Top economist says AI just hasn’t delivered on the productivity hype—and it means a ‘painful repricing’ of markets is very possible

Apollo’s chief economist Torsten Slok warns AI ROI is lagging expectations outside tech, citing Magnificent Seven profit margins rising from ~15% to ~25% (Q1 2023–2026) versus ~10% for the rest of the S&P and ~12% for the Bloomberg 500. He highlights a Harvard/BCG-style concern that only ~5% of companies saw meaningful ROI from generative AI pilots and cites Ricoh as a case where AI increased division productivity but cost overrun drove total expense to ~3x manual work (with ~$500k in outside consultants and ~$200k/month in AI fees). If earnings expectations continue to outrun realized ROI, Slok says markets may face a “painful repricing” as firms slow AI spending without quick payback.

Analysis

This is a duration and valuation problem more than a demand problem. The market has been paying for rapid AI monetization, but the real payoff is gated by systems integration, data hygiene, governance, and workflow redesign — all of which push cash returns out by quarters, not weeks. That favors implementation-heavy vendors and services providers, while leaving the high-multiple AI beneficiaries vulnerable to multiple compression if earnings do not convert capex into visible productivity faster than expected.

NVDA is still insulated by hyperscaler inertia, but the second-order risk is slower growth at the margin, not an outright collapse in demand. META is more fragile because AI spend hits both capex and opex before monetization is clearly separable, so even modest cost creep can pressure the margin bridge. IBM is relatively better positioned if enterprises need human oversight and consulting to make AI usable; Apollo can also benefit if a repricing creates more distressed/growth-credit deployment opportunities.

The consensus is missing that AI can be real economically and still disappoint public-market assumptions because the gains are partly offset by implementation overhead and hidden labor. The key falsifier is not whether AI works, but whether enterprise ROI shows up quickly enough to defend 2026 margin expectations. Watch the next two earnings cycles for narrower AI use cases, token-budget discipline, and any indication that procurement teams are slowing spend rather than broadening it.

Key Decisions for Investors

• Short NVDA on rallies via 1-3 month put spreads ahead of the next hyperscaler capex/read-through cycle; thesis breaks if data-center growth and backlog commentary stay strong for two consecutive prints.
• Pair long IBM / short META over the next 1-3 months: implementation and services should prove more durable than a platform story that needs AI to lift revenue faster than it lifts costs.
• Use APO as a relative hedge or add-on long only if the AI complex starts to de-rate; a repricing would likely increase opportunistic financing and private-market deployment opportunities over 6-12 months.
• Set an alert for any enterprise AI ROI disclosure in the next earnings season: if management teams start cutting tokens, seats, or pilot budgets, extend the short exposure; if they show measurable payback, cover quickly.