Jon Moshier
Jon Moshier / Notes / Data Center Externalities draft
Note · From the Notebook

Data Center Externalities

The costs of the data center buildout that don't show up on the operator's balance sheet: grid strain, water draw, and electricity bills shifted onto everyone else.

draft Started June 6, 2026 sciencecomplexityai

Data Center Externalities

An externality is a cost borne by someone outside the transaction. Data centers generate several large ones, and the AI buildout has made them concrete enough to show up on household utility bills and in regional water tables. The pattern is consistent: the operator captures the value, a diffuse public absorbs the cost.

Grid Strain and the Speed Mismatch

Global data center electricity demand grew 17% in 2025, and the IEA projects total consumption rising from about 415 TWh in 2024 to roughly 945 TWh by 2030, near Japan’s entire annual electricity use. Accelerated AI servers account for nearly half that increase.

The externality isn’t the demand itself, it’s the mismatch in timescales. A data center can be sited and energized in 2 to 3 years. New transmission and generation takes 5 to 10. The gap gets filled by straining existing infrastructure, deferring retirement of fossil plants, and queuing everyone else behind the new load. Gartner projects power shortages will constrain 40% of AI data centers by 2027. The strain concentrates geographically: the US hosts about 45% of global AI capacity by power draw, and EPRI projects data centers could reach 41% to 59% of Virginia’s electricity by 2030.

Cost-Shifting onto Ratepayers

This is the externality with the clearest dollar figure. Under traditional utility regulation, the cost of building and upgrading transmission and distribution for a large new load is socialized across all ratepayers. So households and small businesses effectively subsidize the grid connections of some of the most capitalized companies on earth. A Harvard Law analysis documented the mechanism in detail.

The numbers landed in 2025: US electricity prices rose an average of 13%, and two in three Americans now report utility bills as a source of financial stress. The political response was fast. Lawmakers in more than 30 states introduced over 300 data-center-related bills in 2026, and in March 2026 seven major operators (Amazon, Google, Meta, Microsoft, OpenAI, Oracle, xAI) signed a Ratepayer Protection Pledge committing to negotiate separate rate structures. A pledge is not a tariff; whether Ratepayer Cost-Shifting actually stops depends on what state regulators codify.

Water

Cooling is the second physical externality. Large facilities can consume up to 5 million gallons per day, the residential use of a town of 10,000 to 50,000 people. US data centers’ indirect water footprint (mostly from the power plants supplying them) was about 211 billion gallons in 2023. A 2026 UN report warned AI could use as much water as 1.3 billion people by 2030.

The externality is location, not just volume. More than 160 new AI data centers were built in the past three years in water-scarce regions of the US. Texas data centers used 49 billion gallons in 2025, projected to hit 399 billion by 2030, enough to draw down Lake Mead by 16 feet in a year. Water that is cheap to the operator can be scarce to the watershed, the textbook setup for a [private link].

Why Efficiency Doesn’t Fix It

The intuitive response is that better chips and cooling will solve this. They won’t, on their own. Each compute operation getting cheaper makes compute more useful, which drives more of it: the Jevons Paradox. Google’s per-query energy fell 33x in a year while its total carbon footprint rose 48% since 2019. Efficiency reduces cost per unit; it does not bound the total unless something else caps demand.

This is why externalities are the right frame rather than “is AI efficient.” Efficiency is a property of one transaction. The externality is what the sum of all those transactions does to a shared system: a grid, a watershed, a rate base. Pricing the externality back to the operator (separate tariffs, water charges, interconnection cost causation) is the only lever that engages the total. See Systems Thinking and Critical Systems Heuristics for the framing of who counts as inside vs outside the system boundary, which is exactly the question an externality forces.

See also

Sources

← All notes Read recent essays →