Semis slipped towards a bear market, SpaceX was down 45% from its intraday high on June 16 and Nvidia’s market cap fell below Apple’s for the first time in over a year.
Suffice to say the ground’s shifting. A big part of this, I think, is the creeping suspicion that one or more of the hyper-scalers will feel compelled in the quarters ahead to guide capex lower, or at least to tacitly suggest spending plans are under review.
That’s not the consensus, nor is to say that the AI capex cycle’s anywhere near over. It’s just to state the obvious: The big spenders are under pressure to justify these outlays, particularly in the context of fierce competition from Chinese AI startups. I don’t think they can do it — justify the outlays — in the near-term. Note the emphasis.
Whatever AI’s long run promise, investors are getting antsy. And for good reason. The figure below’s a reminder: Capex is set to exceed free cash flow for the big five spenders.
You can’t spend every spare penny on data centers and compute, which means you have to offset the cash burn somehow.
Oracle aside (and even it’s a blue chip), these are among the most creditworthy entities in the history of capitalism. Pre-ChatGPT, their debt — to the extent they had any — was arguably a safer bet than US Treasurys. So the first thing you do if you’re aiming to preserve cash flow is borrow to fund some of your capex.
But the hyper-scalers are more than $210 billion in on that front since September.
The figure above’s been updated. It’s critical to keep tabs on this borrowing binge.
Consider this: Stripping out Amazon’s euro and loonie deals, and also excluding Google’s sterling, franc, loonie and yen offerings, hyper-scaler supply over the last 11 months still equates to something like 9% of total expected USD high-grade supply for 2026. If the post-Labor Day IG slate were to undershoot estimates, that figure could easily be north of 10%.
Eventually, reception for these deals is going to cool. When it does, the hyper-scalers will have to offer juicier concessions.
As the figure above shows, there was some evidence of indigestion around Amazon’s July 7 offering, which was “only” 1.6x covered, well below the ratio implied when the order book peaked, and meaningfully lower than deals done earlier this year, including Amazon’s March offering.
This is one reason some market participants expect more in the way of Alphabet-style follow-ons (i.e., equity raises), which in turn means more equity supply from a group of companies which normally account for a huge share of demand (i.e., buybacks).
In the event the hyper-scalers don’t want to issue more debt or equity and don’t want to make sundry FCF ratios look even dicier than they suddenly do, they’ll need to at least consider not guiding higher for capex.
Some view Meta’s nascent push to sell excess and/or idle compute as an effort, last ditch or not, to avoid a capex guide down. I don’t necessarily think that’s the best way to look at it, but… well, on Friday, The New York Times reported that Mark Zuckerberg’s in talks with Anthropic on a lease deal that could be worth as much as $10 billion over two years.
The main near-term issue with all of this for markets is that capex speculation is weighing on semis, the go-to “next phase” AI trade in 2026. The big spenders acted as a source of funds for longs in red-hot chipmakers, but now semis are in a bear market before the hyper-scalers have recovered their joie de vivre.
By and by, the hyper-scalers could end up rallying in the event any potential capex cuts are greeted as a sign of discipline by the market, but for now you have a scenario where both the spenders and the beneficiaries of that spending are struggling.
The SOX just had its worst week in over a year, and three of the five hyper-scalers were lower. The Mag7 as a group was weaker this week too, and SpaceX’s market cap is down nearly a trillion from the highs.
I’m not sure where this goes from here, but a lot’s riding on capex commentary from Alphabet, Amazon, Meta and Microsoft, all of “whom” report later this month.
As mentioned here on Thursday, I fear it’s a damned if they do, damned if they don’t scenario. If management expresses anything like caution on AI outlays, markets might take it as a sign that the ROI on money already spent is poor. If, on the other hand, management tips more spending, it’s right back to pondering the questions implicit and explicit in everything said above.
Apple at an all time high! They don’t even have an AI and have not invested in compute! Investors have taken note.
They’ve always played it this way. “Slow to adopt”
As a technologist, I generally take this approach myself. If you want to get actual work done, you generally don’t want to waste time trying out the new tools, you want to get things done using the best things that are at least somewhat robust.
A lot of “really smart” engineers spend a lot of time tricking out their tooling, which I generally view as a waste of time. There’s a lot of film to watch, wine to drink, herb to smoke. And yes, to build castles in the sky of course.
If everyone else is doing something and Apple’s not, it’s never a bad idea to step back and ask, “Who’s the stupid one here, Apple or everybody else?” It’s not a rhetorical question. Apple makes mistakes too. But to just assume that Apple has it totally wrong is probably not the best way to think about it.
Exactly. My opinion of Apple’s management jumped way up when I saw that they were not about to pour billions ( Billions not millions) down the toilet trying to develop and stay in the LLM lead.
They remained focused on customer needs. What a novel concept!
Note that China and now India are following their lead. But what do they know?
I think Apple is in the best position of anyone. They don’t want to build AI, they want to build the machines we use to access it. They are waiting — with a mountain of cash — for one of the other AI companies to falter, so they can buy them out in full at a bargain price.