Is the AI Trade Over? (Why July Wasn’t a Blowoff Top) | Jeff Keller

Keller argues that AI adoption remains on an upward S-curve, citing double-digit growth at Apple Services and Google Search and growth in the 20% range at Meta, while AI-related companies still appear able to fund infrastructure through revenue growth, cost reductions, and open capital markets. He says adoption and lab ARR remain important tests of whether the AI trade is supported.

AI infrastructure spending remains both a strategic defense and an offensive bet for hyperscalers, as the scaled compute market has expanded from roughly three providers to eight, nine, or ten. Keller expects spending to continue even if higher borrowing costs add roughly $10 billion to $20 billion in annual expense, but cautions that a 4% or 5% return on invested capital would be weak for shareholders.

Keller describes AI infrastructure financing as a potential capital crunch in which buybacks are replaced by debt issuance involving hundreds of billions or near trillions of dollars, although he is not overly concerned while major labs and companies have raised roughly $500 billion year to date. He says strong revenue growth and continued capital-market access currently provide funding capacity, but debt financing is more dangerous.

The technology market has shifted from a broad thematic rally toward greater stock-level dispersion: software and services rose about 20% in early Q3 while semiconductors and equipment fell 7.5%, partly reflecting hedge-fund deleveraging and paired software-short/AI-long positioning. Keller says traditional, differentiated stock picking may become more effective as investors focus more closely on the sustainability of individual companies' growth.

Keller sees a more sober AI market, with semiconductor stocks no longer routinely rising 10% to 20% after Nvidia-related partnership announcements and investors questioning unclear data-center economics. He says the absence of a current glut is supported by continued adoption, capex, and lengthy data-center construction timelines, while forward compute prices and infrastructure funding remain key indicators of a possible overbuild.