A state electricity-price ranking does not measure the AI effect

EIA does not publish a state measure called “AI data center electricity price increase.” A defensible analysis instead joins three distinct measures:

1. Observed electricity-price change by state
2. Evidence of data-center load growth
3. Regulatory evidence showing how resulting costs are assigned

Do not treat the first measure as proof of the other two.

What current state price data show

As of September 30, 2026, EIA’s latest monthly state comparison covers July 2026. It reports preliminary average residential prices in cents per kilowatthour. Nationally, the average increased from 17.45 cents in July 2025 to 18.31 cents in July 2026, a rise of 4.9% (EIA Table 5.6.A).

These were the 12 largest percentage increases among states:

Rank	State	July 2025 (¢/kWh)	July 2026 (¢/kWh)	Change (¢/kWh)	Change (%)
1	Hawaii	39.36	48.00	8.64	22.0%
2	New Hampshire	22.79	26.60	3.81	16.7%
3	Maine	27.98	32.41	4.43	15.8%
4	New York	26.22	29.90	3.68	14.0%
5	Maryland	18.83	21.41	2.58	13.7%
6	North Carolina	13.37	15.16	1.79	13.4%
7	Michigan	20.55	23.05	2.50	12.2%
8	Ohio	17.38	19.45	2.07	11.9%
9	Idaho	12.28	13.73	1.45	11.8%
10	Illinois	17.22	19.22	2.00	11.6%
11	Delaware	16.58	18.48	1.90	11.5%
12	Pennsylvania	19.51	21.72	2.21	11.3%

This ranking describes price movement, not its cause. Hawaii’s position, for example, is not evidence of an AI data-center effect. Fuel costs, generation mix, transmission investment, weather, regulation and other demand growth can affect state averages.

The metric is not an individual household’s tariff or bill. EIA calculates average retail price by dividing utilities’ retail-sales revenue by electricity sold; the result includes generation, transmission, distribution, taxes and fees (EIA’s price definition). The July figures are preliminary estimates for one month, so use annual data when the question concerns a sustained trend.

Data-center demand is real, but “AI” is not a state price category

Berkeley Lab’s 2026 national model estimates that data centers could account for 11.8% of U.S. electricity use in 2030, with scenarios ranging from 9.5% to 15.3%. Its sensitivity scenarios vary assumptions such as specialized graphics-chip installations, AI-chip operating life, and AI-server utilization (Berkeley Lab). These are national scenarios, not state estimates of AI-caused retail-price increases.

EIA expects U.S. electricity sales to grow by almost 2% in 2026 and another 2% in 2027, with data-center development and manufacturing activity driving growth in the commercial and industrial sectors. It identifies the West South Central region as the largest regional contributor, not a state-by-state AI price effect (EIA, September 2026).

Virginia offers stronger evidence of local load growth. Its commercial electricity sales increased by nearly 30 million MWh from 2019 to 2025, growth EIA says was largely driven by concentrated data-center development, with electric vehicles and building electrification also contributing. EIA also reports that PJM expects data-center load to drive substantial peak-demand growth in its Dominion zone (EIA, May 2026).

That supports a demand finding. It does not establish that Virginia’s 11.2% increase in average residential price between July 2025 and July 2026 came from AI or data centers.

Build the spreadsheet around supportable claims

Use one row per state and comparison period:

state
price_period_current
residential_price_cents_kwh_current
price_period_prior
residential_price_cents_kwh_prior
price_change_cents_kwh
price_change_percent
data_center_load_evidence
evidence_geography
evidence_period
cost_allocation_status
attribution_status
price_source_url
load_source_url
regulatory_source_url
notes

Calculate percentage change as:

(current_price - prior_price) / prior_price

Keep the underlying decimal value in the file and apply percentage formatting only for display. Use not established rather than a blank in attribution_status when price data exist but causal evidence does not.

For a stronger longitudinal comparison, replace the July snapshot with EIA’s annual state-and-sector data. EIA provides monthly and annual revenue, sales, customer-count and retail-price resources by state and sector (EIA electricity data). Do not compare a monthly value for one state with an annual average for another.

Add regulatory evidence before making a cost claim

A regional grid cost does not by itself show which retail customers ultimately pay it. For PJM’s proposed Reliability Backstop Procurement, PJM says it would allocate costs among load-serving entities in specified zones or service areas; those entities and state regulators would determine how the costs apply to rate classes. PJM also says it cannot allocate retail costs directly to individual data centers (PJM).

Virginia illustrates why this evidence needs its own column. The State Corporation Commission created a GS-5 class for customers demanding at least 25 MW, effective January 1, 2027. Covered customers must pay at least 85% of contracted distribution and transmission demand and 60% of generation demand, among other requirements intended to help insulate other ratepayers from infrastructure costs. In the same proceeding, the commission approved smaller base-rate increases than Dominion requested, resulting in stated monthly increases for a typical residential customer (Virginia SCC).

A publishable state table should label conclusions in tiers:

• Observed: The state’s average residential price changed by a measured amount.
• Demand-linked: An authoritative source connects local load growth to data centers.
• Cost-linked: A regulator or grid operator documents how large-load costs are allocated among customer classes.
• AI-specific: Evidence isolates AI computing from other data-center activity.

Most states will not reach the final tier. Record that as not established; do not convert absence of evidence into a zero.

Publish the cleaned CSV with source and update-date columns intact so readers can sort states without losing provenance. If the table belongs inside a larger story, compare methods to embed a CSV in a website while retaining a link to the complete public dataset.