There’s a number that keeps showing up in earnings calls that no one seems to want to talk about directly.
The hyperscalers are committing hundreds of billions to AI infrastructure. The data center buildout is real, it’s massive, and it’s accelerating. Everyone in the room agrees this is happening.
What nobody in the room seems to want to model is what happens on the other side of that investment.
The consensus
Everyone agrees AI infrastructure spending is unprecedented. The hyperscalers are committing hundreds of billions. Nvidia’s order book is full. Power contracts are being signed for facilities that won’t come online for three years.
The consensus narrative is: this spending is justified by the coming wave of enterprise AI adoption. The demand will come. It always does with transformative technology.
That’s not wrong. The demand probably will come.
The question is when.
The turn
What the consensus hasn’t modeled is the lag.
Enterprise adoption of new compute paradigms has historically trailed infrastructure buildout by 18–36 months.
We saw this with cloud. We saw this with mobile. In both cases, the infrastructure was built years before the enterprise workflows caught up. The hyperscalers who built early won. But the gap was real, and it was painful for anyone who priced in the demand arriving on schedule.
The current AI infrastructure buildout is happening faster than either of those transitions. The capex commitments are larger. The timeline assumptions are more aggressive.
And the enterprise use cases that would justify this spending — the ones that actually move the needle at scale — are still largely in pilot.
What I’m watching
The gap between infrastructure investment and enterprise AI spend is the number I’m tracking. When those two lines converge, the thesis is proven wrong. Until then, it holds.
The model I keep running suggests the gap doesn’t close before late 2026 at the earliest. That’s a long time to be carrying the weight of sunk costs.
What would change my mind
- Enterprise AI spend accelerates past $200B annually by Q4 2025
- A major hyperscaler publicly cuts capex guidance by more than 15%
- A credible third-party model of enterprise AI ROI shows payback periods under 24 months