Dimon Says AI Capital Spending Will Hit $725 Billion in 2026. Here Are the Sectors That Will Win and the Ones That Will Be Left Behind

Jamie Dimon projects the five hyperscalers (Microsoft, Amazon, Alphabet/Google, Meta, Apple) will increase annual AI-driven capital spending from $450 billion in 2025 to $725 billion in 2026 — a >60% rise. Primary beneficiaries are data-center REITs (Equinix, Digital Realty — forward yields ~1.9% and 2.7%), AI chipmakers (Nvidia, Broadcom), foundry TSMC and optical suppliers (Lumentum, Corning), while legacy cloud/software vendors (Salesforce, ServiceNow) and smaller/unprofitable AI software firms (C3.ai) face competitive displacement.

Analysis

The capex wave is changing the unit economics across the stack: training remains GPU-heavy and sticky, while inference and edge deployments are increasingly ASIC-driven, which favors foundries and silicon integrators with long-term supply rights. That creates a two-speed hardware market where ASPs for datacenter GPUs can stay elevated through a concentrated 12–24 month buy cycle even as inference accelerators compress per-inference cost.

Second-order supply effects matter more than headline demand. Tight wafer allocation at leading foundries and prioritized wafer buys for flagship AI customers will compress available inventory for smaller silicon customers, amplifying pricing power for incumbents and lengthening lead times — a scenario that can lift margins for selected chipmakers but signal trough-to-peak cyclicity lasting multiple quarters. At the same time, enterprise software vendors that rely on downstream distribution through clouds face margin and renewal risk as platform owners both bundle AI primitives and internalize R&D.

Key tail risks and timing: macro-driven capex pullbacks or a rapid move higher in real yields will reprice REITs and slow the upgrade cadence within 3–9 months; conversely, supply-side shocks (fabrication outages, equipment lead times) could accelerate pricing power for suppliers over 6–18 months. Regulatory or open-source LLM developments pose a structural reversal over years by changing where compute is consumed (on-prem vs hyperscale) and who captures enterprise spend.

Key Decisions for Investors

• Long NVDA (6–12 months): buy a calendar/LEAP call (e.g., 9–12 month expiries) or a 6–9 month call spread to limit premium decay. Rationale: capture sustained GPU ASPs and upgrade cycle; risk = 100% premium if cyclical demand weakens; reward = asymmetric upside if allocation-led pricing persists (target +30–80% if earnings and guide beat).
• Long AVGO or TSM (6–12 months): buy cash shares or buy protective-call/put collar to reduce drawdown. Rationale: ASIC adoption and foundry exposure; risk = 20–30% on cyclical reset or foundry oversupply; reward = 25–50% if ASIC content share and fab pricing hold.
• Real-asset/infra play with downside protection (12–24 months): buy EQIX with a 12-month, ~10% OTM put to cap large drawdowns from cap-rate repricing. Rationale: exposure to datacenter footprint growth while guarding for rate shocks; risk = put premium + stock downside; reward = total return from rental reversion + limited capital upside of 20–40%.
• Pair trade (6–18 months): long MSFT or AMZN, short CRM and NOW (size to neutralize beta). Rationale: hedge market exposure while expressing divergence between platform bundlers and legacy SaaS margin/retention deterioration; risk = platform execution miss or SaaS resilience; reward = capture multiple compression on legacy vendors versus platform re-rating.