The old tech fairy tale is getting a rewrite
NYU’s Aswath Damodaran — basically the professor version of a laser pointer at a valuation conference — says 2026 is splitting corporate America into two camps: the companies writing the AI checks and the companies cashing them.
The headline numbers sound great at first blush. Earnings are still buoyant, indexes are hanging in there, and capital spending at U.S. public companies hit $507.4 billion in the second quarter, up 35.66% year over year. But that shiny top line comes with a not-so-shiny catch: free cash flow is getting vacuumed up by AI infrastructure spending.
Builders vs. pickaxes sellers
Damodaran’s point is pretty simple: the hyperscalers — Alphabet, Meta, Microsoft, and Amazon — are turning capex into an accounting and cash-flow problem. They can smooth the pain a bit by capitalizing spending and stretching amortization, but the cash still leaves the building. As he put it, there’s basically no world where the year you spend big on capex is the year your earnings magically pop.
Meanwhile, the suppliers get to ride shotgun. Nvidia, TSMC, and Broadcom are the ones turning those giant data-center tab bills into revenue and, in many cases, fat margins. So yes, AI capex can lift earnings — just not necessarily for the companies doing the spending.
What this means for your portfolio
Damodaran laid out four ways to think about the shift:
- go full speed into AI builders and suppliers, while shorting the old economy laggards;
- stay passive with broad index exposure like the S&P 500;
- hide out in defensive pockets like food processing, leisure, or short-term Treasuries;
- or go full contrarian and bet the AI trade eventually cracks.
His broader warning is the one investors should hear over the noise: the market is treating these megacap tech names like they’re still featherweight software businesses, when they may be morphing into capital-heavy utilities with a fancier logo.
Big picture: the AI boom may still be a winner, but the winners are starting to look a lot less evenly distributed than the bull case suggests.