Data-center electricity demand is growing several times faster than global electricity consumption, and AI-focused facilities are growing faster still.
Big Tech's AI buildout has a power problem, and it is no longer confined to data centers. The shortage is moving through the grid, into utility-rate filings, onto the balance sheets of energy-intensive industries, and straight into the cost structures of software companies that built their business models on the assumption that computing was cheap.
For investors, the question is not whether AI is inflating energy costs. That debate is settled. The question is who absorbs those costs, who can pass them on, and which companies sit on either side.
Let's start with the grid itself. A Bloomberg analysis found that wholesale electricity costs near major data-center clusters have risen sharply since 2020, with some areas seeing increases of more than 250%. PJM Interconnection, the grid operator serving much of the U.S. Mid-Atlantic and Midwest, shows how quickly that pressure can move from local demand into marketwide pricing. Its capacity prices rose from under $30 per megawatt-day for 2024-'25 to more than $300 for 2026-'27.
Once new load pushes against available generation and transmission, the cost does not stay neatly attached to the companies creating the demand. It spreads through capacity markets, rate filings and industrial power bills.
The International Energy Agency puts the trajectory plainly: Data-center electricity demand is growing several times faster than global electricity consumption, and AI-focused facilities are growing faster still. The capex cycle behind that demand is enormous. Alphabet $(GOOG)$ $(GOOGL)$, Amazon.com (AMZN), Microsoft $(MSFT)$ and Meta Platforms (META) are on track to spend around $700 billion in capital expenditure in 2026 alone, up roughly 77% from the prior year. That spending is landing on the grid, and the grid is passing it along.
Consumers already are feeling the hit. Utilities filed a record $31 billion in rate-increase requests in 2025, while academic estimates suggest data-center load could push average U.S. electricity bills significantly higher by 2030, with far larger effects in the most constrained markets. U.S. states are beginning to respond. Texas Senate Bill 6 is one example: Large-load customers are now being forced to shoulder more of the infrastructure risk they create.
The most useful framework for investors is not simply who uses energy, but who has the leverage to pass on rising costs and who is forced to absorb them.
The hyperscalers are largely insulated. Amazon, Microsoft, Google and Meta have the scale to negotiate long-term power-purchase agreements directly with generators, build generation capacity behind the meter and lock in rates before the open market reprices. Their capital budgets allow them to absorb, hedge or pre-empt energy-cost inflation in ways smaller companies cannot. The more interesting investment question is what happens to everyone else.
On the physical side, energy-intensive manufacturers are the most exposed group that the market consistently underweights. A steel plant or aluminum smelter competing for electricity in a regional market where data-center demand has collided with tight supply and pushed capacity prices sharply higher does not have many options. It is a price-taker - and the price is going up. The margin squeeze does not resolve until new generation capacity comes online - which, given gas turbine lead times now running up to seven years, is not soon.
On the software side, the dynamic is different but the conclusion is similar. The AI boom has restructured the cost of computing in ways that traditional SaaS economics did not anticipate. Where conventional software companies operate at 80% to 90% gross margins because marginal cost per additional user approaches zero, AI-native companies run structurally lower - around 50% to 60% - because every query incurs real inference costs. Mid-tier software companies building AI features on top of third-party application programming interfaces are caught in the middle: paying per token; passing some of that cost through pricing models that are still being calibrated, and competing against hyperscalers who own the computing power and can always undercut them on price.
The clearest investment case sits in the infrastructure that has no option except to be built. Utilities directly in the path of hyperscaler capital are the most evident plays. Dominion Energy (D) and Entergy $(ETR)$ are both guiding for strong earnings growth through 2029, backed by contracted data-center load.
5 stocks to watch
Grid equipment and power management companies carry more asymmetric upside. GE Vernova (GEV) has effectively sold out its gas turbine production through 2030, with a backlog that hit 100 gigawatts in the first quarter of 2026; management expects the remaining slots will be gone by year's end. Eaton Corp. $(ETN)$ supplies electrical management systems that every data center requires regardless of what happens to any individual AI company. Quanta Services $(PWR)$ builds the grid connections these facilities need, a bottleneck that in some regions now rivals chip supply as a limiting factor. Trane Technologies $(TT)$ is the equivalent play on cooling - a non-discretionary cost that scales directly with computing density.
On the generation side, Bloom Energy $(BE)$ went from clean-energy alternative to critical infrastructure provider after operators discovered that fuel-cell installations could bypass interconnection queues that stretch years in constrained markets - coming online in a fraction of the time. A $5 billion partnership with Brookfield Asset Management in late 2025 validated the position. Finally, NextEra Energy $(NEE)$, with a 33 GW renewable and storage development backlog, is the large-cap version of the same thesis.
Unsettled environment
One caveat worth tracking: The regulatory environment is not settled. Constellation Energy (CEG), for example, sold off sharply in a single session in January when regulators signaled interest in capping electricity prices. The political economy of electricity is not friendly to uncapped pricing when consumers are already seeing higher utility bills. Companies with significant exposure to wholesale market pricing rather than long-term contracts carry regulatory risk that current valuations may not fully price in.
AI is repricing energy across the U.S. economy, but not uniformly. The repricing concentrates where computing is dense, flows outward through grid pricing mechanisms, and lands unevenly depending on whether a company can negotiate around it, pass it through, or is simply exposed to it as a cost it cannot control.
The companies worth owning are those with contracted demand, captive supply chain, or irreplaceable positions in the physical infrastructure that no amount of software optimization can eliminate. The ones to watch carefully are energy-intensive manufacturers in AI-dense grid regions and software companies running inference at scale on third-party computing, with pricing models that have not yet caught up to their actual cost structure.
The grid is the substrate on which the AI economy runs. Its constraints and its pricing are now investment variables, and investors who treat them as such will have an edge over those still reading this purely as a tech story.
Comments