One thing we learned over the past three or so months is that broad US equities can’t make new all-time highs without the hyper-scalers participating in the rally.

“Learned” probably isn’t the right word. Because “broad” is a misnomer in the context of cap-weighted equity benchmarks. Said differently: Of course it’s difficult for a cap-weighted stock index to score new records when the most heavily-weighted names are under pressure.

As discussed here on several occasions of late, the irony of the July momentum unwind is that it actually “allowed” the S&P to scale new highs as the “long AI beneficiaries versus big spenders” trade reversed.

The light grey shaded area in the figure above shows you the dynamic. Once the hyper-scalers found their footing, the index managed new records.

For what it’s worth, the reversal that blew up the momentum trade in July (i.e., the multi-sigma drawdown for the semi / memory / AI “bottleneck” plays versus hyper-scalers long/short trade) may itself be in the process of reversing — and just in time for Nvidia earnings later this month. But for our purposes here, the point is just that “parabolic semis” were insufficient to get us to new records if a corollary was “hyper-scalers as a source of funds.”

One interesting aspect of the sideways grind that all of the hyper-scalers succumbed to at one time or another in 2026 (Amazon’s now up sharply for the year and Alphabet respectably, but every name in the group’s trailing the Nasdaq 100) is that earnings, both realized and expected, are sturdy, which means the group’s uneven performance was accompanied by meaningful multiple compression.

The figure above, from SocGen, says a lot. A picture’s worth a thousand words, as the old saying goes.

Mechanically, it’s the juxtaposition between meandering share prices and buoyant earnings that’s compressing the multiple, but behind the stocks’ bumpy ride is the perception that capex is out of control and that the borrowing binge to offset the associated cash burn is likewise off the rails.

One way to illustrate that is to contrast hyper-scaler CDS — which has nearly quadrupled since Oracle changed the game in September of 2025 — and the group’s multiple — which compressed 12 turns over the same period.

As it turns out, borrowing a quarter trillion dollars over 12 months is the sort of behavior that warrants a discount. Or at least makes people pause before bidding up your shares commensurate with rising profit expectations.