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

AI infrastructure spending is projected to reach ~$750B in 2026, supporting Cummins’ power generation opportunity as data centers face grid constraints. Cummins reported power systems division revenue up 19% YoY in Q2 2026 (vs. 6% growth in the engine business), as its engines increasingly power and provide always-on/backup electricity. The article flags valuation risk with a ~30x P/E versus a 5-year average of 17.5x, but frames the AI-driven demand outlook as a growing tailwind.

Analysis

Cummins is being re-rated less as a truck-cycle industrial and more as a constrained-power supplier to AI buildouts, which is the right framing if you expect grid interconnect delays to persist. The market is likely underappreciating the mix shift: backup generation can become prime power for off-grid data centers, and that mix usually carries better pricing discipline and aftermarket pull-through than core engine sales. The key question is not demand, but durability—whether this is a multi-year capex leg or a temporary bridge until utilities catch up.

Second-order winners are the gas-fueled distributed-power ecosystem and adjacent service providers; the losers are regulated utilities and grid-dependent developers if customers keep bypassing the grid. A more subtle implication is for AI hardware names: if power is the bottleneck, GPU demand is not destroyed, but shipment-to-revenue conversion can slip by quarters, which can soften near-term expectations for names like NVDA. SPCX-style private, self-powered data centers validate the theme, but they also show how fragmented and non-scalable the solution can be.

The contrarian risk is that the consensus is capitalizing a shortage into a permanent AI moat. That is usually wrong in industrials: when interconnect queues shorten or projects secure utility PPAs, the scarcity premium fades fast and the multiple compresses before growth disappears. The bullish case is falsified if power-systems growth does not stay meaningfully ahead of the core engine business for the next 2-3 quarters, or if margin expansion stalls despite strong order commentary.

Key Decisions for Investors

• Long CMI on a pullback rather than chasing the multiple: use weakness into the next industrial tape break to build a 3-6 month position, with the thesis invalidated if power-systems growth decelerates back toward core engine growth.
• Pair trade: long CMI / short XLU for 1-3 months as a relative-value bet on grid bypass and distributed generation; exit if utility-capacity additions or faster interconnect approvals reduce the need for customer-owned power.
• Avoid paying up for NVDA purely on the power-shortage narrative; use any rally in the stock to trim if data-center deployment delays start showing up in guidance, since the bottleneck can defer revenue recognition even when end-demand remains intact.
• For higher beta, small starter long NGS as a secondary beneficiary of gas-fired distributed power, but only if upcoming order commentary confirms AI-related demand; keep size modest because the story is more cyclical and less proven than CMI.