AI boom faces reality check as returns lag behind massive spending

Goldman Sachs says roughly 95% of companies are seeing little to no ROI from AI despite tens of billions of dollars in spending, underscoring a widening gap between adoption and monetization. The report highlights that consumer usage is growing quickly, but enterprise profitability remains weak, with semiconductor firms capturing most of the gains while model developers, cloud providers, and corporates struggle to justify outlays. The message is cautious for the AI investment theme and suggests current capex levels may be unsustainable without clearer returns.

Analysis

The market is still pricing AI as a straight-line capex supercycle, but the more important signal is that returns are concentrating at the component layer while the software/application stack remains economically weak. That is usually a late-cycle pattern: infrastructure vendors capture near-term scarcity rents, but once buyers realize enterprise workflows are not changing fast enough, spend growth decelerates and multiple compression follows. The strongest second-order effect is not on model developers first, but on companies exposed to AI-capex budgets and cloud digestion cycles where revenue recognition lags hardware orders by quarters.

For Goldman specifically, the near-term read-through is mixed: AI enthusiasm can support advisory and underwriting activity, but if the market starts to question whether incremental AI investment is value-destructive, the premium embedded in large-cap tech and growth equities becomes more vulnerable. The most exposed names are those with the largest announced spend plans and weakest evidence of monetization, because investors may start demanding proof of ROI rather than rewarding optionality. That creates a window where beneficiaries of the buildout can outperform even as the broader AI basket underperforms.

The contrarian view is that consensus is still underestimating how long the spend can outrun monetization. In prior platform transitions, adoption penetration looked economically disappointing for years before compounding utility became visible in P&Ls, so this may be more of a valuation air pocket than a structural thesis break. A reversal likely needs one of two catalysts: either a visible enterprise killer app that converts pilots into workflow lock-in, or a capex pause from a major hyperscaler after two quarters of slowing bookings. Until then, the asymmetry favors selling expensive beneficiaries of the narrative and owning the picks-and-shovels with real pricing power.

Key Decisions for Investors

• Short a basket of high-multiple AI software names versus long AI semis for 3-6 months; the pair works if the market continues rewarding tangible revenue capture over vague enterprise adoption. Favor NVDA/AVGO as the long leg against a basket of unprofitable AI application names.
• Reduce exposure to hyperscalers with the heaviest AI capex trajectories if next earnings show accelerating spend but flat-to-slower cloud growth; use a 1-2 quarter horizon and keep tight stops if bookings re-accelerate.
• Buy downside protection on mega-cap growth indices via 3-6 month put spreads; the setup is a valuation reset, not a broad market crash, so defined-risk hedges are more efficient than outright shorts.
• Overweight semiconductor equipment and select power/infrastructure names where spending is already contracted and less dependent on enterprise ROI proofs; these benefit even if software monetization disappoints.
• If GS is trading at a discount to other bulge-bracket firms, consider a relative long GS / short a more AI-exposed broker-dealer only if underwriting and advisory remain resilient; otherwise, avoid making GS the expression of the AI debate.