risk and imposes regulations and transparency requirements on the first two categories respectively. Read our briefing on the implementation of the first provisions of the AIA here.

However, there have recently been signs of a change of tone from the EU. In February, the European Commission withdrew its planned AI liability directive as the Commissioner for Tech Sovereign, Security and Democracy, Henna Virkkunen, announced that the EU is "committed to cutting bureaucracy and red tape". The upcoming AI Codes of Conduct (secondary guidance and technical standards designed to help companies comply with the AIA) Virkkunen said, will be designed to "help and support" AI companies, not create further reporting obligations for them. This change in regulatory tone comes alongside a major increase in investment. At the 2025 Paris AI Summit, the President of the Commission launched InvestAI, a public-private partnership mobilising €200 billion of investment into four key areas: financing AI gigafactories, a €20 billion fund for AI startups and research institutions, streamlining data access across the EU and AI talent development. President Macron also announced a €109 billion investment into French AI infrastructure, describing it as France's equivalent to Project Stargate.

More recently, the Commission announced the "AI continent" action plan (the Plan), looking to leverage the EU's strengths and place the continent at the forefront of the next phase of AI development. The Plan comes with concrete targets, including setting up at least 13 "AI factories" (collaborative ecosystems that unite computing power, data, and talent to drive innovation and development in AI across various sectors in Europe) and investing €20 billion to help establish 5 AI "gigafactories", large-scale facilities with significant computing power and data centres. The Plan also include supporting companies and EU countries with the implementation of the AIA, launching an AI Act Service Desk to serve as a point of contact for businesses seeking clarification and guidance on the AIA. Together, these efforts suggest a shift in the EU's focus, prioritising AI investment alongside prudent regulations.

Conclusion  

The launch of R1 has shifted the dynamics of the global AI race. By offering a low-cost, open-source alternative to incumbent models, subject to any potential national security concerns, it presents a unique opportunity for deployer states like the UK and those within the EU to close the technology gap and strengthen their domestic AI sovereignty. It remains to be seen which region – the UK with its light-touch regulatory approach and efforts to unlock domestic capital, or the EU with its focus on balancing regulation and investment – will be more successful in capitalising on this opportunity.