The business model of a data center

Strip away the racks and the blinking lights and a data center is a rental business: it buys expensive machines, houses them, and rents out their working hours. Everything about whether it makes money hangs on four numbers.

The four numbers

1. Price per GPU-hour. Set by the market — chip generation, contract length, and how scarce capacity is. Long-term contracts trade some price for certainty; the spot market pays more but swings.

2. Utilization. A rack earns nothing while idle. The difference between 60% and 90% utilization is roughly the difference between a struggling facility and a thriving one. Serious operators secure anchor tenants — long-term contracts covering the base — and sell the remainder flexibly.

3. Power cost. AI racks are power-hungry, and cooling multiplies the draw. This is why location matters enormously: cheap, stable electricity (hydro in Scandinavia, for instance) is a structural advantage that never stops paying.

4. Depreciation. The silent giant. Hardware loses value fast as new chip generations arrive. A sound operating model plans the refresh cycle and prices it in — a model that ignores it is telling you a story, not a plan.

From revenue to investor returns

Revenue (price × rented hours) minus power, staff, maintenance, financing, and depreciation leaves the operating margin. That margin — not the revenue — is what can be distributed to whoever financed the facility. It's why two data centers with identical revenue can produce completely different returns.

Questions this should make you ask

When anyone offers you data-center exposure: What utilization does the plan assume? What's the power price and source? Who bears the hardware refresh? A specific, written answer to those three questions is the difference between an investment and a pitch.

Educational content — not investment advice. Capital at risk.