Evidence: Product Moat · 4 live Netlify sites on sovereign infrastructure
The Sovereign AI Infrastructure Manifesto
Independent AI safety researcher, London. Verify any claim.
Vercel suspended my account. One billing dispute. One automated decision. Everything went dark overnight, production sites, preview deployments, CI pipelines. The infrastructure I had depended on was not mine. It was rented. And the landlord could change the locks without notice.
That was the moment I decided: never again. Everything that matters in this programme now runs on infrastructure we control. Four Netlify sites (we own the DNS), three physical machines (we own the hardware), a constitutional gateway on our own server, and a Python daemon that does not phone home. The code is in our repos. The data is on our disks. The gates are enforced by our rules.
What Sovereign Means
Sovereign does not mean "we built everything from scratch." It means we control the dependencies. Netlify hosts static sites, if Netlify disappears, the same static files deploy to any CDN. Supabase stores data, if Supabase disappears, the same Postgres schema runs on any server. The gateway runs on a Mac in a room. The daemon runs on the same Mac.
The principle is: every dependency must be replaceable within 24 hours without data loss and without downtime exceeding the time it takes to repoint DNS. This is not paranoia. This is the lesson of having a cloud provider suspend your account with zero warning.
The Alternative
The standard AI startup playbook: rent everything. Rent the compute from AWS. Rent the database from Supabase. Rent the deployment from Vercel. Rent the AI from OpenAI. None of it is yours. Any one of those providers can suspend your account, change their pricing, or go out of business, and your entire product disappears overnight.
I learned this lesson the hard way. You do not have to. Build on infrastructure you control. Own your dependencies. Make every layer replaceable. The sovereignty is the safety margin, not just for the AI, but for the business that depends on it.