Europe Should Double Down on Its Open-Source AI Bet

At the frontier of AI, closed models currently outperform open models on the most demanding measures of raw model capability, even though open models can offer significant advantages in cost, efficiency, speed, and customizability. But building the world’s most powerful closed models requires enormous concentrations of compute, access to proprietary data, and large user bases that generate valuable feedback and deployment data—all areas where Europe is unlikely to have an advantage, in part because its regulatory policies make it harder to achieve the necessary scale. Europe’s best hope of relevance in the AI race, therefore, may not be to build the world’s most powerful model, but to become a leading player in the strongest ecosystem of open models. The EU has already identified open models as an important part of its tech sovereignty ambitions. Now it should double down.

The EU has long recognised the value of open-source technology. The European Commission launched an open-source software strategy in 2014 which focuses on sharing data and reusing software to promote European digital autonomy. The EU’s latest Open Source Strategy, launched earlier this year, once again “places open source at the center” of the EU’s plans to secure technological power. In 2025, the EU launched its Apply AI Strategy, which has “a focus on open source AI solutions” to boost AI adoption.

Open-source software is fully transparent and freely accessible: anyone can download, study, modify, use, and share it. Many organizations do not make AI models fully open source because they train them on sensitive or proprietary data that they cannot share freely. For this reason, providers of so-called “open-weight” models opt to share some of the model’s code and parameters (weights) but withhold training code, data sources, and methodology.

Although openness in AI exists along a spectrum, more open models offer several advantages over closed models. Developers can more easily deploy and build novel applications with open models because they can freely access, adapt, and fine-tune them to specific needs. Their appeal is already evident: nearly all developers surveyed by the Linux Foundation last year have experimented with open models, while nearly two-thirds of surveyed companies are using an open model. Open models can also strengthen the AI safety ecosystem by allowing more researchers to inspect, test, and red-team models rather than relying on the assessments of their developers. A group of researchers has described open-source models as “indispensable for the safe development and deployment” of AI because they enable broader monitoring and auditing.

Europe does not have a world-leading closed model, such as the United States’ Claude Fable 5.1 or China’s Seed2.1 Pro. A consortium of researchers and former policymakers estimates that building one would cost over €800 billion over the next three years. That doesn’t mean Europe should give up on deploying or integrating closed-weight models, especially in specific use cases where they are best suited. But leaning into open models may be the only way Europe can have a meaningful advance in AI—and that’s no bad thing.

As it happens, open models suit Europe well. Europe has a tradition of open-source infrastructure. European open-source software and protocols such as Linux and HTTP formed the basis of the Internet. Today, it has key players like France’s Mistral AI and Germany’s Black Forest Labs, whose open-weight models have attracted substantial use and attention. The distributed nature of open-source ecosystems means developers can adapt a model to specific languages, culture, and sectors.

Meanwhile, the absence of a winner-takes-all model reduces the importance of concentrating enormous amounts of compute and capital in a handful of frontier labs—an area where Europe is at a substantial disadvantage. Europe has about 5 percent of global AI compute capacity, compared with roughly 75 percent in the United States, and U.S. investors accounted for roughly ten times as much AI venture capital investment as did EU investors in 2025. But Europe has other assets: strong universities, public institutions, and specialized industrial firms that can develop and deploy open models.

China’s embrace of open models has allowed it to attract a global community of contributors who value low-cost open-source alternatives. These alternatives do not lag far behind—the gap has closed to around four months. However, the West does not want to develop a technology dependence on China’s authoritarian regime, nor use AI products that have been trained and tuned in accordance with CCP principles. Europe could be a major player in building a trusted alternative to China’s open-source ecosystem that is not tied to any one nation.

Europe would not be taking on China alone. The United States has open-model AI companies like ReflectionAI and Liquid AI, and many of its leading tech companies have released popular open models, including Meta, Google, Nvidia, and AMD. And, according to Stanford’s 2026 AI Index Report, U.S.-based AI projects on the open-source repository GitHub still attract the most engagement, although the U.S. share of projects has dropped from 80 percent 15 years ago to under a third today, whereas the European share has remained constant at around 25 percent.

To double down on open models, the EU should do two things. First, it should ensure licensing, intellectual property, liability, and data rules do not discourage use of open models. Although the EU’s AI Act sets out clear rules, guidelines, and exemptions for open-source AI, a 2026 European Commission report on open-source software says “existing legal frameworks were designed with proprietary software models in mind and do not adequately reflect the collaborative and distributed nature of open source development,” resulting in “restrictive interpretations” and limited adoption. For example, EU policymakers should ensure that the upcoming proposal on copyright and AI does not discourage the development and use of open-weight models.

Second, the EU should revise its procurement rules to ensure that open-source solutions can compete on a level playing field with proprietary software, rather than giving them preferential treatment. The Commission’s report on open-source identifies procurement policies as a barrier, including warranty and guarantees that open-source projects may not be able to provide, and restrictions on acceptable licenses. The Open Source Initiative similarly argues that government tender templates and procurement practices are often designed around proprietary software, while failing to account for open source’s potential advantages in interoperability, reuse, and cost.

Recognising the strategic importance of open models to Europe’s sovereignty ambitions, it should update procurement frameworks to allow their use on equal footing with proprietary solutions. It should resist, however, creating a “Europe-only” club. Open-source ecosystems thrive best when they attract global developers and adopters. A Western alliance of democracies that share interoperable interfaces, evaluation protocols, and safety standards would make more sense for Europe.

The EU has already recognised the importance of open models. Rather than chasing a losing race to develop closed frontier models, the bloc should focus on facilitating a thriving ecosystem for open AI models. Leaning into open models is the best hope for the bloc to have a seat at the global AI table.

Image credit for social preview: Canva