The A1 Manifesto
AI Should Belong to the People
AI is the defining technology of this era. The decision that matters is who controls it.
Right now, the answer is a handful of companies. It does not have to be.
It's Not Magic
An AI model is compute: large matrices of numbers multiplied at scale on standard chips. There is no special ingredient, and nothing about it requires a remote facility.
Yet today it runs inside a few corporate data centers. They consume the water and electricity of small cities, and they meter access back to you by the month and by the token.
It Was Trained on Your Data
These models were trained on public and personal data at scale: text, images, code, and queries produced by billions of people. That collective output was enclosed, and access to it is now sold as a subscription.
Centralization Is the Risk
When every query routes through a few providers, those providers hold full visibility and full control: what the model answers, what it refuses, and what changes between releases. That is concentrated power over information.
Critical infrastructure shouldn't sit behind a single gatekeeper, and access to it shouldn't require a recurring fee.
The Alternative
Distribute the compute. The model runs across a network of independent servers operated by many different people, instead of one company's data centers.
Each server is a node, contributing a share of the work and connecting to others by choice. No central authority, no paywall, no single point of failure.
The Common Objections
The claim that AI can't run outside Big Tech echoes earlier claims about the personal computer, the internet, and smartphones. Each one proved false.
The claim that only a centralized data center has enough capacity ignores the math: a network of independent servers holds more aggregate compute than any single facility, and a distributed network has no central switch to shut off.
What This Requires
Run a node, or use the network instead of a centralized provider. The barrier is adoption, not capability.
The compute is distributed. The training data was collective. Control of the result should be too.
"Infrastructure this important should not be owned by a few."
The technology to decentralize AI exists today. What remains is the decision to use it.
- The A1 working group