Macro & Market
·Follio Research

AI Capex Cycle — Where Are We in May 2026

Hyperscaler infrastructure spend is decelerating sequentially for the first time since 2022. Why it matters for NVDA, the power complex, and the speculative quantum/space tail.

macroAIsemiconductorspowercycle

The defining trade of the last 24 months — long AI infrastructure, short almost everything else — is showing its first cracks. Not collapse. Deceleration. The distinction matters.

The setup

Microsoft, Meta, Alphabet, and Amazon together guided combined FY26 capex of roughly $240 billion, up from $190 billion in FY25 and $130 billion in FY24. That is still a parabolic curve in absolute terms. But the sequential growth rate — the second derivative most equity markets actually price — is decelerating.

NVIDIA's most recent quarter (Q1 FY27, ending January 2026) printed +19.5% QoQ revenue growth. The prior quarter was +22%. The one before, +27%. The trend line is unambiguous: the rate of acceleration is fading, even as absolute dollars keep ripping.

This is not bearish. It is the shape of a maturing capex cycle — the same shape we saw in dotcom 1999, in fracking 2013, in semis 2018. The market historically pays the highest multiple right at the moment of peak second derivative, then derates even as numbers continue to grow.

What's confirming the shift

Three independent signals worth tracking:

  1. Insider behavior at NVDA. $165m+ of net executive sales in a single two-week window in March 2026 (Kress, Puri, Stevens). One window means nothing. A pattern means something. We are watching for the next print.

  2. Dark pool ratios across semis. Off-exchange volume as a percentage of total has crept up from ~38% in late 2024 to ~44% in Q1 2026 for the top five AI infrastructure names. This is institutional hedging, not retail capitulation.

  3. The power complex is leading. VST, CEG, OKLO, SMR — the second-derivative beneficiaries — are outperforming the primary beneficiaries (NVDA, AVGO) on a 90-day basis. That historically happens at the back half of a thematic cycle, not the front.

What this implies for positioning

What would change the view

Three things would force us back to "full risk-on":

Until then: accumulate quality on weakness, trim quality on strength, avoid junk on either.


This post is opinion grounded in observable data. It is not a recommendation to buy or sell any security. Position sizes, stop-losses, and risk tolerance are yours alone.