Europe keeps losing the race to build the biggest closed AI models. The smarter play: dominate the open model ecosystem instead.
Closed models from American and Chinese labs still beat open alternatives on raw performance benchmarks. Yet open models win on cost, speed, efficiency, and the freedom to customize. Those advantages matter more for a continent that controls only about 5 percent of global AI computing power. U.S. investors poured roughly ten times more into AI startups than their European counterparts in 2025.
The EU has backed open source for over a decade. A 2014 European Commission strategy promoted sharing code to boost digital autonomy. The bloc doubled down in 2025 with its Apply AI Strategy, which puts open source AI solutions at the center of adoption plans.
Open-weight models share parameters but keep training data and methodology private. That compromise lets developers fine-tune systems for local languages, industries, and cultural contexts. Nearly all developers surveyed by the Linux Foundation have tried open models. Close to two-thirds of companies now use one. Broader access also means more researchers can test models for safety flaws, not just the companies that built them.
Building a frontier closed model would cost Europe more than €800 billion over three years, according to a consortium of researchers and former policymakers. That won’t happen. However, Europe already has serious open model players, including Mistral AI in France and Black Forest Labs in Germany. Its universities and industrial firms provide additional assets. The continent’s open source roots run deep: Linux and HTTP both came from European developers.
China’s open model push has attracted a global community drawn to cheap alternatives. The performance gap between open and closed systems has narrowed to roughly four months. But Western governments don’t want dependence on models trained under CCP principles. Europe could anchor a trusted alternative.
Two policy moves would help. First, fix legal frameworks designed for proprietary software that now constrain open source adoption. The upcoming copyright proposal needs care to avoid chilling open-weight development. Second, update procurement rules so open source can compete fairly with proprietary vendors. Current tender templates often demand warranties and licenses that open projects can’t provide.
Rather than chase a race it cannot win, Europe should build the ecosystem where open models flourish.
