Who owns AI?
Nobody owns AI. It is a field of research, not property — the core methods, including the 2017 transformer design behind every major chatbot, are published publicly and free to use. What companies own are specific models, brands, and trained weights: OpenAI owns ChatGPT, Google owns Gemini, Anthropic owns Claude.
Why — the first-principles explanation
Three different things get called "AI," and only one of them can be owned.
The field cannot be owned. AI is a branch of computer science, launched at Dartmouth in 1956. You cannot own a field any more than you can own chemistry. The founding proposal was published; the founding conjecture — that intelligence can be described precisely enough for a machine to simulate it — is public property. The methods are mostly not owned either, and this surprises people. The transformer architecture that powers ChatGPT, Gemini, Claude, and Grok was published by Google researchers in a public paper in 2017. Google did not lock it up. Backpropagation, attention, gradient descent — all published. Software patents on core AI methods exist but are weak and rarely enforced against research. Anyone with a textbook can build a small language model this weekend.
The trained models are owned, and that's where the money is. A specific model — the actual numbers inside it, called weights — is a company's property, protected as a trade secret, not a patent. Training one costs hundreds of millions of dollars in electricity and chips. That expense, not legal exclusivity, is the moat. As of July 2026 the major owners are OpenAI (ChatGPT), Google DeepMind (Gemini), Anthropic (Claude), Meta (Llama), and xAI (Grok), plus Mistral in France and DeepSeek and Alibaba in China. Some — notably Meta and Mistral — publish their weights for anyone to download and run, which means a piece of frontier-class AI is genuinely unowned.
So the real structure is: the recipe is public, the ingredients are expensive, and the finished cake is private. Concentration in AI comes from capital, not copyright. That's also why the ownership questions people actually fight about are narrower ones — who owns the training data, who owns AI output. On that last one, the U.S. Copyright Office has held that purely machine-generated work without human authorship isn't copyrightable, meaning some AI output is owned by no one at all.
An example that makes it click
Nobody owns pizza. The recipe is free — flour, water, heat. Anyone can look it up. But Domino's owns Domino's: its ovens, its stores, its brand, its specific dough.
AI is the same. The recipe went public in 2017 when Google printed the transformer paper. What OpenAI owns isn't the concept of a chatbot — it's their particular oven, the couple hundred million dollars of electricity they baked with, and the name on the box. And Meta does something odd for a pizza chain: it gives away its finished dough for free. Anyone can download it and open a shop.
Key facts
- AI began as an academic field at the Dartmouth Summer Research Project in 1956; the founding proposal was public and no one holds rights to the discipline.
- The transformer architecture underlying every major 2026 chatbot was published openly by Google researchers in 2017 and is free for anyone to implement.
- Trained model weights are protected as trade secrets, not patents — the barrier is training cost, which runs into the hundreds of millions of dollars for frontier models.
- As of July 2026: OpenAI owns ChatGPT, Google DeepMind owns Gemini, Anthropic owns Claude, Meta owns Llama, xAI owns Grok.
- Meta's Llama and Mistral's models publish downloadable weights, so anyone with sufficient hardware can run them without permission.
- Anthropic is a Delaware public benefit corporation partly governed by an independent Long-Term Benefit Trust; OpenAI Group PBC is controlled by the nonprofit OpenAI Foundation.
▶ The 60-second explainer (script)
Who owns AI? Nobody. And that's not a technicality — it's the most important fact about the industry. Three things get called AI, and only one of them can be owned. First, the field. AI is a branch of computer science, founded at a workshop at Dartmouth in 1956. You can't own a field. Nobody owns chemistry either. Second, the methods. And this is the part that surprises people: the transformer — the architecture behind ChatGPT, Gemini, Claude, all of them — was published by Google researchers in a public paper in 2017. They gave it away. Backpropagation, attention, gradient descent, all public. You could build a small language model this weekend with a textbook and a laptop. Third, the models. This is the part companies own. Not the idea — the actual numbers inside a specific trained model, called weights. Those are trade secrets. And what protects them isn't law, it's cost: training a frontier model runs into the hundreds of millions of dollars in chips and electricity. So the structure is: the recipe is public, the ingredients are brutally expensive, and the finished cake is private. AI is concentrated because of capital, not copyright. Oh, and one more twist — Meta and Mistral publish their model weights for free. So some frontier-grade AI is owned by literally nobody. You can download it right now.
What authoritative sources say
People also ask
Can anyone own AI technology?
You can own a specific trained model, a brand, and your code. You cannot own the field or, in practice, the core methods — most were published openly, including the 2017 transformer paper behind every major chatbot.
Who owns the biggest AI companies?
OpenAI's for-profit arm is controlled by the nonprofit OpenAI Foundation, with Microsoft holding about 27% equity. Anthropic is a public benefit corporation with an independent trust electing board members. Google DeepMind is owned by Alphabet, a public company.
Who owns what AI creates?
Contested and jurisdiction-dependent. The U.S. Copyright Office has held that works without human authorship cannot be copyrighted, so purely machine-generated output may be owned by nobody. Vendor terms typically assign you rights to your outputs, but that doesn't override copyright law.
Is any AI free to own and run?
Yes. Meta's Llama and Mistral's models publish downloadable weights you can run on your own hardware. Licenses vary and some restrict commercial scale, so read the terms.
Does the government own any AI?
Governments fund and procure AI heavily — U.S. defense agencies have bankrolled the field since the 1960s — but no government owns the field or the leading commercial models.