AI earnings quality
The S&P’s first-quarter net margin was 14.8%. Yardeni calls it housecleaning. Jakab calls it steroids. A third story says the kill-switch is credit.
In the first quarter the S&P 500’s net margin was 14.8% — the highest on record, about twice the postwar average. Information technology and communication services are already 45% of the index. Three stories explain the same print. They cannot all be the one that matters.
The margin everyone is arguing about
14.8% is Jakab’s first-quarter print. The ~7% postwar average is the mean-reversion Yardeni’s story has to beat. Same gap, opposite stories.
Stop the music
Ed Yardeni, a veteran Wall Street strategist selling a research product and bullish since 2009, does not treat 14.8% as a sugar high. AI, he says, is evolutionary, not revolutionary — a step in a digital revolution that started with the mid-1960s IBM mainframe. Data is a fourth factor of production, and there will never be a shortage of it. What AI forced was a freeze: stop hiring, stop firing, walk the company floor. Which departments stay. Which shut.
I don’t think that we’re taking earnings away from the future. I think we’re kind of building on the productivity of these companies.
Ed Yardeni, The Compound, August 2026
Those IT and communication firms, 45% of the S&P, do creative destruction better than anybody else. Margins trend up. They do not mean-revert to 1985. Operating earnings, stripped of Anthropic-style write-ups, are real; the hyperscalers wish they had more compute today. 1999 was a price-to-earnings rally without the earnings. “This time around it’s grounded on earnings.” Michael Batnick’s unusual tape: the six-month forward multiple is down 8% while the index price is up 11%.
The arithmetic if there is no recession: $403 of earnings times 20 at year-end 2026, about 8,250. By the end of 2029, $500 times 20 is 10,000.
Hell to pay later
Spencer Jakab, in the Journal and on The Compound, reads the same margin as steroids. Mid-quarter, analysts raised S&P 500 earnings forecasts 3.4%. Over the prior forty quarters the average move was a 2.7% cut.
Two accounting effects do the work. Hyperscalers spend now and depreciate slowly, so the buyer’s cost drips through the P&L while the supplier books the revenue today. And GAAP makes public holders of private AI stakes — Nvidia, Alphabet, the rest — mark those stakes up every time a funding round reprints the valuation. Bao Lan Wang at the University of Florida put the write-ups at about 12% of first-quarter S&P net profit, and preliminarily two to three times that for the second quarter. Hyperscalers booked about 30% of income earlier this year from the same markups.
Hold the multiple constant and cut the margin from 15% to 11%, Jakab said, and the market should be 25% lower. “There’s going to be hell to pay for that later on when it goes into reverse.”
Michael Santoli, on The Compound six days after Yardeni’s appearance, named the same worry without picking a side. “I’m more worried about over earning and overstating of earnings than anything else. But they’re doing earnings are just supporting things.” On a forward price-to-free-cash-flow basis, he said, the S&P is at 30. “Don’t love that.” Commentator color on that episode. Not a filing, and not a desk mark.
The kill-switch was the Fed
A third story says the argument about productivity is the wrong argument. Chamath Palihapitiya’s “token costs doubling every 45 days” runs the wrong way on the unit price: blended AI cost fell 67% year-on-year, from $18.40 to $6.07 per million tokens, on a later pass that still wants a primary price series. Volume is what is exploding. Realized enterprise ROI of “0 to 2%” traces to a contested MIT NANDA study. Neither Yardeni’s productivity nor Chamath’s near-zero return is measured.
Nick Colas’s 1999 correction is the load-bearing one. Spender stocks rolled first. That did not end the cycle. Leadership rotated. The thing that killed it was “110% the Fed” — the cost of capital. The live tells are in credit: S&P cut Oracle to BBB- in July 2026 on capex and cash-flow strain; Meta’s free cash flow collapsed to $784 million from $12 billion a quarter earlier.