Is solar's growth finally slowing in the US?

US electricity demand rose 2% year over year over the first seven months of 2026, slowing from a 3% increase over the comparable period a year earlier despite data-center growth and broader electrification from EVs and heat pumps. Efficiency gains and data centers generating some power on-site may be limiting the load visible on the grid. The data indicate demand growth remains manageable rather than reaching the extreme levels suggested by some data-center headlines.

Analysis

The key investable distinction is between metered utility load and total data-center power consumption. Behind-the-meter generation shifts economics away from regulated volumetric demand and toward gas turbines, switchgear, backup generation, fuel logistics, and on-site power-management equipment. That is relatively constructive for GEV, CMI, CAT, ETN and Vertiv (VRT), while weakening the argument that every utility with a data-center service territory deserves a demand-driven multiple premium.

For merchant generators, the near-term implication is more nuanced than headline AI-load forecasts suggest. CEG and VST still benefit from scarcity pricing and long-term capacity contracts, but their upside requires contracted load to translate into incremental grid purchases rather than self-supplied generation; otherwise forward power-price and capacity-market assumptions can prove too aggressive. The largest valuation risk is in regions where utilities have already embedded unusually high load-growth forecasts in capital plans, especially if interconnection queues reveal that projects are being served on-site or delayed.

The contrarian read is not that AI power demand is false, but that it may be less investable through broad electricity-demand proxies and more investable through the infrastructure required to make unreliable or constrained grid service usable. On-site generation can also preserve gas demand while bypassing grid load statistics, creating a lag between weak reported utility sales and strong orders for turbines, generators and electrical distribution equipment. Over 6-18 months, permitting, emissions limits, gas-pipeline availability and local reliability rules are the principal constraints that could force these loads back onto the grid and revive the utility/capacity thesis.

Key Decisions for Investors

• Avoid adding to a broad long utility-beta trade solely on AI load-growth narratives over the next 1-3 months; scrutinize AEP, DUK, SO and EXC for earnings-call disclosure of signed load, interconnection deposits, on-site generation and actual delivered MWh rather than announced data-center pipeline.
• Consider a 6-12 month relative-value position: long GEV or CMI versus short VST, sized modestly. The thesis is that self-generation monetizes equipment demand sooner than merchant-grid demand; exit if ERCOT/PJM forward power and capacity prices rise materially alongside verified delivered load, which would favor VST.
• Maintain ETN and VRT as higher-quality picks-and-shovels exposure, but enter on order-related volatility rather than chasing AI headlines. Falsification is a sequential decline in data-center electrical backlog, book-to-bill below 1x, or customer commentary that power constraints are delaying facility fit-outs.
• Set an alert around upcoming utility regulatory filings and capacity-auction results: a material reduction in utility load forecasts or capital plans is a catalyst for multiple compression in AI-premium utilities, while binding reliability mandates that restrict behind-the-meter generation would reverse the relative-value thesis.