There are two ways to participate in the artificial-intelligence buildout. You can operate the compute, buying accelerators and renting them out as cloud capacity, or you can own the physical foundation those operators depend on and lease it to them. The two look similar from a distance. Up close, they are different businesses with different risk profiles and very different durability.
Operating compute is a hard place to stand. The hardware depreciates quickly, the next generation is always months away, and price competition is relentless. You are exposed to technology risk at the most volatile layer of the stack, in a market where today’s premium product is tomorrow’s discount.
Owning the durable layer
The landlord owns something slower-moving and more defensible: the power, the cooling, and the buildings. That capacity is leased under long-term contracts to the firms racing to train and run the next generation of models. The economics look less like software and more like industrial real estate, with recurring, contracted cash flows tied to something every AI company needs and almost none can build for itself.
The operators fight over a depreciating asset. The owner of the building collects rent through the entire cycle.
History rewards this position with unusual consistency. In the gold rush, the durable fortunes were made selling tools and provisions, not panning for gold. In the early internet, the firms that laid fiber and built the data centers quietly underwrote everything that came after, long after individual applications rose and fell. The pattern repeats because it reflects something structural: infrastructure outlives the products that run on it.
This is not an argument that operating AI compute cannot be lucrative. It is an argument about where risk is concentrated and where value is durable. The landlord converts a volatile technology race into stable, contracted income, and sits beneath the entire industry rather than competing on its most exposed front line.
As the AI buildout matures, the question of who captures the lasting value will not be settled by who had the best model in any given year. It will be settled, in large part, by who owned the ground it ran on.
The views expressed are the author’s own and are offered as general commentary. They do not constitute legal, tax, investment or accounting advice.