Big Tech Is on Pace to Spend $735 Billion on AI Data Centers in 2026. These 3 Industrial Stocks Collect the Checks.

AI capex tailwinds are spilling into industrial infrastructure: Amazon plans $220B capex for 2026, Microsoft $175B, Alphabet up to $205B, and Meta $135B. Caterpillar’s Power & Energy revenue rose 17% YoY to $8.2B, while Eaton posted record Q2 revenue of $8.5B (+21% YoY) with data-center-related Electrical Americas data center revenue up 65% and electrical backlog up 43%. GE Vernova reported +22% revenue growth with +88% orders, and raised 2026 guidance (total revenue $45.5–$46.5B; free cash flow $11.5–$12.5B) after data center orders exceeded $5B YTD (more than double 2025).

Analysis

The market is still mispricing the AI spend stack as if it ends at chips. The cleaner monetization is upstream in power conversion, backup generation, switchgear, and grid interconnects, where lead times and installed-base relationships create pricing power and backlog visibility. That argues for a longer-duration earnings upgrade cycle in ETN and GEV than in the more cyclical, less pure CAT exposure; CAT can participate, but it is more sensitive to construction timing and less to the actual power bottleneck.

Second-order winner dynamics favor suppliers that can capture the same dollar of capex multiple times: equipment sale, installation, service, and replacement parts. The likely losers are the parts of the ecosystem that depend on grid throughput improving quickly — utilities with constrained transmission queues, and eventually some GPU-demand timing if data center commissioning slips. In other words, the AI thesis is not derailed, but monetization shifts from semiconductor volume to infrastructure scarcity, which can extend the cycle and support industrial multiples longer than consensus expects.

The main risk is that investors front-run a multi-year build-out with a few quarters of backlog data and overpay for names where the AI mix is still a small percentage of revenue. If hyperscalers slow capex growth, or if permitting/interconnection eases faster than expected, the scarcity premium in ETN/GEV can compress quickly. The contrarian view is that CAT is probably the least differentiated expression here; the move is likely overdone if investors are buying every AI-adjacent industrial indiscriminately rather than isolating the power bottleneck beneficiaries.

Key Decisions for Investors

• Long ETN on pullbacks for a 6-12 month hold; best risk/reward from backlog conversion and pricing power in electrical infrastructure. Falsify if data-center-linked orders decelerate for two consecutive quarters or organic growth guide rolls over.
• Long GEV vs short CAT as a 3-6 month relative-value pair. Thesis: GEV has the tighter tie to turbine scarcity and grid build-out, while CAT’s AI exposure is more diffuse and more exposed to general construction cyclicality. Exit if CAT reaccelerates relative to GEV on new power-gen orders.
• If already long AI semis, rotate a portion into ETN/GEV rather than adding to NVDA here; the marginal dollar of AI capex is increasingly captured by infrastructure, not compute. This is a lower-beta way to stay exposed to the same capex theme.
• Set a watch item on utility and grid capex commentary over the next 1-2 earnings seasons; any evidence of faster-than-expected interconnection relief is a negative catalyst for the scarcity premium in ETN/GEV.
• Avoid chasing CAT after strength unless you get confirmation that power-generation demand, not just construction, is driving the upside; otherwise the name risks becoming the 'AI industrial' that trades on narrative faster than fundamentals.