Hyperscalers Will Spend $750 Billion on AI This Year. Cummins Sells the Backup Power.

AI infrastructure spending is projected to reach ~$750B in 2026, and Cummins is positioned to benefit as data-center power demand strains the grid. In Q2 2026, Cummins’ power systems division grew revenues 19% YoY as AI-related customers increasingly need backup/prime power, while the engine business grew 6% YoY. The article flags valuation risk with Cummins trading at ~30x P/E vs a 5-year average of 17.5x, implying upside is tied to continued AI-driven demand.

Analysis

The real mechanism here is not “AI helps an industrial,” but that hyperscale buildouts are creating a parallel power market where speed-to-deploy matters more than traditional grid economics. That favors vendors with installed service networks and fuel-flexible generation, and it should pull mix toward higher-margin systems/service rather than commoditized engine sales. Over the next 1-3 quarters, the market will likely reward any evidence that this is recurring backlog rather than one-off project work.

Second-order, the trade is bigger than CMI: distributed generation demand can crowd out near-term utility load growth, lift natural-gas infrastructure volumes, and extend the life of gas-fired equipment suppliers. But it also creates a policy/permitting overhang; if local emissions rules tighten or grid interconnect queues clear faster than expected, on-site generation demand can normalize quickly. In that sense, this is a 6-18 month theme with a sharp reversal risk if power availability catches up.

Consensus likely gets the direction right but may overpay for the duration. CMI’s multiple already reflects a good deal of the AI narrative, so the risk/reward is better on dips than on momentum continuation, especially because the core business is still cyclical and not purely AI-exposed. What would falsify the thesis is a stall in power-systems orders, power-segment growth converging back to low-single-digit industrial levels, or management failing to raise backlog commentary on the next print.

Key Decisions for Investors

• Do not chase CMI at current levels; wait for a 5-10% pullback or a weak industrial tape before building exposure, since the AI power story is real but already partially capitalized.
• If you want exposure now, express it as a relative-value trade: long CMI / short XLI for 1-3 months to isolate the AI power beneficiary versus the broader industrial multiple risk.
• Set an earnings-alert on CMI for next quarter: only add aggressively if power-systems backlog and order intake reaccelerate; if not, trim because the multiple can revert quickly toward the mid-teens.
• Watch for spillover beneficiaries in electrical infrastructure and gas-linked names; if on-site generation becomes a sustained trend, rotate into the second-order winners rather than paying peak multiples for the first-order beneficiary.