Apollo Global Management, one of the world's largest alternative asset managers, has circulated a presentation whose title is the thesis: no signs of AI payoff outside the tech sector. The implication lands hard given the scale of capital flowing into AI infrastructure — estimates now exceed $200 billion annually in combined capex from hyperscalers alone. The structural claim is straightforward. Technology companies are spending aggressively on AI compute, chips, and model development. Some of that spending is showing up as revenue and margin expansion inside the tech sector. But in the sectors where AI was supposed to deliver transformative productivity — healthcare, manufacturing, logistics, financial services, education — the macro data shows no inflection. Productivity growth outside tech remains anemic, tracking pre-AI trends. This is the J-curve argument's most uncomfortable moment. Proponents of large AI investment have long argued that productivity gains lag capex cycles by years — electricity took decades, the internet took a decade. But the electricity and internet analogies cut both ways: both also produced massive capital destruction among firms that invested early and wrong. The question is whether we're in 1996 (early but directionally right) or 1999 (late and about to correct). Apollo's framing matters because of who they are. This isn't an academic exercise or a tech-skeptic blog post. Apollo manages over $600 billion in assets and makes allocation decisions based on exactly this kind of analysis. When a firm with that much capital at stake says the productivity signal isn't showing up, the audience is institutional investors deciding whether to keep funding the buildout. The extraction pattern is clear even if the presentation is short on specifics. Nvidia, the hyperscalers, and the model labs are capturing value from AI spending. Everyone else is paying — either directly through enterprise AI contracts that haven't yet delivered ROI, or indirectly through inflated equity valuations that assume productivity gains that haven't materialized. The costs are real and current; the benefits remain speculative and concentrated. The deeper concern is capital misallocation at scale. If hundreds of billions in annual investment are flowing into infrastructure that primarily benefits the infrastructure vendors themselves, that capital isn't going to housing, energy transition, healthcare capacity, or other areas where the productivity drag is acute and measurable. Opportunity cost is the invisible tax on AI hype. None of this means AI is useless or that productivity gains will never materialize. It means the honest read on the data right now — per one of the most sophisticated capital allocators on the planet — is that we're in a seller's market for AI infrastructure and a buyer's desert for AI outcomes. That gap is the story.
