How Much More Power Do AI Data Centers Need Right Now? Morgan Stanley Says 38 Gigawatts. Here Are 3 Stocks Cashing In on That Opportunity.

AI data centers face a potential ~38 GW electricity shortfall through 2028, boosting demand for on-site generation solutions. GE Vernova led the list with natural gas turbine orders of $16.7B in the quarter ending June, up 134% YoY, and expects 2026 revenue of about $46B (vs. $38B last year). Bloom Energy posted record $1.06B revenue last quarter (+166% YoY), while NuScale’s SMR pathway is positioned for longer-term growth despite being pre-profit and slower to deliver; its $12.59 12-month price target implies >30% upside.

Analysis

The market is likely underappreciating how much of this theme is a capacity-supply story rather than a generic AI winner story. The earliest monetization accrues to vendors with bankable, shippable equipment and service attach rates, so the scarcity premium should sit with large turbine and electrical balance-sheet names before it migrates to more speculative “new power” architectures. That argues for multiple expansion in the proven OEMs, while the second-order beneficiaries are the grid interconnection, switchgear, transformer, and gas infrastructure supply chains that get pulled forward by onsite generation demand.

The key distinction is revenue quality and time-to-cash. Fuel cells can win because they solve siting and emissions constraints, but their economics depend on installation velocity, project finance, and uptime assumptions; that makes them more sensitive to slippage than the headline growth rate suggests. Small modular reactors are even more option-like: the market is paying today for a 2030+ narrative, which means valuation is vulnerable to any delay in permitting, financing, or utility/customer procurement.

Near term, the risk is that the trade becomes crowded and the names gap up on narrative without enough evidence of margin accretion. Over 1-3 months, watch for backlog conversion, gross margin progression, and whether AI operators are actually signing multi-year onsite power contracts versus issuing LOIs. Over 6-18 months, the main falsifier is a change in power capex economics: if grid upgrades accelerate, gas prices spike, or regulators speed interconnect approvals, the urgency premium for onsite solutions could compress.

Consensus is probably too bullish on the whole basket, but in different ways. GE Vernova looks like the cleanest fundamental expression because it monetizes existing demand with the least execution risk; Bloom has the most asymmetric near-term upside if it proves product margins are durable; NuScale is the most over-optionalized and should trade like a long-duration call, not a core equity story. The best risk/reward is separating “deliverable now” from “promised later.”

Key Decisions for Investors

• Long GEV on 1-3 month horizon; add on pullbacks, target continued backlog/guide revisions. Falsify if order conversion slows or turbine lead times normalize faster than expected.
• Pair trade: long GEV / short SMR into strength. Thesis is near-term cash-flow realization versus long-dated option value; cover the short if SMR wins a real utility contract with financing attached.
• Tactical long BE for 3-6 months if the market is rewarding product-led growth; use tight risk controls because the thesis breaks if gross margin or cash conversion disappoints.
• Avoid chasing SMR as a core position; treat it as a catalyst watchlist name only. Reassess after evidence of bankable project financing, not on design approval alone.
• Watch supplier beneficiaries in the electrical backbone space for a delayed second-order trade; if data center onsite power demand persists, the next leg likely shows up in grid equipment rather than the headline OEMs.