The AI narrative still revolves around chips.
Every earnings cycle turns into another discussion about GPU demand, model performance, or who has the best frontier stack.
But the market increasingly looks saturated on that narrative.
The more interesting shift may be underneath it.
AI is starting to look less like a software cycle and more like an infrastructure cycle constrained by physical systems.
That matters because markets tend to misprice bottlenecks until they become unavoidable.
Right now, the constraint may no longer be compute availability alone.
It may be power.
NVIDIA continues posting extraordinary numbers, and hyperscalers are still spending aggressively on AI infrastructure. But even bullish AI investors are beginning to acknowledge that data center expansion is colliding with grid limitations, cooling requirements, financing pressure, and transmission delays.
That changes the shape of the trade.
When the market believes a cycle is purely about semiconductors, capital concentrates into chip manufacturers and model leaders.
When the market realizes deployment itself is constrained, the opportunity set broadens dramatically.
Utilities.
Transmission.
Cooling.
Power equipment.
Grid engineering.
Natural gas infrastructure.
Nuclear optionality.
Not because these businesses suddenly became exciting.
Because they became necessary.
That distinction matters.
What stands out to me is how quickly hyperscaler behavior has changed.
This no longer looks like incremental cloud capex.
The scale increasingly resembles sovereign infrastructure spending.
Large operators are locking in land, securing long-term power agreements, financing dedicated energy capacity, and competing for interconnection access years in advance.
That creates second-order effects the market may still be underpricing.
The first-order AI winners were obvious.
The second-order winners may emerge from whoever enables deployment at scale.
And the psychology around this still feels early.
Most investors still treat power constraints as a temporary friction point.
I’m not sure that’s right.
Grid expansion timelines are measured in years, not quarters. Transmission buildouts remain politically difficult. Many major data-center regions are already experiencing delays tied to power availability.
Meanwhile, AI demand keeps accelerating.
That mismatch is where structural scarcity develops.
And scarcity tends to drive pricing power.
One thing I’m watching closely is whether capital rotation inside AI starts broadening beyond the obvious mega-cap beneficiaries.
We may be approaching the phase where investors stop asking:
“Who builds the models?”
And start asking:
“Who enables the infrastructure required to sustain them?”
Those are very different positioning regimes.
Especially if rates stay structurally higher and financing costs begin to matter more.
Because eventually the conversation shifts from growth narratives to return on invested capital.
The market is still rewarding AI exposure.
But I think investors are beginning to differentiate between companies benefiting from narrative momentum and companies controlling actual bottlenecks.
That’s usually where the more durable asymmetry lives.
— Connor
Alpha Before It Prints
© 2026 Alpha Before It Prints
Unsubscribe