AI data center research updates and corrections

+Published a revised Crowding Index score of 110.3 (January 2024 = 100), up from the v1 score of 96.8. The v1 score is preserved below and republished in the sensitivity table on /index.

·Demoted the semiconductor PPI from a scored component to a published indicator on exposure-validity grounds. BLS PCU334413334413 surveys domestic producers, but the AI-relevant chips, TSMC-fabricated accelerators and mostly Korean-made HBM and DRAM, are imports outside that survey frame, and the series carries a decades-long quality-adjusted downward drift. Its 82.1 v1 reading measured the instrument, not the market.

·Switched the control basket from the all-commodities PPI to the core PPI, final demand less foods and energy (FRED WPUFD4131). The all-commodities control rose 14.1% since January 2024, with roughly ten points of that arriving in the 2025-26 energy and metals surge, a market AI data-center demand itself plausibly contributes to. A control that contains the treatment subtracts the signal.

+Added two scored components with locked January 2024 baselines: commercial electric power PPI (WPU0542) and industrial electric power PPI (WPU0543), both already graded markets on the site.

·This revision was made after observing the data, motivated by the instrument problems above. All cross-specification variants, including the ones that read less favorably than 110.3, are published on /index alongside the current score.

+Published a computed Crowding Index score of 96.8 (January 2024 = 100), replacing the illustrative 125 preview.

+Scored two components against a matched all-commodities PPI control: storage 114.2 and semiconductors 82.1, combined as an equal-weight geometric mean.

+Added the control series, January 2024 baselines and the composite to the downloadable baseline CSV so the score is independently reproducible.

·Transformers stay an indicator: the current BLS series has no January 2024 observation to rebase against.

+Added /statistics: twenty-eight source-linked figures, each with an attribution grade or Context label and a stated “supports / does not support” scope.

+Added one-click citations that name the primary source, its publication date and the check date.

+Added opposing causal estimates of data centers’ historical residential-rate effect so the disagreement is visible instead of averaged away.

+Added ratepayer protections, local tax and investment boundaries, siting and noise exposure, generator emissions, data-center efficiency and generation-queue realization.

·The original fifteen figures and their attribution grades remain unchanged.

+Added ten resource ledgers and five claim report cards.

+Added June 2026 LBNL electricity outlook: 11.8% reference case for 2030.

+Added June 2026 storage PPI observation: 65.056.

+Added FERC’s June 18 large-load tariff reform orders.