Self-Powered AI: a practical standard so AI growth doesn't raise power bills (www.thepricer.org)

AI growth risks driving big new peak-hour costs and infrastructure spending onto neighbors unless data centers bring firm, hourly-matched local clean power. The piece outlines the problem: as AI demand rises, the grid’s worst hours—think 6–9 p.m.—drive expensive wholesale energy and capacity spikes, costly new substations/lines, and water stress from cooling. Examples: PJM capacity prices jumped from ~$29/MW‑day to ~$329/MW‑day, turning a 100‑MW campus’s annual capacity obligation into roughly $12M; Google’s Oregon campus used 355M+ gallons in 2021. Left unchecked, those costs are recovered through riders and rate-base growth and disproportionately hit low-income households and stressed communities. The solution proposed is an operational, finance-grade standard so AI sites don’t socialize their peak risk. Developers must prove Attributable Additional Clean Supply (AACS), publish Hourly Clean Coverage (HCC) and Energy-weighted HCC (EHCC) plus Clean Matching Shortfall (CMS), demonstrate Firm Self‑Supply Availability (FSSA) for peak windows, and keep Scarcity‑Adjusted Import Exposure (SAIE) ≈ 0 in the worst decile of price/CO₂ hours. Verification requires an hourly ledger, project CODs and contracts, and telemetry showing multi-hour self-supply. The technical implication: move beyond annual RECs to 24/7, locally attributable, dispatchable clean energy and storage paired with smarter siting and tariff design so AI scales without raising neighbors’ bills or emissions.