The AI capex digestion narrative is finally visible in the hyperscaler print — sequential capex growth has slowed at three of the four. The bear case was always demand-driven: that enterprise AI adoption was overstated, that ROI was thin, that the capex was front-loaded relative to a demand curve that would disappoint. None of that has actually shown up. The slowdown is instead supply-side — power availability, transformer lead times, and silicon allocation are forcing a slower build pace than ambitions would dictate.
Key takeaways
- Hyperscaler capex growth has decelerated, but not because of demand.
- Power availability is now the binding constraint on data-centre builds.
- Transformer lead times have stretched to multiple years.
- Enterprise AI adoption metrics continue to surprise to the upside.
Why the bear case was wrong about the cause
The bears predicted a demand miss. The demand has held up — enterprise inference workloads, sovereign AI deployments, and consumer-facing AI applications are all growing in line with or ahead of expectations. The miss is on the supply side. Power-availability constraints have forced the hyperscalers to pace their builds. Transformer shortages have stretched delivery timelines. Silicon allocation has gated training-cluster construction.
- Power: binding constraint at premier sites
- Transformers: multi-year lead times
- Silicon: still allocation-constrained
- Demand: tracking expectations or better
What it implies for the supply chain
The pacing of capex is now set by infrastructure, not by demand. That favours suppliers with backlog visibility and pricing power — power equipment, transformers, grid infrastructure. The narrative-driven AI semis remain story-stock-sensitive.
Where digestion is real
Speculative-tier data-centre developers without secure power.
Where it isn't
Hyperscalers with locked-in power.
Hyperscaler quarterly capex growth
| Period | YoY capex growth (avg top 4) |
|---|---|
| 2024 | +45% |
| Q4 2025 | +38% |
| Q1 2026 | +28% |
AI capex digestion is a supply-side story, not a demand story.
Frequently asked questions
Is the demand really intact?
Yes, on every metric that matters.
What ends the constraint?
New power capacity and transformer supply.
How long does the constraint last?
Multi-year.
The bottom line
The capex digestion is real but not for the reason the bears claimed. The bottleneck is power and transformers.
