Sovereign AI

Four companies control the world's AI. That's a dependency problem, not a technology problem.

If OpenAI, Google, Anthropic, or xAI changes their model, their pricing, or their terms of service, your business adapts. If they decide not to serve you, you're done. You don't own your intelligence. You're renting it from someone who can evict you.

I wrote about where this leads in how the software industry changes when development costs near zero. When building software becomes cheap, generic SaaS loses to custom software built per company. The same shift applies to intelligence itself: when AI is powerful enough to run a company's operations, renting a generic brain from four vendors won't be competitive with owning one that knows your business.

I also wrote about the path from LLMs to AGI. We're closer than most people think, and the mechanical path is clearer than the philosophical debate suggests.

The thesis: every company needs its own AI, trained on its own data, running on its own infrastructure, under its own control. Not a rented API. Not a shared model. A sovereign intelligence that belongs to the company the way its codebase and customer list do.

This is what I'm building with Aether. A runtime that lets any company run its own agent, on a server it controls, with full visibility into what it does and why. The long-term goal is a network of independent, cooperating agents, each owned by the entity it serves.

The technology for this exists. The bottleneck isn't capability. It's trust: companies need to trust their own AI the way they trust their own employees. That's a sovereignty problem, not a model-size problem.

Get in touch

• If this thesis resonates, whether you're an investor who sees the shift, a company that wants to own its intelligence, or an engineer who wants to build it, let's talk.