solution · S06

Open-Source AI & Antitrust Enforcement

Combining open-source AI model development with antitrust enforcement to prevent monopolistic control of AI capabilities and ensure broad access.

Updated 2026-09-20

Observatory assessment: 42 / 100. Scaling

How assessments are made

Deep dive

Open-source AI models have emerged as the primary counterweight to corporate AI concentration. A global ecosystem of developers, including Meta (LLaMA), Mistral AI, and China's Moonshot AI (Kimi), demonstrates that competitive AI can be developed outside the closed Western corporate lab paradigm. In 2024-2025, top open-source models consistently matched proprietary performance on many benchmarks.

However, open-source models alone are insufficient without antitrust enforcement and diversification of the hardware layer. The vertical integration of AI companies—controlling compute (cloud), models (APIs), and applications—is facing challenges from regulators like the FTC and EU, though recent analysis suggests existing merger laws are ill-equipped to handle the novel ownership structures in the AI sector. This is compounded by new hardware players like Zhipu AI reducing reliance on a handful of chip designers, and national strategies like 'Swiss Sovereign AI' signaling new forms of public intervention. The challenge remains to coordinate these technological, regulatory, and national efforts effectively.

Trajectory and projections

Near term

By 2026, open-source models achieve frontier performance for most tasks. EU antitrust investigations into AI market structure conclude.

Mid term

By 2030, publicly funded foundation models (EU CERN-for-AI initiative) provide a non-commercial alternative. Mandatory API interoperability is enforced.

Long term

By 2035, a competitive AI ecosystem exists with 50+ capable model providers. AI is treated as critical infrastructure with regulated access requirements.

Key players

• Meta (LLaMA open-source models)
• Mistral AI (European open-source leader)
• Hugging Face (model distribution platform)
• FTC & EU DG Competition (antitrust enforcement)
• Linux Foundation (AI & Data governance)
• Moonshot AI (Kimi open-source models)
• Zhipu AI (hardware & model development)
• National AI Initiatives (e.g., Swiss Sovereign AI)
• Public Good Advocates (e.g., Bruce Schneier)

Barriers

• Open-source models still require massive compute to train and fine-tune
• Safety concerns: open models can be fine-tuned to remove guardrails
• Antitrust law struggles with platform dynamics in AI markets
• Corporate lobbying against regulatory intervention
• Privacy erosion from 'always-on' AI tools integrated into core software
• Policy uncertainty from the 'Open-Weight Paradox', where restricting models for safety may undermine broader security research and innovation.
• Inadequacy of existing merger and antitrust laws to address novel AI ownership and integration structures.
• Debate over the sufficiency of 'open-weight' models versus 'truly open-source' for achieving societal goals.

Top evidence