Stocks will tumble much further as AI investment will slow next year then fall in 2028 and drag GDP down too, economist predicts

Capital Economics expects hyperscaler AI capex to fall about 10% in 2028 after continued but slower growth next year, with Redwood estimating a 10-fold increase in external AI revenue by 2030 would be needed to justify current investment plans. She forecasts the S&P 500 will be 12% below its current level by end-2027, or 17% below its early-October peak, and expects larger declines in tech-heavy indices. Redwood estimates the capex slowdown could weigh on U.S. GDP, but does not expect a major downturn; she projects AI will lift U.S. productivity by about 1% annually for several years.

Analysis

The key risk is not that AI demand disappears; it is that infrastructure suppliers have priced a continuing acceleration in spend while buyers can preserve high absolute investment and still slow its growth. That distinction makes quarterly capex-growth revisions, utilization, and returns on deployed capacity more important than headline budget totals. The external-revenue test may also understate strategic value: internal labor savings and product retention can justify spending without a proportional new AI-revenue line. But those benefits accrue unevenly and may take longer to appear than hardware orders.

The most exposed are capacity-sensitive parts of the AI supply chain—accelerator vendors, semiconductor equipment, memory, networking, and data-center construction—where delayed orders can propagate through inventories and utilization. Taiwan and South Korea face a larger macro transmission channel through export and investment exposure. Conversely, continued high spending for replacement, inference growth, or custom silicon could support infrastructure demand even if hyperscaler budgets flatten; power and cooling constraints may redirect spend rather than eliminate it.

Near term, this is a positioning and earnings-revision risk, not a standalone recession call. Over 1–3 months, track hyperscaler capex guidance alongside cloud growth, AI service monetization, and supplier backlog conversion. Over 6–18 months, the split between productivity beneficiaries and infrastructure sellers should widen. The contrarian offset is that capex can decelerate from an exceptional pace while remaining elevated, and productivity gains may cushion the macro drag. The thesis weakens if capex guidance holds, deployed capacity utilization rises, and AI-related revenue or customer savings become measurable; it strengthens if guidance is cut while supplier inventories rise or financing conditions tighten.

Key Decisions for Investors

• Trim or hedge concentrated AI-infrastructure exposure rather than shorting the entire technology complex. A conditional relative-value expression is short accelerator/equipment-heavy exposure against cash-generative enterprise software or diversified technology, sized modestly until capex revisions confirm the thesis.
• Use upcoming hyperscaler earnings as the catalyst window: watch changes in capex growth, cloud margins, depreciation, utilization, and backlog conversion. Treat stable absolute capex with slower growth differently from outright cuts; the latter is the stronger negative signal for suppliers.
• Avoid extrapolating the U.S. GDP estimate to a broad recession trade. Taiwan- and South Korea-sensitive exposure has greater direct capex/export sensitivity, but verify actual export orders and earnings revisions before adding regional shorts.
• For an options hedge, consider defined-risk put spreads on highly AI-capex-sensitive exposure ahead of guidance events rather than open-ended shorts; unwind or reduce if spending guidance and supplier order trends remain intact.