AI Data Centers Are Competing With Residents for Power — and Utilities Are Choosing Data Centers
Sierra Pacific Power, the utility serving Lake Tahoe, is considering a proposal to redirect existing transmission infrastructure — currently relied on by approximately 50,000 residents — to serve an incoming AI data center, according to reporting confirmed by state filings. The proposal does not involve building new grid capacity. It involves rerouting power that homes and businesses already use.
That detail puts the Tahoe situation in a different category from earlier grid disputes, which typically involved arguments about who should fund new infrastructure. This one is about taking what is already there.
Four States, One Pattern
The Tahoe case is not isolated. The same structural conflict is playing out across at least four other US jurisdictions:
Montana: NorthWestern Energy serves 413,000 customers and is under pressure to extend capacity to an Atlas Power AI data center outside of Butte. Consumer advocates have warned the existing customer base — many of whom absorbed recent rate hikes — are likely to fund the infrastructure expansion through their bills.
San Antonio: CPS Energy is revising its forward capacity plan to accommodate data center load growth that has already exceeded three-year-old forecasts. Wind generation, originally planned as the buffer, has been displaced in favour of additional natural gas and solar — a costlier path that ultimately lands on residential ratepayers.
Washington State: Grant County Public Utility District has filed eminent domain lawsuits against nine landowners to seize portions of their properties for transmission lines. The stated purpose is power delivery to the county’s data center concentration — now one of the densest in the Pacific Northwest.
The Structural Problem
Grid conflicts are a direct consequence of speed mismatch. AI data centers negotiate directly with utilities for dedicated supply contracts, often securing firm capacity commitments that take power off the table for other customers. Residential users have no comparable leverage. They absorb what remains — or, in Tahoe’s case, absorb the infrastructure reduction.
US hyperscalers are on track to spend approximately $830 billion in combined AI infrastructure capex in 2026. That compute requires power. The US grid was not designed to absorb a multi-hundred-gigawatt step change in industrial electrical demand over three to five years. The transmission infrastructure, in many regions, predates modern data center load by decades.
Regulatory Positioning
FERC has an active complaint related to Maryland ratepayers subsidising Virginia data centers — the first federal-level signal that ratepayer cross-subsidy is a live regulatory issue. Several state utility commissions are now considering rules requiring data center developers to fund dedicated grid upgrades rather than drawing on existing residential capacity.
Federal FERC rulemaking is running parallel, at slower pace. Until those rules land, the conflict between AI compute demand and residential power supply will be resolved utility by utility, deal by deal.
The Tahoe case will be the test. If a utility can redirect existing residential transmission infrastructure to a data center without triggering regulatory intervention, similar proposals will follow in every constrained grid region in the country.