Opinion
You will never own the AI. You can still own what it cannot.
The real AI divide is not rich versus poor. It is the few who own the models and the many who will rent them.
Imagine everyone building frontier AI and you would not need a stadium. You would need a few rooms. A few hundred researchers are writing the future, and they are paid like no employees in history.
Look at what it costs to move one. In 2024 Google paid roughly $2.7 billion in a deal dressed up as a license. What it really bought, the Wall Street Journal reported, was one man: Noam Shazeer, who co-wrote the paper the field rests on, brought back to run Gemini. He reportedly walked away with hundreds of millions. Then, in June 2026, he left again, for OpenAI. One researcher, fought over by rival labs and priced like a scarce natural resource.
It is not a one-off. In June 2025 Meta put $14.3 billion into Scale AI for a 49 percent stake, mainly to install its 28-year-old founder atop a new lab and pull more than twenty researchers from OpenAI, DeepMind, and Anthropic, some on packages worth hundreds of millions. You no longer even need a product: Ilya Sutskever raised $2 billion for Safe Superintelligence at a $32 billion valuation with nothing shipped, and Mira Murati raised $2 billion at $12 billion the same way. The asset being priced is the people.
For the first time, the means of production of the era's defining tool are concentrating, not spreading. Every earlier wave came down to us: the printing press, the factory, the internet. AI runs the other way. Training a frontier model costs billions and demands chips and power almost no one can build. The engine is owned by a few, and the rest of us rent it. Renting is a fine business, for whoever owns the meter.
We tell ourselves a comfortable story: an AI superpower, talent unmatched, the world flying to Tel Aviv to hire. All true, and all on the wrong side of the line. Safe Superintelligence took an office in Tel Aviv and hired Dr. Yair Carmon from Tel Aviv University. Israeli minds, feeding a frontier that will be owned in California. A talent superpower can still be a renter, and we are at risk of being exactly that.
The answer is not despair, and not a race you can win. You will not out-build the people racing to AGI, so stop trying. The model is becoming electricity, and almost nobody got rich owning the current. The fortunes went to those who built what ran on it. If the only thing your company adds is a prompt wrapped around someone else's model, you own a thin layer anyone can copy, sitting on an engine someone else controls. They can raise the price, change the rules, or build your feature themselves, and a cut of every dollar you earn flows back to them with every token you run. That is renting. Which leaves the question worth the whole essay: what can you actually own that the model cannot take back?
When building is cheap, ownership is the whole game
The ability to build, once scarce and well paid, is collapsing in price. One person can now ship in a weekend what once took a funded team a year. What is no longer scarce is no longer a moat. Value moves to whatever the model cannot copy: a niche the giants overlook, a channel you built, the customer, the data.
Maor Shlomo is the proof. Almost alone he built Base44, which lets anyone create software by typing a sentence. He did not train a model; he wired existing ones into a product, grew it to more than 250,000 users with almost no marketing, and sold it to Wix in about six months for roughly $80 million, as sole shareholder. He owned the niche, distribution, customer, and data. The value landed with him.
It is not a fluke. Doti, founded in 2024 by two former Wix engineers, built an AI layer over a company's scattered internal knowledge, raised about $7 million, and sold to Salesforce a year later for an estimated $100 million. Solo or team, consumer or enterprise, the move is the same: own what sits on top of the model.
The race to build AGI will be won by a few hundred people in a few rooms. That race is closed. The race to build durable, human things on top of it is wide open, and it belongs to everyone reading this. The question of the next decade is not whether you use AI. Everyone will. It is whether you own anything it cannot.
Oran Yehiel is a CPA (Israel) and US Enrolled Agent. For fifteen years he has run US operations and finance for Israeli companies building into America. He writes from Austin, Texas.
