DeepSeek’s AI breakthrough challenges US dominance, disrupts global competition, and raises urgent policy questions for Europe.
The recent launch of new Artificial Intelligence (AI) models by Chinese startup DeepSeek has disrupted the global AI landscape and raised significant questions. Until now, the prevailing narrative held that developing large language models required billions of dollars in high-end AI chips, making the field accessible only to the largest US-based technology firms, such as Microsoft, Meta, and Google.
DeepSeek has overturned this assumption by releasing two AI models, V3 and R1, that rival the latest systems from OpenAI and Meta while being significantly cheaper to develop and operate. The total cost of training these models is estimated at six million US dollars, with operating expenses between twenty and fifty times lower than those of their Western counterparts. On 27 January, the DeepSeek AI chatbot, powered by the DeepSeek-V3 model, became the most downloaded application on the Apple App Store. Meanwhile, the DeepSeek-R1 model, designed for advanced reasoning tasks in mathematics, science, and computer programming, has drawn the attention of experts worldwide.
DeepSeek’s breakthrough stems from a novel approach to training large language models. In a research paper published on 22 January, the company outlined how it enhanced AI reasoning capabilities through a methodology combining multi-stage training, pure reinforcement learning, and self-evolution. In the same publication, DeepSeek announced the open-sourcing of DeepSeek-R1-Zero, DeepSeek-R1, and six distilled models derived from DeepSeek-R1. By enabling developers to customise these models, DeepSeek fosters collaborative innovation and builds a broader ecosystem of programmers and users. This strategy sets it apart from competitors such as OpenAI, which has largely resisted open-source approaches.
The United States’ Stargate Project
DeepSeek’s advances coincide with the launch of the United States’ four-year, five-hundred-billion-dollar Stargate Project. Described by President Trump as a private sector initiative designed to promote innovation, Stargate seeks to limit regulatory burdens, enhance the country’s AI infrastructure, and address the energy challenges posed by AI’s growing computational demands. Data centres, which are integral to AI model training and deployment, consume vast amounts of electricity, raising concerns about power grid sustainability.
Beyond financial investment, the Trump administration has also introduced a distinct AI policy framework. A recently issued executive order rescinded previous directives that prioritised AI safety, citing them as obstacles to American innovation. This policy shift signals a move towards a deregulated, market-driven AI sector, departing from commitments made in transatlantic cooperation initiatives such as the EU-US Trade and Technology Council and the Council of Europe’s Framework Convention on Artificial Intelligence and Human Rights, Democracy, and the Rule of Law. The latter, signed in Vilnius on 5 September 2024, had seen active US participation in shaping global AI governance. By reversing its stance, the United States risks undermining common international standards rooted in shared democratic values, including human rights protections.
The US administration’s objective is to maintain a competitive edge over China. However, DeepSeek’s emergence complicates this strategy, raising questions about its true development costs and the nature of the computing resources available to the company. Regardless of these uncertainties, DeepSeek’s entry into the AI race has reshaped the geopolitical landscape, narrowing the technological gap between the United States and China. It also casts doubt on the effectiveness of Washington’s deregulation-led approach, which depends on large-scale private investment and ever-expanding natural resource consumption.
Can the European Union Bridge the AI Divide?
In this contest between global powers, the European Union faces a critical question: can it compete in AI innovation? In his recent report on European competitiveness, Mario Draghi identified AI as an opportunity to address Europe’s lagging innovation and productivity while revitalising its manufacturing base. The Competitiveness Compass, unveiled by European Commission President Ursula von der Leyen on 28 January, explicitly calls for the EU to lead in AI development. The initiative, primarily focused on regulatory simplification, includes up to four billion euros in AI research and deployment funding for the period from 2024 to 2027. It outlines measures such as providing AI startups and industry with access to supercomputing resources through an AI Factories initiative, promoting industrial and public-sector AI applications via an Apply AI Strategy, and consolidating efforts under a new European AI Research Council.
However, this strategy raises concerns about the future of the EU’s AI Act. Although designed to foster European AI innovation, the Act may be at risk of dilution. Its legal provisions, harmonised standards, and forthcoming guidance instruments face potential simplification, which could undermine their core objective of ensuring safe and trustworthy AI for both society and the environment.
For AI to deliver tangible benefits to European citizens and workers, the European Commission must ensure that the AI Act remains a robust regulatory framework rather than becoming a symbolic measure. While Europe lags behind both China, which has rapidly repositioned itself, and the United States, where private companies are investing at an unprecedented scale, technological capability alone is not the defining issue.
Instead, the focus must be on AI’s societal impact—on democracy, the rule of law, employment, and fundamental European values. DeepSeek’s rise underscores how quickly the AI landscape can be disrupted. In this evolving environment, Europe’s priority should be to maintain a strong governance framework and implement digital regulations that prioritise AI safety. By doing so, it can ensure that AI’s benefits are broadly shared while mitigating its economic, social, and environmental risks.
Aida Ponce Del Castillo is a senior researcher at the European Trade Union Institute.