The entire tech industry, save for Anthropic, has come out in favor of open source AI. This isn't a niche stance: companies like Meta, Mistral, Stability AI, and others have publicly backed open models. Anthropic's opposition, rooted in safety concerns, now faces a reality where its lobbying push seems unlikely to change course. But the phrase "Nobody is trying to ban open source" is a textbook case of gaslighting, as regulatory proposals often include heavy-handed compliance that effectively shuts down self-hosted deployments.

For enterprises deploying LLMs on-premise, this isn't a distant policy debate — it's a direct threat to their infrastructure strategies. Open source models are the backbone of self-hosted AI: without unrestricted access to models like Llama, Mixtral, or Qwen, organizations cannot build sovereign pipelines that keep data within their own facilities. Quantization, VRAM optimization, and local inference are the tools that make on-prem AI viable, but they depend on the freedom to distribute and modify model weights. Regulatory capture could quickly turn local inference into a liability nightmare, forcing companies back to cloud APIs, undermining TCO calculations and data residency requirements.

The gaslighting tactic is insidious: proponents of stricter rules claim they only want "responsible" AI, but the mechanisms under discussion — pre-release model evaluations, mandatory licensing, liability for downstream misuse — would disproportionately crush the decentralized ecosystem. A small startup or a mid-sized enterprise cannot afford the legal and technical overhead that such regulations impose. Yet the hyperscalers and closed-model vendors, with their massive compliance budgets, would sail through. The result is a market locked into a few providers, exactly the opposite of what Europe's GDPR and data sovereignty movements demand.

What does this mean for hardware and deployment? If open source models become harder to access legally, the demand for high-memory GPUs like the A100 or H100, which are purchased specifically for self-hosting, could shift. Instead, enterprises might double down on proprietary APIs, reducing the need for local compute. This would alter the competitive landscape in the AI hardware supply chain, affecting everything from NVIDIA's enterprise GPU sales to the viability of on-prem solutions offered by vendors like Dell or HPE with integrated AI stacks.

AI-RADAR's analysis of LLM deployment consistently points to a fundamental truth: open source is not a luxury, it's the enabler of true sovereignty. Without it, on-premise AI is reduced to a fragile adjunct, dependent on the goodwill of a few key players. The current lobbying battle will shape the next decade of enterprise AI infrastructure. The question isn't just whether Anthropic will change its stance — it won't — but whether the regulatory frameworks being crafted today will allow the self-hosted ecosystem to survive. The gaslighting has begun, and the real victims might be the very organizations that need AI to work on their own terms.