The $3 Trillion Question
Big Tech's AI buildout now carries about $1.5 trillion in purchase commitments plus nearly $1.5 trillion in lease obligations - close to $3 trillion in total. Investors keep asking whether this counts as hidden CapEx.
Scale and risk are two different things. Most of these obligations sit in purchase commitments, leases, and power contracts, but they do not start until the actual data center is energized and powered. The central question: will AI revenue and utilization grow fast enough to keep up with demand?
What Can Be Canceled
How much of these commitments is legally unavoidable, and how much can companies cancel, renegotiate, or walk away from? It depends on the contract and how it was set up. Certain items can be renegotiated. The most sophisticated players build in flexibility, phased commitments, termination rights, assignment rights, capacity step-downs, and even repricing mechanisms.
When a Commitment Becomes Debt-Like
A purchase commitment is not the same as funded debt, because it does not kick in until there is power to the data center. But it starts to look debt-like when it is long-dated, non-cancelable, and tied to capacity for the company. Investors should ask how much is fixed, how much flexibility and optionality is built in, and what happens when the technology changes.
The technology will change. There is a technology mismatch and a duration mismatch between the physical data center asset and the technology inside it. The technology needs a refresh roughly every 18 months.
The Duration Mismatch
Some commitments run 10, 15, even 20 years. If the tech needs refreshing every 18 months, what happens if AI economics change sharply within, say, 5 years?
Structure these contracts with a long-term view. Both sides of the equation matter: the power demand and energization on one side, the chips and the AI on the other. The real risk is the gap between the 15 to 20 year life of a data center and the much shorter technology cycle - and deciding who takes the refresh risk. Phased pricing, capacity rights, and how the technology refresh is handled all need to be settled when the contract begins, not 15 to 20 years later.
Who Absorbs the Risk
If AI demand falls short, who pays - the hyperscaler, the data center operator, the power provider, or the shareholder? It depends on how the deal is structured, because the contract spells out who assumes which risk. Sometimes hyperscalers take the utilization risk. Sometimes the developer, operator, or power provider takes it. Sometimes financing parties take part of the risk on pricing or minimum commitments.
If all these risks are mismatched - the outcome good structuring tries to avoid - shareholders feel it. The key is knowing who is contractually on the hook when demand does not show up.
A Bullish Read
The demand and the energization are there. The AI cycle is real. I am very bullish on it. The whole thing comes down to structuring each deal correctly at the start. Hyperscalers are turning into infrastructure and energy companies as these deals keep appearing, and individual contract terms matter far more than the headline aggregate number.
