Everyone Wants to Own the Machine
Meta wants to be a cloud, Washington wants a slice of the labs, and Coinbase wants 1,200 agents on payroll. The AI business is quietly reorganizing around who owns what.
The story this week is not about a shiny new model. It is about ownership. Who owns the compute, who owns the labs, and who owns the work once the software basically writes itself. Three separate threads all point the same direction: the value in AI is migrating away from clever demos and toward the boring, capital-heavy assets underneath. If you want a single lens for the noise, that is it.
Meta wants to be a landlord
The headline that moved markets was Meta building a cloud business to rent out its AI compute. Meta is developing plans for a cloud infrastructure business that will sell access to AI computing power and models, setting up a new vector of competition with Amazon Web Services, Microsoft Azure and Google Cloud. The reflex read was that Meta is drowning in overcapacity and about to dump it on the market. That is the wrong take.
The more convincing view is that this accelerates Meta's buildout rather than signals a glut. Meta has been contracting enormous volumes of capacity, and the cloud angle simply gives it more ways to monetize whatever superintelligence ambitions do not immediately pay off. Think of it as optionality. The compute still feeds frontier model training and a massive ramp in ads recommendation complexity, but if those bets underdeliver, renting capacity at a premium is a very high-margin fallback. That is why the market's initial panic looks misplaced.
The irony is thick. Meta's decision to sell off excess compute comes weeks after SpaceX, via xAI, announced similar plans. In early May, SpaceX signed a deal with Anthropic to buy out all of the compute capacity at SpaceX's Colossus 1 data center. Everyone with a data center now wants to be a neocloud. The investors who dumped CoreWeave and Nebius on the news may have the causality backwards. The winners of the AI race may not be the ones providing the best models and services, but rather the ones who own the data centers. When a company that reportedly agreed to pay neoclouds tens of billions for GPUs starts eyeing the same business, the message is that infrastructure, not intelligence, is the scarce asset.
Washington wants a cut
The second ownership story is stranger and more consequential. OpenAI has proposed handing the U.S. government a 5% stake in the company, as the startup seeks to defuse mounting political pressure in Washington. A 5% holding would be worth roughly $42.6 billion, after the lab closed a record round in March at a post-money valuation of $852 billion. Altman's framing is that the public should share the upside. The more interesting detail is the scope. The broader arrangement would have Washington hold 5% of each of the leading U.S. AI developers via a government vehicle, envisioning Anthropic, Google and Meta ceding similar stakes through a sovereign wealth fund.
Why now? Because Washington has started treating frontier labs as strategic assets, not just companies to police. The timing is not subtle. This lands right after the government spent most of June with a literal off switch over Anthropic's top models. Anthropic spent most of June with its Claude Fable 5 and Mythos 5 models disabled worldwide under the first U.S. export controls ever applied to an AI model rather than to hardware; access was restored yesterday. A government that can turn a model off, and now wants equity in the company that makes it, is both regulator and shareholder. That blurs a line that used to matter.
The precedent is already set elsewhere. The administration took a 9.9% stake in Intel last August by converting CHIPS Act grants into equity, and AMD and Nvidia agreed to hand over 15% of their China chip revenue in exchange for export licenses. For anyone tracking where AI margins ultimately land, add a new stakeholder to the cap table: the state. That is a structural change to the industry's economics, not a headline.
The org chart is next
The third thread is what all this compute is actually doing to how companies run. Brian Armstrong says Coinbase now operates with roughly 1,200 full-time AI agents, counted the way you would count staff, with small pods of humans supervising fleets of agents that open pull requests and ship work. Whether the exact number holds up, the direction is real, and you see the same pattern in smaller shops where non-engineers now push code daily.
The smarter operators are not editing the agent's output. They are updating the underlying instructions so the next attempt lands clean, treating institutional knowledge as the real asset. That is the same ownership logic playing out at human scale: the durable value is not the code an agent produces in an afternoon, it is the accumulated context and judgment that only you control.
So what
Strip away the model launches and the through line is unmistakable. In an era where building software is nearly free and models are converging, the moat is whatever cannot be copied over a weekend: the data centers, the equity, the proprietary context. Meta is buying the first, Washington is claiming the second, and the sharpest companies are hoarding the third. If you are deciding where AI value accrues over the next few years, stop watching the demos and start watching the deeds.