Open AI closes the gap as Europe bets on sovereignty and infrastructure

Mozilla's latest report reveals that open-source and open-weight AI models have progressed from experimental stages to becoming integral components of the global AI landscape. The September 2026 assessment evaluates the state of open AI across capability, adoption, infrastructure, investment, regulation, and sovereignty, with a focus on Europe's progress.

Open AI transitions from a promising to a ready state, as demonstrated by various European developments, such as an open medical model developed in Lausanne in collaboration with the Red Cross. Mozilla's Chief Technology Officer, Raffi Krikorian, highlights several instances where European researchers independently created open AI models without seeking permission or renting them from others.

On Hugging Face, there are 2.5 million public models, 13 million users, and a third of the Fortune 500. OpenRouter, where developers route real production traffic, saw open-weight models increase from a minority to nearly a third by late 2025, accounting for 25 trillion tokens weekly. The report concludes that open models are "ready" for deployment, with the best closed model scoring 60 compared to 54 for the best open models.

Although Europe participates in open AI, it does not lead at the frontier, with Chinese models like Kimi K3, GLM-5.3, and Qwen 3.8 dominating. European player Mistral is further behind in capability, with a modified MIT license that imposes revenue restrictions. The report suggests that open AI can provide European organizations with technological sovereignty, enabling them to maintain ownership and control over models, data, and infrastructure.

In Switzerland, a public consortium successfully trained a national model using public supercomputers and released the weights, data, and training code, making it widely accessible for copying and adapting. However, only 51% of open-model users deploy them into production, compared to 63% for closed models, suggesting a gap in deployment infrastructure, trust, and operational tooling.

Europe's opportunity lies in deployment rather than model capability, with companies focusing on deployment infrastructure, orchestration, compliance, security, and enterprise tooling. The UK is investing £500 million in sovereign AI through the Lumen Sovereign project, trained on proprietary datasets covering over 30 regulated workflows for air-gapped deployment within customers' infrastructure by the end of 2026.

This move indicates growing demand for sovereign AI in the UK, with the Cohere company planning to triple its office footprint in the country. The EU AI Act, which uses training compute as a proxy for systemic risk, may struggle to keep up with increasingly efficient models, prompting further analysis of the regulatory framework's effectiveness.