When AI capex becomes macro: what the buildout means for buyers
AI infrastructure spending is set to top half a trillion dollars in 2026, roughly 2 percent of US GDP, and carried most of first-quarter growth. For everyone buying AI rather than building it, the boom changes the calculus.
TL;DR
- Goldman Sachs expects AI companies to invest more than 500 billion dollars in 2026, and analysts put US AI and data-centre investment at roughly 2 percent of GDP this year, approaching the defence budget.
- AI-related capex accounted for an estimated three-quarters of US GDP growth in the first quarter of 2026.
- For the vast majority of firms, which buy AI rather than build it, the buildout means falling unit costs, rising strategic dependence, and a real need for scenario planning.
The AI story of mid-2026 is no longer only a technology story; it is a macroeconomic one. Goldman Sachs projects more than half a trillion dollars of AI investment this year. Analysts estimate the United States will devote around 2 percent of GDP to AI and data-centre infrastructure in 2026, within sight of what it spends on defence, and AI-related capex is credited with roughly 75 percent of first-quarter GDP growth. Commentators reach for the railways, the interstate system, and the late-1990s telecom buildout for comparisons. Whatever the right analogy, the number is now big enough to shape the economy your business operates in.
Reading the buildout from the buyer's side
Almost every company we work with sits on the demand side of this boom. From that seat, three implications matter more than the bubble debate.
Capability keeps getting cheaper. Massive competing investment in chips, data centres, and models has driven the cost of a unit of machine intelligence steadily down. Planning assumptions set in 2024, about what is affordable to automate, are stale. Business cases that failed at last year's prices deserve a rerun.
Dependence is deepening. The flip side of buying ever more capability from a handful of infrastructure owners is concentration risk: pricing power, capacity allocation in shortages, and platform lock-in. The buildout strengthens suppliers' hands precisely because everything downstream will run on their capital stock.
Correction risk is a planning input, not a prophecy. When investment this large is premised on future demand, the late-1990s parallel cuts both ways: the telecom crash bankrupted builders, but the fibre they left behind made the internet economy cheap for everyone after. A funding correction would slow the frontier without un-inventing the capability already deployed.
Questions worth an hour of your leadership team
| Question | Why now |
|---|---|
| Which shelved AI business cases flip positive at current prices? | Unit costs have fallen since they were written |
| Where would a supplier price shock or capacity squeeze hurt first? | Concentration is rising with every contract |
| What is portable in our stack if a key provider stumbles? | Corrections punish the un-hedged, not the users |
| Are we capturing the deflation, or paying last year's rates? | Procurement rarely reprices as fast as the market |
The operator's stance
- Treat AI pricing like an input cost: review it on a cycle, renegotiate, and rerun rejected business cases annually.
- Diversify where it is cheap to do so, and keep exit costs visible where it is not.
- Plan for both branches: capability keeps compounding if the boom holds, and keeps working if the financing does not.
The half-trillion-dollar buildout is being financed by someone else's balance sheet. The discipline is making sure your organisation is positioned to harvest it either way.
Related reading
Source: Why AI companies may invest more than $500 billion in 2026, Goldman Sachs.