AI Data Center Power Gap Hits 38 Gigawatts, Industrial Stocks Benefit

Morgan Stanley projects a 38-gigawatt shortfall in U.S. data center power supply by 2028, opening a $2.9 trillion construction pipeline for industrial equipment makers.

AI data centers face a 38-gigawatt power shortfall through 2028, and Morgan Stanley says industrial equipment makers GE Vernova, Eaton, and Vertiv stand to benefit most.

The bank expects data center developers to increasingly build on-site natural gas turbines, fuel cells, and other behind-the-meter generation because grid connections in some regions now take five to seven years, according to the Morgan Stanley research report published Aug. 20.

GE Vernova's gas power equipment backlog reached 116 gigawatts in Q2, up from 100 gigawatts, with management projecting at least 125 gigawatts under contract by year end. Eaton's data center orders jumped 85 percent year over year in Q2, while Vertiv's revenue climbed 24 percent to $3.27 billion with adjusted EPS up 60 percent to $1.52.

Morgan Stanley expects nearly $2.9 trillion in global data center construction through 2028. GE Vernova, Eaton, and Vertiv sell the turbines, transformers, switchgear, and cooling systems that no AI chip can replace — making them the physical backbone of the AI buildout.

GE Vernova's Backlog Jumps to 116 Gigawatts on AI Power Demand

GE Vernova manufactures natural gas turbines that generate electricity for utilities and, increasingly, large data centers. Morgan Stanley identifies natural gas turbines as one of the largest potential solutions to the power shortage, estimating they could provide roughly 15 to 20 gigawatts of capacity through 2028.

The company is already seeing that demand translate into orders. Data center orders in its Electrification business exceeded $5 billion in the first half of 2026, more than double what it booked in all of 2025. The backlog growth from 100 to 116 gigawatts in a single quarter reflects the acceleration in AI-driven power procurement.

The behind-the-meter shift is particularly significant because it bypasses the traditional utility interconnection queue. Data center developers are effectively becoming their own power producers, which changes the economics of AI infrastructure — power generation is no longer a utility cost but a capital expenditure that companies like GE Vernova directly benefit from.

Eaton and Vertiv Capture the Distribution and Cooling Layer

Generating electricity is only half the battle. Eaton manufactures switchgear, circuit breakers, transformers, busways, and backup power systems required inside data centers. Its Electrical Sector data center orders increased approximately 85 percent year over year in Q2, while revenue jumped roughly 65 percent. The Electrical Americas segment's backlog rose 33 percent from a year earlier, providing considerable visibility into future demand.

Eaton also completed its $9.5 billion acquisition of Boyd Thermal in March, adding liquid-cooling technology for power-dense AI data centers. Boyd Thermal is expected to generate roughly $1.7 billion in 2026 sales, including about $1.5 billion from liquid cooling, giving Eaton another revenue stream as AI servers consume more electricity and generate more heat.

Vertiv sells power-management equipment, uninterruptible power supplies, thermal-management systems, and liquid-cooling technology. Its Q2 adjusted free cash flow more than tripled to $925 million, and management now expects roughly $14 billion in 2026 revenue at the midpoint of guidance, with organic sales growth of approximately 31 percent. The Americas and APAC regions both grew 29 percent in Q2, with EMEA returning to positive growth.

Vertiv shares have gained 81.3 percent year to date, trading at 10.05 times trailing sales. The company's 2026 consensus EPS estimate of $6.64 implies 58.1 percent year-over-year growth, according to Zacks data. The company is also collaborating with Nvidia on high-density AI data center architectures, including 800V DC power distribution systems.

None of these stocks is cheap, and there's always the possibility that data center construction slows as hyperscalers become more disciplined with capital spending. But a roughly 40-gigawatt power shortage isn't solved with software — it requires turbines, transformers, switchgear, cooling systems, and billions of dollars in industrial equipment. For investors, the question isn't whether AI will need more power; it's which companies will supply the physical infrastructure to deliver it.

This article is for informational purposes only and does not constitute investment advice.