The S&P 500's 32% Boom Is Narrower — and More Circular — Than It Sounds

The headline says the S&P 500's earnings are growing 32% this year and AI did it. That much is true. What the headline leaves out is who is paying for it — and that is the part that decides whether the boom is real or just a circle of companies selling to each other.

Start with why the number sounds extraordinary. Streaks of 30%-plus earnings growth have happened only ten times since 1936, according to Bank of America. Almost every previous streak came on the rebound from a deep earnings recession — the 2008 crash, the 2020 Covid collapse. This one had no recession at all. It's the rare case where the index grew that fast on its own, in the middle of a normal economy, which is exactly why Wall Street keeps marking it up: the 2026 forecast is 32%, revised from 24% before the second-quarter season and roughly 15% at the start of the year.

But here is the part worth your attention: the growth is not spread across the market the way a booming economy usually is. It is concentrated in a handful of companies. Three AI hyperscalers — Alphabet, Amazon, and Meta — account for about 70% of the S&P 500's expected earnings growth for 2026. The rest of the index is doing fine. Those three are carrying the year.

That concentration matters more once you look inside the "record" quarters the headlines were built on. The second quarter produced the strongest year-over-year earnings growth the S&P 500 has reported since 2021. Yet a large slice of that number was not operating profit at all. Alphabet took a roughly $98 billion gain in other income, mostly unrealized gains on its equity investments; Amazon booked a $53.4 billion gain on its stake in Anthropic. Take those two one-time, mark-to-market items out and the quarter's growth falls from about 50% to roughly 32%. JPMorgan's analysts run the same test at the full-year level, stripping the effect of private-company stake valuations to get a more honest growth number.

Which brings us to the mechanism underneath — and this is the part a beginner should understand before trusting the story. A big share of the "earnings boom" is a closed loop. The hyperscalers are spending enormous sums on AI infrastructure — more than $754 billion of capital expenditure forecast for 2026, heading toward $900 billion by year-end and over a trillion in 2027. That spending shows up as revenue and profit on the other side of the loop: the Nvidia chips, the memory, the power and construction. One group spends, the other books the sale, and the earnings growth is real on both sides of the ledger — but it all depends on whether someone outside the loop eventually pays for it.

That is the whole question, and there are early signs the loop is already straining. Nvidia — the biggest beneficiary of all this spending — has warned its profit margins will narrow as memory costs surge. Apple blamed rising memory prices for a disappointing sales outlook. When the two most visible AI companies are simultaneously reporting that their input costs are inflating, that is the price of the build-out making itself known. It has not broken the loop yet, but it is the first sign that building is expensive long before the payoff arrives.

Set that against what the market is charging for the boom. By one measure the index looks cheap: its forward price-to-earnings ratio sits near 20, below its own five-year average, because the "E" keeps climbing faster than the "P." That is how the bulls argue a market at record highs is still a bargain. But the backward-looking measures tell the opposite story. The Buffett indicator — total market value against GDP — is at an all-time high, and the Shiller CAPE ratio is near dot-com levels. Bank of America warns that if investors ever stop treating AI capex as a permanent industrial revolution and start treating it as an ordinary spending cycle that must eventually prove its returns, the multiple could de-rate by 15% to 30%. A multiple that high is the market betting the growth compounds without interruption.

So what changed with this report? Nothing about the direction of the boom — it is real, it is rare, and the numbers are honestly reported. What the data exposes is the composition of the growth. Three companies generate most of it, a meaningful slice was one-time accounting gains, and the durable core is a spending circle that has not yet been validated by final customer demand. That is not a reason to panic, and it is not a reason to sit out. It is the difference between buying a boom and understanding who is funding it. The number that will tell you whether the circle is closing is not earnings — it is whether the AI spend converts into bought-with-cash revenue instead of just circulating. Watch the hyperscalers' next capital-spending guidance and their margins. If the spend keeps rising and the monetization keeps arriving, the growth compounds and the price was cheap. If the spend rises and the returns don't arrive, the earnings were always going to be the easy part — and three companies having already decided the outcome is not the safety the index's 32% makes it feel like.