AI-driven data center developers are increasingly turning to on-site power generation and battery storage as electricity grid constraints threaten to delay projects in key US markets, according to a Wood Mackenzie note published Friday.
The consultancy said demand for electricity is surging in northern Virginia, Texas and other major data center hubs as hyperscale operators expand AI infrastructure.
It said access to reliable power, rather than computing capacity, is now the main constraint on AI infrastructure growth.
Lengthy transmission upgrades and slow grid interconnection processes are lagging behind data center development.
This has prompted some developers to consider temporary solutions such as interruptible grid connections or behind-the-meter generation to bring facilities online years earlier than if they waited for permanent utility connections.
Wood Mackenzie said most developers still favor conventional grid connections but are increasingly evaluating bridge solutions despite technical, commercial and regulatory uncertainties.
The consultancy warned that operating AI data centers with dedicated on-site power systems is significantly more complex than simply matching generation capacity with expected demand.
Rapid fluctuations in AI workloads require sophisticated controls to maintain voltage and frequency, while poor system design can damage equipment or trigger outages.
It also said the industry faces a shortage of specialist power system engineers capable of designing and modeling increasingly complex electrical systems.
In some cases, commercial agreements are being reached before technical risks have been fully assessed.
Although relatively few large-scale AI facilities using BTM generation have been completed, Wood Mackenzie said some projects have already experienced major operational failures, including turbine damage and site-wide blackouts linked to system instability or incorrectly configured protection systems.
It said successful BTM projects will require integrated power systems that combine generation, battery storage, advanced inverter controls, and voltage management technologies to cope with the rapid swings in electricity demand typical of AI computing workloads.