UBS analysts have concluded that electricity supply can keep pace with AI chip demand through 2030, provided that power projects now in early-stage development are actually completed on schedule.

The assessment addresses one of the central constraints on the AI hardware buildout: the risk that data center expansion will outrun available grid capacity. UBS's analysis frames power supply not as a hard physical limit but as a project-execution question, with the timeline hinging on whether developers bring early-stage generation and transmission capacity online in time.

If those projects are delivered, the bank's modelling indicates sufficient electricity will be available to run the fleet of AI accelerators manufacturers plan to ship over the remainder of the decade. The conclusion ties the semiconductor demand curve directly to the power project pipeline, treating the two as coupled systems rather than separate markets.

The condition attached to the forecast is significant. Early-stage power projects — generation plants and grid infrastructure that have not yet begun construction or reached final investment decisions — carry execution risk, permitting delays and interconnection backlogs. UBS's scenario therefore depends on the utility and development sector converting announced capacity into operating assets before the AI chip installed base grows past the point that available supply can support it.

For chipmakers and data center operators, the analysis offers a measure of reassurance on a constraint that has moved to the top of infrastructure planning agendas. Power availability has become a gating factor for data center siting, with operators competing for grid connections and turning to on-site generation to bridge gaps. A supply outlook that holds through 2030, conditional on project delivery, gives the industry a planning horizon aligned with current roadmaps for AI accelerator production.

For utilities, grid operators and power project developers, the finding signals sustained demand growth tied to AI compute buildout — demand that remains contingent on the semiconductor sector's shipment volumes but that, on UBS's numbers, the power sector can serve if its own pipeline performs.

The 2030 endpoint of the analysis coincides with the period over which AI accelerator shipments are projected to scale substantially, making the alignment between power project completion schedules and chip delivery schedules the operative variable in the scenario. Any slippage in the power pipeline, under UBS's framework, would tighten the constraint before the end of the decade rather than at some indefinite future date.