Demand for power from AI data centers continues to rise, but the investment case around “AI power trades” is becoming more selective. In 2024, investors focused on independent power producers able to sell into wholesale markets and equipment suppliers involved in data-center infrastructure. Since June, valuation multiples for several companies in these groups have contracted, even as some businesses have raised earnings and guidance.
The pullback in power-related stocks has not eased the underlying shortage. Data-center projects still face constraints on electricity supply, equipment lead times and skilled labor, all of which can delay construction. For investors, the key question is shifting from how much power AI will consume to which companies can convert that demand into durable revenue and cash flow.
Power constraints continue to delay data centers
Electricity remains one of the hardest requirements to add quickly as AI data centers expand. A new facility needs more than generation capacity. It also requires substations, transformers, transmission connections and qualified workers able to build and commission the necessary infrastructure. Existing supply chains cannot respond immediately to every newly announced project.
Delivery times for substation transformers have exceeded 160 weeks. Even after a data-center operator has selected a site and committed capital, equipment delivery and grid connection can determine when the facility actually begins operations. A shortage of skilled electricians is another constraint. Power infrastructure cannot be completed simply by increasing capital spending if the people and equipment needed for construction are unavailable.
Higher BNEF forecasts do not confirm construction
Forecasts for AI-related electricity demand still require a distinction between announced plans and physical construction. BNEF previously raised its demand projections in part on the basis of expansion plans disclosed by companies and data-center operators. Those plans can nevertheless be changed, delayed or canceled, meaning announced capacity does not automatically translate into realized power load.
That distinction affects how power producers, equipment manufacturers and grid companies assess future revenue. If a data-center project is postponed, power purchases, equipment orders and grid-connection schedules may also move back. Evaluating AI-related electricity demand therefore requires more than tracking announced investment or planned capacity. Investors also need to watch whether projects have started, whether equipment is being delivered and whether facilities are connecting to the grid.
Vistra and Talen became early market favorites
In 2024, independent power producers selling electricity into wholesale markets—and particularly those positioned to serve AI data centers—attracted strong investor interest. Vistra and Talen were among the most prominent names in the trade. Their appeal rested on the prospect of rising electricity demand and the possibility of long-term purchasing arrangements with data-center customers.
Companies involved in power supply and infrastructure, including Bloom Energy Corporation, also drew attention. As earnings and guidance improved across parts of the group, investors might have expected valuations to continue expanding. Instead, valuation multiples for some power suppliers and equipment manufacturers have compressed since June. The change suggests that the market is drawing a sharper line between announced orders, construction progress and profits that have actually been realized.
Three business models face different earnings paths
With the power bottleneck still unresolved, there is no single way to benefit from rising AI electricity demand. The clearest routes include regulated utilities, independent power producers selling into wholesale markets and equipment suppliers serving behind-the-meter demand.
Regulated utilities generally generate revenue through grid investment and approved returns on regulated assets. Independent power producers have more direct exposure to wholesale electricity prices, generation-asset utilization and demand from data-center customers. The behind-the-meter market refers to generation or power systems configured by customers on their side of the grid connection. Data centers may turn to these systems to speed deployment or reduce waiting times for grid access.
The three groups differ in how orders are recognized, how regulation affects returns, how prices fluctuate and how long construction takes. Rising AI power demand alone therefore does not imply identical returns for every related stock. The more immediate facts remain whether projects actually start, whether transformers arrive on schedule and whether companies can secure enough qualified workers to complete the work.