The Price of Intelligence

Abstract

Utility regulators, state and local officials, corporate real-estate and energy teams, and business leaders navigating AI infrastructure decisions.

AI feels digital. Its electricity demand is physical, continuous, and increasingly city-scale. U.S. data centers consumed 176 TWh in 2023 — roughly 4.4% of national electricity — and DOE/LBNL projects that range to reach 325–580 TWh by 2028, up to 12% of national use. Single-site requests of up to 4.5 GW — comparable to the average demand of an entire state — are arriving faster than new generation and transmission can be built. When load precedes supply, wholesale and capacity markets tighten, and costs can shift across the ratepayer base before regulators or communities have a chance to respond.

The policy challenge is not whether AI is economically valuable. It is whether the deal structure lets benefits concentrate while risks spread. This report maps the evidence on cost mechanics, political dynamics, and state and federal regulatory action through July 25, 2026, including the FERC large-load orders, the White House Ratepayer Protection Pledge, and H.R. 9340 — the Ratepayer Protection Act, which advanced from the House Energy and Commerce Committee 52-0 on July 21, 2026.

The Cybernomics view is straightforward: proceed with guardrails. The best policy is neither an unconditional subsidy nor a blanket ban. Require large-load customers to pay the costs they cause, guarantee long-term payments, add supply or flexibility, disclose public obligations, and fund measurable community benefits. The tariff and contract matter more than the press release.

Key findings

• U.S. data center electricity use rose from 58 TWh in 2014 to 176 TWh in 2023 and is projected at 325–580 TWh in 2028 — a wide range that reflects genuine uncertainty about AI workload growth.
• PJM's 2025/26 capacity-market auction attributed $9.33B in impact to data center load, an early sign of how concentrated demand can move regional capacity prices.
• The political center is forming around enforceable cost causation: large loads should pay for the grid assets, capacity reservations, and risks they create — not pass them to households.
• H.R. 9340, the Ratepayer Protection Act, advanced from the House Energy and Commerce Committee 52-0 on July 21, 2026, signaling rare bipartisan consensus on the cost-shifting issue.

What's inside the full report

• Demand data: how data center electricity use tripled since 2014, the 2028 DOE/LBNL forecast range, and why timing — not just scale — is the grid's core challenge.
• Cost mechanics: a scenario test comparing a weak deal (tax abatement + socialized grid build) to a guardrailed deal and a best-in-class structure, with ratepayer outcomes for each.
• Evidence standards: what the research does and does not prove, including conflicting academic estimates on historical retail-price effects and the clearest forward-looking risk.
• Political landscape: what the Trump administration, bipartisan Congress, state executives, and local officials are each saying — and why pro-growth and pro-ratepayer are no longer opposing positions.
• Regulatory tracker: FERC large-load orders, Virginia GS-5, Georgia large-load rules, Oregon HB 3546, Florida protections, the New York moratorium, and the White House pledge.
• Seven guardrails: dedicated large-load rate, take-or-pay contract, real project milestones, additional power, flexible compute, community terms, and transparency with clawbacks.
• A local approval scorecard: the minimum evidence required for each approval test, so decision-makers can verify cost causation, stranded-cost protection, and community benefit before signing.