Much of the 2026 surge in US corporate profits was one company's chip-and-datacenter bill becoming another company's revenue, not new money actually earned.
In September 2026 the US ten-year Treasury yield climbed from 4.75% to 5.29%. Across 770 monthly moves since 1962, that lands in the top 10%, and stocks still rose that month, adding $2.5 trillion in total market value.
Profits did jump, just not evenly
Strip it down: in the second quarter of 2026, net income at US listed companies went from $576 billion a year earlier to $904 billion, up 57%.
But more than half of all companies reported falling profits. Technology, financials and communication services carried the market. Capital spending across listed companies rose $133.4 billion year on year, up 36%, with technology, communication services and consumer discretionary all above 50%.
The money moved from one set of books to another
Cash spent on land, datacenters and chips is an expense for whoever builds them. It drags down profit and free cash flow.
The same money becomes revenue at Nvidia and TSMC, which is why Nvidia has been the biggest winner, with power producers and datacenter landlords taking a cut too.
One accounting detail matters here: datacenter spending is deducted over many years, and the longer the schedule, the lighter the hit to profit today.
Paper gains versus cash in hand
Profits soared; money handed back to shareholders did not follow. For two decades, S&P 500 companies returned 80% or more of earnings through dividends and buybacks, sometimes over 100%.
By 2026 that ratio had fallen to 63%, the lowest since 2004. Where did the rest go? Into AI.
The real question is whether those datacenters eventually earn more than the cost of the money borrowed to build them. If yes, everyone is fine. If no, accountants will write down the spending, stock prices will fall first, and the books will admit it later.
The author's own choice: he still holds Amazon, Alphabet, Meta and Microsoft, all heavy AI spenders, but new money has gone mostly into 4% short-term Treasuries. Sleeping well beats squeezing out every last point.
One plain sentence to end on: if AI fails to deliver, waiting for the write-down before you sell means you waited too long.
Why it matters
The easiest mistake in this earnings season is reading one company's expense as another's revenue and concluding that the whole economy is richer. Telling apart genuinely new demand from money merely changing pockets is what decides whether this is a bet worth adding to.
Curated from high-quality sources, with concise summaries and key takeaways.
