The global economy is experiencing the largest capex cycle ever, with nearly $5 trillion seen by the end of the decade—and it’s not all AI spending

Global capital spending is being driven by two large cycles: AI hyperscaler capex and the energy transition, with the latter estimated at nearly $5 trillion by decade-end. Bank of America sees hyperscaler capex topping $800 billion this year and potentially reaching $1 trillion next year, supporting chipmakers and data-center suppliers. The article also highlights beneficiaries like Caterpillar and GE Vernova, with gas turbine demand reportedly sold out to 2030.

Analysis

The market is underpricing how much of this capex cycle is structurally sticky because it is being funded by cash-rich incumbents rather than balance-sheet-stressed utilities. That matters: hyperscaler spend is increasingly tied to revenue-generation and power-availability bottlenecks, which makes data-center buildout and grid interconnects less cyclical than typical tech capex. The second-order beneficiary is not just the obvious equipment vendors, but the entire “time-to-power” bottleneck stack — gas turbines, transformers, switchgear, EPC services, and industrial automation should retain pricing power for multiple quarters even if AI software sentiment cools.

CAT and GEV look like the cleanest ways to express the thesis because both sit closer to the physical bottlenecks than the AI narrative itself. The key nuance is margin durability: when lead times stretch into years, backlog quality improves and mix shifts toward higher-margin aftermarket/service revenue, which can keep earnings growth ahead of unit growth. The risk is less about demand disappearing and more about execution friction — if supply chains normalize faster than expected or if project delays push revenue recognition out, the market could de-rate these names despite healthy order books.

On the hyperscaler side, the winners are likely to be semis and networking vendors with pricing discipline, but the underappreciated loser is anyone exposed to power-constrained deployment timelines. If grid access becomes the gating factor, capex can move from a linear spend story to a lumpy “catch-up” cycle, creating intermittent disappointments in revenue ramps. BAC is a lower-conviction read-through: better for financing activity and capex velocity than for directional equity upside.

The contrarian takeaway is that the market may still be too focused on AI as a pure software-duration trade and not enough on it as an industrial-capex and power-shortage theme. That means the upside could persist even if AI multiples compress, because the physical buildout has its own multi-year inventory and infrastructure replacement cycle. The main reversal trigger is a sharp slowdown in hyperscaler revenue or a policy shock that relaxes power constraints faster than expected, which would hit the scarcity premium embedded in the industrial beneficiaries.

Key Decisions for Investors

• Long CAT vs short a broad industrial basket (XLI) for 3-6 months: CAT should capture the highest-quality backlog and aftermarket mix from the capex supercycle, while the short hedges a generic industrial slowdown if the theme is over-owned.
• Long GEV into any 5-7% pullback over the next 2-4 weeks: the risk/reward is attractive because turbine scarcity and long lead times should sustain pricing, but size modestly due to execution and delivery-timing risk.
• Buy MSFT/AMZN calls out 6-9 months vs short-dated puts on an electric utility ETF: this expresses the view that AI demand is driving power bottlenecks faster than the market expects, benefiting hyperscaler capex more than regulated load growth.
• Pair long CAT / short BAC for 3 months: CAT has direct operating leverage to physical capex, while BAC is mostly a financing facilitator with limited standalone earnings torque from the cycle.
• Avoid chasing GOOGL/META/AMZN/MSFT after sharp post-earnings rallies; instead wait for a 5-8% consolidation to add exposure, since the next leg higher depends on capex staying paired with accelerating revenue rather than just spend alone.