Chamath Palihapitiya said the AI boom may be hiding the biggest capital allocation mistake in history, arguing that only 0% to 2% of the S&P 493's recent earnings growth came from AI productivity. For companies spending heavily on AI, that leaves a harder test than adoption alone: the spending has to beat the return on cash left on the balance sheet.
9% earnings-per-share growth is what the S&P 493 has produced since generative AI entered the mainstream, Palihapitiya said, but he attributed almost all of that gain to inflation and buybacks instead of AI productivity. On a recent episode of the All-In podcast, he said the AI return-on-investment chickens are finally coming home to roost.
PwC survey on 56% of CEOs
56% of CEOs in the PwC CEO Survey said they saw no AI revenue or cost benefit, while 12% said AI delivered both higher revenue and lower costs. That split gives Palihapitiya's argument a second leg: broad AI spending is still moving faster than measurable payback for many buyers, even as the corporate sell side keeps selling the infrastructure and tools.
Hundreds of billions of dollars are still flowing into AI buildouts, and Nvidia cannot manufacture AI chips fast enough to satisfy demand. Palihapitiya's point is that buyers and sellers should not be lumped together: the companies selling AI can still benefit from the surge while the companies buying it have to show something tighter than enthusiasm.
Cash on the balance sheet
Palihapitiya said AI adopters must now prove the technology generates returns above the risk-free rate or justify why cash was not left on the balance sheet. In practical terms, that means a company deploying AI has to show more than activity or headcount savings; it has to clear the hurdle any low-risk alternative already offers.
For Corporate America, the decision now looks less like a race to buy and more like a capital-allocation test. If AI spending does not raise earnings, lift costs down, or surpass a basic return benchmark, the safer move is to keep the cash and wait for a clearer payoff.
