The rotation argument has a specific claim embedded in it: money is moving out of Nvidia and into other names, not out of the AI trade altogether. That is a testable distinction, and it is the one the desk flagged on 6 July 2026 when it noted a 12-month hardware delay to Nvidia's Kyber NVL144 rack was a supply-side risk distinct from the demand question Nomura was defending the same day. The prior piece said the Nasdaq Composite had not yet shown which risk the market was pricing. It has now moved, and not gently.
Nvidia's own valuation reset is described as a fall to a pre-AI boom low, a phrase that says more than a market cap figure alone. A cap loss can be rotation if capital resurfaces in Broadcom or another chip name; a valuation reset to pre-boom levels is a different claim, one about the multiple the market is willing to pay for the entire AI infrastructure story, not just for Nvidia's share of it.
Oracle supplies the corroborating data point the rotation thesis needs to explain away. Oracle stock is down 25% in the first half of 2026, and the wire attaches a specific mechanism: some of Oracle's biggest data center customers might be unable to meet their obligations. That is a demand-side warning about the AI infrastructure buildout's counterparties, not a supply-side hardware delay like the one at Nvidia. Two different failure modes, both landing on AI capex names in the same week, is harder to wave off as one company's rotation.
A trillion-dollar cap loss dressed as rotation and a valuation reset to pre-boom levels cannot both be true; one of them is the market's real verdict on the AI-capex story.
Apple's announced $30 billion purchase of US-made chips from Broadcom cuts the other way, at least for Broadcom specifically: capital is still moving toward chip supply agreements, just not toward Nvidia's. That is consistent with the rotation reading in the narrowest sense, one buyer redirecting spend within the semiconductor complex. It does not resolve whether the broader AI capex thesis is intact, because a single procurement deal is evidence about Apple and Broadcom, not about aggregate AI infrastructure demand.
The Russell 2000's 1.75% intraday drop on 8 July 2026, its steepest of the three major indices cited here despite carrying the least AI concentration, is the detail that most complicates a clean rotation story. If this were capital moving within the AI trade, small caps with little AI exposure should be relatively insulated. They are not. The softest point in this reading is that one session of broad-based weakness ahead of the FOMC minutes, due at 18:00 UTC on 8 July 2026, could simply reflect rate-sensitive positioning unwinding across the board rather than anything specific to AI capex. What would confirm the reassertion thesis has failed: semiconductor and AI-linked names underperforming the broader index over the coming sessions, as the desk's 6 July note already flagged; what would save it is Nvidia and its chip peers stabilising or outperforming once the FOMC minutes are digested, leaving 8 July as a rate-driven air pocket rather than an AI-specific repricing.
