The fork in the road.

Look back at the big turns in technology and you see the same fork again and again. For a while, something new is wide open. Anyone can build on it, and the ideas pour in.

Then, as we see it, one path wins. A few companies grow big enough to set the rules, and everyone who comes after pays a toll to get through their gate. The builders, the businesses and the customers all end up inside a couple of walled gardens, and it slows down the people with the next good idea.

We believe AI is standing at that fork right now. This time, we think we can take the other path.

Each of these was a great technology. Each one started open. And each time, most people ended up inside a few walled gardens.

Before going any further, I want to say this plainly: the AI companies are creating a lot of value, and we should not deny it.

Their models help people write, research, plan, build software and do real work across a business. A lot of people are hating on AI way too much right now. We disagree with them.

Nothing in this article is an argument against AI, or against the companies that build it. We believe every business should use their models. The question is not whether to use AI. It is where your business keeps the thing it builds with it.

They are worth looking at together. Only one of them is the problem this Foundation was started to solve, but a business owner should know both.

Trend one: your data helps improve their models.

Some AI services can use what customers send them to improve their models. Whether they do depends on the service, the plan you are on, a setting, or the terms you agreed to. Terms vary from one service to the next, and they change over time.

So read them. Even when the use is small, or anonymised, you are helping the tool get better as you use it. There is a little symbiosis in that, and many people will find it a fair trade. But it is a trade, and as a business owner you should make it on purpose.

This is a real concern, and it is not the one we solve. What we do make easier is switching: a factory you own can move work to an open model you run yourself. We believe that, once you look at the cost, open models are not yet worth it for most businesses for most work. The option still matters.

Trend two: the model companies want to hold your factory.

The second trend is the one we exist to answer. The companies that make AI models are becoming product companies too. The biggest AI companies are launching products that ask you to keep your skills, tasks and agents on their platform. They are seamless, easy to use, and quick to start.

What you build there is not a chat. It is your operating system, what we now call a factory: every skill, agent, workflow and recurring task that runs your business, often with your data, and sometimes with your customers. And it sits where that company, and its models, can see it.

Here is why that matters. Once an AI can read all of your data and understands your processes, it can come close to running your business. It may not have your relationships. But whoever holds your factory could help someone else rebuild it, and undercut you.

Picture a competitor, five years from now, asking an AI to build them a great system for your industry. If that AI has learned from yours, it may know how to build one as good as yours, or better. It could even offer your way of working as a starting template: start with these tasks, these agents, these workflows.

We are not saying any company does this today. We are saying that a business should not have to rely on a promise that it never will.

I believe the right analogy here is not the SaaS revolution. It is the computer.

When computers came into business, owners did not hand their computer to a third party that would also sell it to their competitors. They bought their own computer, and they installed what they needed on it to run their business. That is what worked.

We believe AI needs the same thing. Call it AI sovereignty: your own factory, holding all the work your business does, on hardware you own or control. Then plug the AI providers in. They can advise you. They can do individual tasks. And you decide, by your own judgment, how much to involve them, or how little.

This is a trade-off, and I want to say it plainly. As we see it today, the smartest models and agents are closed, not open source. Many businesses will choose them, and I respect that choice.

Our advice is only this: keep your operating system on a system of your own, and plug those models into it. Then, if you ever want to change models, you change what runs your factory, not the factory. That is what a factory you own is for: no rebuilding, no moving house.

Connections to your data are another switching cost. Rebuilding them each time you change AI provider is part of what keeps businesses where they are.

And if you want to move to open models, you can do that too. In fact, we believe open models run much better once a factory is well defined: a clear data model, clear processes, everything well refined. That suggests a path.

Build and refine your factories on frontier models. Test them. Once the processes are defined and you are mostly maintaining them, you can swap that work to open models.

We think that once a business runs factories, its AI usage could grow ten or even a hundred times, and that much more of it will be profitable use of AI.

Much of what a model company sells is usage: AI work, priced by the token. We believe businesses that run factories will use far more of it. Even if a business moves a large share of that work to open models, we think the total it buys from the model companies still grows. If you are in the business of selling tokens, you win in this future.

Their product teams, who are excited about building operating systems on top of their own products, may like this less. That is fair. I am sure they will keep building fun, exciting, easy products, and many people will love them. We are glad they exist.

My advice to business owners stays the same, and I give it almost as a fiduciary would: keep your operating system separate, on something you own, and invite the best models in.

So far this article has been about open source and walled gardens, and about businesses. I believe the stakes go further. This technology is powerful enough that the fork may be one for Western civilisation: whether AI becomes a centralising force or a decentralising one. Whether it strengthens the individual, or leaves the individual beholden to a few massive corporations, or to the government.

There is a lot of talk of AI doom, and we believe most of it is unfounded. But this risk is real: whether AI concentrates power or spreads it. If we get this one right, and AI strengthens the individual rather than big corporations or governments, we believe much of the rest falls into place.

None of this makes the AI companies villains. As we said earlier, they create real value. The question is one of structure, not of bad actors: who holds the factory. Power that gathers in one place tends to stay there, whoever holds it and however good their intentions.

Two paths, for real people.

Here is how the two paths look for a worker, a small business and an organisation. It comes down to three things: bargaining power, dependency, and who can see your work.

Who can see your work.

One difference is quiet but real. In United States law, records you hand to a third party have generally had weaker privacy protection than records you keep yourself. Courts call this the third-party doctrine. The Supreme Court applied it to a customer's bank records in United States v. Miller (1976), and to the numbers a person dials, as recorded by the phone company, in Smith v. Maryland (1979).

In Carpenter v. United States (2018), the Court narrowed it: the government generally needs a warrant to obtain historical cell phone location records from a carrier. The Court called that decision a narrow one, and the law here is still developing.

This is not legal advice. Our point is modest: what you store with a company can be easier for the government to reach than what you keep yourself. That is one more reason to keep your factory where you decide.

An old distrust of concentrated power.

Western tradition has long distrusted concentrated power of any kind, whether a company holds it or a government does. Antitrust law is one example. In 1911 the Supreme Court found that Standard Oil had broken the Sherman Act and upheld an order to dissolve it. In 1982 the United States settled its antitrust case against AT&T with a decree that broke up the Bell System, and on 1 January 1984 the local Bell telephone companies were separated from AT&T.

We take no side in any current political debate. Our point is about structure: when a few hands hold the tools everyone depends on, the individual loses leverage, whoever those hands belong to. AI can go either way. We believe it should spread power out.

We believe this is like the industrial revolution, and even bigger. A change that size sets its patterns early, while everything is still being built.

So we believe AI sovereignty should be baked in at the beginning. Businesses and organisations should own their own AI from the start: the factory that runs their work, on hardware they control, with any model they choose.

Agreeing to that now is cheap for everyone, the AI companies included, and as we argued above, everybody can win: the businesses, the AI model companies, and everybody else.

If we wait five years, it may be too late. The companies that gain a foothold by then will be strongly motivated to keep it, and what is cheap to agree today will be expensive to undo.

That is why the Infinity OS Foundation exists: to make owning your AI the easy path from the start, so that everybody in this industry wins, long term.

Our mission:

Every business should own its AI. The Infinity OS Foundation exists to help business owners build, own and run their own AI factories, on hardware they control, with any model they choose.

Because what we publish is free and open source, I encourage you to take it. It is early: today that means the early public code and the open starter, all MIT licensed. If our mission and our principles fit how you want to run your business, start with one of the three ways in.

---

Infinity OS is being built and is not yet released. Nothing in this article describes a released product.

Sources

• StatCounter Global Stats: desktop operating system market share, worldwide, read 16 October 2026.
• W3Techs: usage statistics of Linux for websites, read 16 October 2026.
• The Register: Tim Berners-Lee says regulation of the web may be needed, 12 March 2018, reporting his open letter for the web's 29th birthday.
• StatCounter Global Stats: mobile operating system market share, worldwide, read 16 October 2026.
• Apple: App Store Small Business Program and Google Play: service fees, read 16 October 2026.
• United States v. Miller, 425 U.S. 435 (1976), Supreme Court opinion, Cornell Legal Information Institute, read 16 October 2026.
• Smith v. Maryland, 442 U.S. 735 (1979), Supreme Court opinion, Cornell Legal Information Institute, read 16 October 2026.
• Carpenter v. United States, No. 16-402 (2018), Supreme Court opinion, 22 June 2018; also at Cornell Legal Information Institute, read 16 October 2026.
• Standard Oil Co. of New Jersey v. United States, 221 U.S. 1 (1911), Supreme Court opinion, Cornell Legal Information Institute, read 16 October 2026.
• United States Department of Justice: press release on the 1982 AT&T consent decree, 28 February 1996, and Pacific Bell, annual report on Form 10-K for 1993, on the divestiture effective 1 January 1984, read 16 October 2026.