LBNL reference case; scenario range 9.5%–15.3% of U.S. use.
Electricity ledger →Source-linked, independently graded research
So — who pays for AI?
Sometimes you. Less often than the headlines say. AI data centers compete for power, chips, equipment, and money — this site grades where that competition is real, where it’s shared with other demand, and where AI is just getting the blame.
Every claim gets a ruling on the first line — true, false, or unproven — plus an attribution grade, the sources and a check date. 4 of 5 are settled outright.
Claim report cards
The claims people actually repeat — ruled on
True“AI is driving the return of U.S. electricity-demand growth.”AI data centers are the largest single driver of new U.S. electricity demand.→Unproven“AI raised my household electricity bill.”No published analysis puts an AI-attributable dollar figure on a household bill.→False“AI caused the transformer shortage.”The transformer shortage began years before the AI buildout.→
Prices in the markets AI buys from are rising about 10% faster than in comparable markets it doesn’t.
Crowding Index
How the score works
The index compares AI-exposed input markets to a matched core-PPI control, baselined at 100 in January 2024.
See the score, basket and calculation rules →Note the outlier: semiconductors broadly are cheaper than in January 2024. The squeeze is specific — memory, storage, and grid equipment — not “chips” in general.
DOE’s 2024 distribution-transformer range; large units can reach 3–4 years.
Equipment ledger →Micron’s reported HBM-to-DDR5 production trade ratio.
Memory ledger →GE Vernova backlog plus slot reservations at Q1 2026.
Turbine ledger →Latest resource grades
Markets with the clearest signal
SemiconductorsAI-driven
Memory (DRAM & HBM)
3:1HBM-to-DDR5 capacity trade ratio reported by Micron
AI memory demand is squeezing the same production capacity used for ordinary DRAM.
See the evidence →SemiconductorsAI-contributing
Storage (NAND & SSDs)
+30.4%BLS storage-device PPI, Jan 2024 to Jun 2026
Storage prices rose 30.4% in 30 months while AI and the server refresh cycle drew on the same NAND capacity.
See the evidence →Grid equipmentAI-contributing
Transformers & Switchgear
1–2+ yrdistribution-transformer lead time in 2024
Data centers are adding demand to a transformer shortage that started earlier.
See the evidence →Generation equipmentAI-contributing
Gas Turbines
100 GWGE Vernova backlog plus slot reservations, Q1 2026
Data centers occupy turbine slots within a broader power-construction boom.
See the evidence →How a grade is made
The same four checks for every grade
- 1Write down the claim
Use the wording people actually repeat.
- 2Trace how it would happen
Identify the market, the other buyers and the supply limit.
- 3Check the best sources
Start with public data, agency work and company filings.
- 4Review it by hand
Tools draft updates. The editor approves every publication.
Public changelog
Every revision gets a date
Grade changes, source swaps, and corrections — including ones that don’t change the conclusion — go into one public record.
Read the changelog →About this project
Pro-growth, and honest about the bill.
David Veksler is a working AI architect who thinks the buildout is worth it — which is exactly why the accounting should be straight. Software flags new data and drafts updates; David reviews every publication. No ads, sponsors, or affiliate links. Affiliations and the capital-markets conflict are disclosed. About & disclosures →