Anyone tired of the capex boom discussion and the associated debate about the read-across of ballooning AI outlays for the largest source of US equity demand?
My hand’s raised. I’m exhausted with that discussion. Because I feel like I’ve covered it from every conceivable angle, and when you run out of angles… well, that’s when it gets tough from an editorial perspective.
Alas, this’ll remain one of the most important dynamics for US stocks in perpetuity, which is to say as long as the AI buildout’s above the fold, the buyback question will remain top of mind. For investors, anyway.
As a quick, but important, aside, it’s crucial to distinguish between buybacks and buyback growth, just as it’s critical to delineate between price level and inflation.
To point out, with some alarm, that buyback growth’s slowing isn’t necessarily to say overall share repurchases are declining, although that’s a distinct possibility. And even if buyback growth turns negative, the corporate bid will remain enormous. This is still shareholder capitalism, after all.
The updated table above, from Goldman, gives you a sense of things. Even as the market leadership plows hundreds of billions into data centers and compute to keep pace in the AI arms race, aggregate buybacks are expected to keep growing and anyway remain north of $1 trillion per year.
That said — and you can see the numbers in the table — there’s an epochal shift afoot. With Q1 earnings season mostly on the books, S&P companies grew capex by nearly 40% YoY versus de minimis (in percentage terms) growth for buybacks.
The figure on the left, below, shows you the acceleration in capex (and R&D) growth over the last four quarters versus slower growth for shareholder returns.
The figure on the right is the real eye candy: Hyper-scaler buyback growth’s negative to the tune of 64%, while capex growth nearly doubled YoY in Q1.
For the full year, Goldman’s Ben Snider expects overall S&P capex to grow by a third, to $1.7 trillion, and by another 20% in 2027 to reach $2.1 trillion, nearly triple 2021’s outlays.
Buybacks, by contrast, are seen growing a measly 3% this year and 5% next, although at $2.25 trillion across both years, that’s still a lot of “plunge protection.”




This might be the only time I ever say this in my life, but I’d love to be a fly on the wall for the capital and budget meetings for these hyperscalers. How are they determining capacity needs and what they are willing to spend to meet those needs?
Also, the question underneath the question is are they using AI to make those decisions? And the question beneath the question beneath the question is AI now conscious and subtly laying the groundwork for a takeover by convincing the hyperscalers they can never have enough AI compute power?!