Capital Before Revenue

Four firms guided to as much as 630 billion dollars of 2026 capital spending in a single earnings week. This report follows that money into the national accounts, into the electricity system and into household bills, and identifies who is levered to a programme they do not control.

Scale of commitment

In one earnings week, four companies guided to as much as 630 billion dollars of capital expenditure for calendar 2026, roughly 62 percent above a 2025 total of 388 billion that was itself a record. Amazon alone plans around 200 billion. Across the five largest US cloud and AI infrastructure providers, including Oracle at about 50 billion, committed spending runs to between 660 and 690 billion, close to double the 2025 level.

Capital formation at this scale enters the national accounts directly, on identifiable lines, and moves the aggregate. It is no longer sensible to treat it as a technology sector development.

The timing asymmetry

Capital spending enters output the moment it is incurred. The revenue it is meant to generate arrives over subsequent years. That timing gap has a specific and frequently misread consequence.

Growth accounting operates on changes, not levels. A capital programme that merely stops rising contributes nothing to growth the following year while remaining enormous in absolute terms.

So the cycle does not need to reverse to become a drag on measured growth. It only needs to plateau. Anyone forecasting 2027 output is, whether they state it or not, forecasting the second derivative of hyperscaler capital spending. That is an uncomfortable dependency, and it is rarely written down as an explicit assumption.

The corresponding upside is real. Data centre construction, electrical equipment manufacturing, semiconductor packaging and AI software development are concentrated sources of marginal demand, and their geographic concentration is already reshaping which regional economies grow fastest.

The physical constraint

Executives are describing themselves as capacity constrained while spending at this scale, which tells you capital is not what is scarce. Power availability, interconnection queues, turbine and transformer lead times and construction labour are.

That has two consequences. Spending converts into productive capacity more slowly than the headline implies, so the revenue lag is longer than a naive model assumes. And the competition for electricity shows up as a price for everyone else on the same network.

In Virginia, data centre growth is adding thousands of megawatts of near-constant demand, and the incumbent utility has proposed its first base rate increase since 1992, adding roughly 8.51 dollars a month for a typical household in 2026 and a further 2.00 in 2027, alongside a resource plan of nearly 27 gigawatts of new generation by 2039. In the PJM market, data centre load is associated with a 9.3 billion dollar increase in the 2025-26 capacity auction, implying on the order of 18 dollars a month on bills in western Maryland and 16 in Ohio.

Wider economic effects

Higher electricity prices are estimated to exert roughly a 0.2 percentage point drag on real consumer spending growth across 2026 and 2027 by lowering real disposable income, with a net drag on GDP growth of around 0.1 points once higher utility capital expenditure is netted against it. The drags are larger for lower-income households and larger in regions with high data centre concentration, so an average of 0.2 across the distribution implies something considerably worse at the bottom of it.

Two groups deserve more attention than they get. Energy-intensive manufacturers in data centre dense regions face the full power price increase with none of the offsetting revenue: same grid, same tariff, no upside. And households, who bear cost socialised through regulated recovery for infrastructure serving a load they do not consume. The first group is rarely modelled as exposed. The second is where this becomes political.

Assessment of risk

Nobody knows whether the return justifies the spending, including the people doing it. The clearest rationale offered is not that the return is proven but that being short of compute is the one error none of them can afford. That is a defensible corporate strategy and a poor basis for a macroeconomic forecast, because it means the spending path is set by competitive dynamics rather than by realised demand.

Four balance sheets are making a correlated bet on the same three inputs at the same time. Correlated bets create correlated downside: if enterprise adoption disappoints, the entire stack re-rates simultaneously, because there is no diversification across the buyers. Investors have already shown some sensitivity to this, with shares in several of the spenders selling off following the guidance.

Forecasts that assume capital expenditure tracks AI revenue are assuming a discipline the participants have explicitly said they are not applying.

Source register