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

Morgan Stanley estimates U.S. data centers face a potential 38GW power gap (68GW needed vs. 30GW supply/under construction through 2028), pushing demand for behind-the-meter generation and grid/electrification. GE Vernova reported Gas Power backlog/slot reservations rising to 116GW (from 100GW) with guidance for at least 125GW under contract by year-end, alongside >$5B data-center electrification orders in 1H 2026. Eaton’s data center electrical orders jumped ~85% YoY with revenue up ~65% in Q2, and Vertiv delivered Q2 results with revenue up 24% to $3.27B, adjusted EPS up 60% to $1.52, and adjusted free cash flow up to $925M; 2026 revenue guidance midpoint is ~$14B.

Analysis

The market is still treating AI as a semiconductor story, but the tighter bottleneck is shifting to power delivery and thermal density. That favors infrastructure vendors with multi-year visibility, but the best economics likely accrue to the names that control scarce lead times rather than the names with the broadest AI branding. In that frame, GEV has the cleanest scarcity value because behind-the-meter generation is the fastest workaround to interconnect delays, while ETN sits on the critical path for every incremental megawatt regardless of whether the source is grid or onsite generation.

The second-order effect is that the power shortage may widen pricing spreads inside the sector: companies tied to electrical balance-of-plant and cooling should compound faster than general industrials, while pure-play compute suppliers face more capex scrutiny if power becomes the gating item. VRT is the highest-beta expression because its revenue mix benefits from rising rack density and liquid cooling adoption, but it is also the most vulnerable to any pause in data center starts or to a shift toward lower-density architectures. Over 1-3 months, the catalyst is backlog/order commentary; over 6-18 months, the real question is whether customers keep pre-ordering capacity or start deferring projects as returns normalize.

The consensus is probably underestimating how long power constraints can persist, but it may be overpaying for certainty already visible in order books. The main falsifier is not a single weak quarter; it is evidence that hyperscaler capex is being re-phased, that turbine or switchgear lead times are easing, or that utility interconnect approvals start shortening materially. If that happens, the trade shifts from scarcity premium expansion to normal industrial multiple compression.

Key Decisions for Investors

• Long GEV / short a broad industrial basket (e.g., XLI) for 3-6 months: GEV has the most direct exposure to the fastest power workaround, while XLI dilutes the theme and should lag if AI power scarcity persists.
• Long ETN on any 3-5% post-rally pullback; use a 6-12 month horizon. Risk/reward is attractive because the electrical distribution bottleneck is harder to substitute than generation, and order visibility should support a premium multiple.
• Trade VRT tactically, not structurally: buy only on a pullback or after a confirmed backlog beat. It offers the highest upside to rack-density growth, but if data center starts slow even modestly, the stock can de-rate faster than ETN/GEV.
• Pair long ETN vs short a lower-quality electrical contractor or diversified capital-goods name if you want a cleaner relative-value expression of AI capex without taking broad market beta; thesis should be tested by next two earnings prints and order growth.
• Set a falsifier alert: if hyperscaler capex guides down or interconnect lead times shorten meaningfully over the next 1-2 quarters, reduce exposure across GEV/ETN/VRT because the scarcity premium is the main driver, not near-term revenue already booked.