Morgan Stanley Says AI Data Centers Face a 38-Gigawatt Power Gap. These Industrial Stocks Fill It.

Morgan Stanley flags a potential 38 GW U.S. data-center power gap (68 GW needed for 2026–2028 vs. ~30 GW available/under construction), which would likely push developers toward faster behind-the-meter generation. GE Vernova reported a Q2 gas power backlog/slot reservations of 116 GW (vs. 100 GW prior year) and expects at least 125 GW under contract by year-end; Eaton’s Q2 data-center orders rose ~85% YoY with ~65% revenue growth, and it is expanding AI liquid-cooling exposure via its $9.5B Boyd Thermal deal expected to drive ~$1.7B in 2026 sales; Vertiv posted Q2 revenue up 24% to $3.27B, adjusted operating profit up 51%, adjusted EPS up 60% to $1.52, and adjusted FCF rising to $925M, with 2026 guidance around $14B revenue at the midpoint. Net: the thesis is structurally bullish for power-generation, grid/electrical, and cooling equipment tied to AI data-center buildouts.

Analysis

The cleanest expression is not “AI beta,” but a bottleneck trade on power delivery. The market is still underappreciating how much of the next leg of AI capex migrates from semiconductors into electrical balance-of-plant, which favors ETN and VRT on a faster revenue conversion path, while GEV gets the most torque from the power-generation substitution effect. Second-order winners also include copper, specialty transformers, gas-turbine service, and EPC capacity; the losers are hyperscalers and colocation operators if they cannot secure power fast enough, because that pushes out utilization and delays monetization.

The main catalyst path is 1-3 months, not 1-3 years: order announcements, backlog conversion, and guidance raises matter more than headline construction plans. The risk is that a lot of the good news is already in the multiples, especially for ETN and VRT, so even strong fundamentals may not expand valuations if the market decides the constraint is supply-chain, not demand, and therefore only shifts revenue timing. For GEV, lead times on turbines and interconnect equipment can create a mismatch between backlog and cash conversion, so the stock is more exposed to execution slippage and working-capital drag than the narrative suggests.

Contrarian view: the consensus is probably too linear on the size of the addressable market and too optimistic about timing. A power shortage does not automatically translate into immediate earnings upside if utilities, permitting, gas infrastructure, and transformer availability remain the true bottlenecks; in that case, the winners are still real, but the path to EPS is slower and more volatile than the market implies. The thesis is falsified if hyperscaler capex growth decelerates, if order growth normalizes below high-teens/low-20s, or if supply constraints push out delivery schedules enough to cap near-term revenue recognition.

Key Decisions for Investors

• Long ETN / short XLI as a 3-6 month relative-value trade: ETN has the clearest mix of demand visibility and pricing power, while the broader industrial complex is less directly exposed to AI power bottlenecks.
• Initiate a starter long in VRT on pullbacks, with a 1-3 month horizon into next earnings: the stock is the most execution-sensitive but also the cleanest beneficiary of higher rack density and liquid-cooling adoption; use a tight stop if order growth fails to hold above current run-rate.
• Buy GEV only on confirmation of further backlog or contract conversion, not on the headline story alone: the risk/reward is best if management reiterates turbine slot tightness and data-center-related electrification orders, but the position should be smaller than ETN/VRT because cash conversion is more back-end loaded.
• Avoid paying up for pure semis here; consider a relative short in NVDA vs. ETN/VRT only if market starts rotating capex from compute to infrastructure, because the thesis is a mix-shift trade rather than a broad AI-universe bearish call.