Why Electricity Bills Are Rising in 2026: Data Centres vs. Grid Upgrades vs. Weather

Executive Summary

U.S. electricity prices are rising faster than the volume of electricity sold. In July 2026, EIA's all-sector average revenue per kWh was 14.99 cents, up 4.4% from July 2025, while retail sales rose 1.9%. Residential price rose 4.9% against 1.0% sales growth. No single driver explains this. The data reviewed here show three structural pressures plus one that is easy to overlook.

Weather: short-run demand

Cooling degree days from January to August 2026 were 1,195, which is 209 above normal (21%) and 84 above the same period of 2025 (8%). ERCOT set a record hourly peak of 91.1 GW on July 22, and SPP set a record of 57.9 GW on July 27. Weather shifts how much electricity is used and how stressed the system is.

Data centres: structural, concentrated demand

LBNL estimates U.S. data centres used 176 TWh in 2023, or 4.4% of U.S. electricity, up from 76 TWh in 2018. Its 2030 reference case is 649 TWh (11.8%), within a modeled range of 521 to 843 TWh.

Grid upgrades: long-lived costs

Utilities spent $27.7 billion on transmission and $50.9 billion on distribution in 2023. DOE announced $5.25 billion of SPARK projects in September 2026, and 1,312 GW of generation capacity was waiting in interconnection queues at end-2025.

Fuel costs: the fourth driver

LBNL reports that the price of natural gas delivered to electric generators rose 36% in 2025 from the 2024 low, and it assessed higher gas prices as a driver of retail price increases in 21 states. A story limited to data centres, grids and weather would leave this out.

Our reading: the public data support each driver's existence and scale but not a split of the price increase among them. This article is explicit about which figures are reported, forecast, modeled or calculated.

Price Growth vs. Consumption Growth

EIA does not collect retail prices directly. It calculates average revenue per kWh from sales revenue and volume and uses it as a proxy for rates, with retail sales as a proxy for consumption. Comparing the two for July 2026 separates a price effect from a volume effect.

Source: U.S. EIA, Electricity Monthly Update (July 2026 data, released September 24, 2026). The last column is a NextMSC calculation. Estimates are preliminary.

Price growth exceeded volume growth in the residential, industrial and all-sector figures. Commercial is the exception, with sales up 4.3% against a 3.4% price rise. EIA's sector data cannot isolate data centres, so the commercial result is consistent with load growth but does not prove it.

The national average hides wide local differences

EIA reports that 44 states and the District of Columbia saw higher average revenue per kWh than last July, and six states saw declines. Hawaii rose 25.4%, Ohio 15.2% and Maryland 14.8%, while Connecticut fell 12.4%, New Mexico 8.3% and Louisiana 6.1%. Among contiguous states, California (30.71¢), Massachusetts (26.56¢) and Rhode Island (25.65¢) had the highest average revenues and North Dakota (9.09¢), New Mexico (9.52¢) and Louisiana (9.74¢) the lowest. California's figure is 3.4 times North Dakota's (NextMSC calculation). Drivers that are local, such as clustered data centres, state rate decisions and regional fuel prices, therefore matter as much as national ones.

Implied Revenue and Sector Mix

Multiplying EIA's July 2026 average revenue per kWh by retail sales gives an implied monthly revenue for each sector. This is NextMSC arithmetic on EIA's rounded figures, and it measures what customers paid in total, not an individual bill. Combining the price and volume changes gives an implied revenue growth rate.

Source: U.S. EIA, Electricity Monthly Update (July 2026). All columns are NextMSC calculations: revenue = price x sales; growth = (1 + price change) x (1 + sales change) − 1; price share = price change / implied revenue growth.

Residential customers account for 40.9% of kWh sold but about half (49.9%) of implied revenue, because their price per kWh is the highest of the large sectors. Residential revenue growth of 5.9% is mostly a price story: price explains about 82% of it. Commercial revenue grew fastest of the large sectors at 7.8%, and there volume (+4.3%) contributed more than price (+3.4%). Industrial revenue rose only because price rose, since volume fell 0.3%.

A hypothetical illustration

EIA's residential average revenue of 18.31 cents implies a July 2025 level of about 17.45 cents (18.31 divided by 1.049). For a hypothetical household using 1,000 kWh in a month, that is $183.10 against about $174.55, or roughly $8.55 more. The 1,000 kWh figure is an assumption for illustration, and EIA does not publish household usage in the data used here.

Reading the sector mix

The sector split matters for the three drivers. Weather mainly moves residential and commercial cooling load. Data-centre load would appear in whichever customer class a utility assigns it to, which EIA's sector data do not reveal. Grid costs are spread across all customer classes through rate design, so a rise in the sector with the highest implied revenue share, residential, has the largest effect on total collections.

Longer-Run Price Trend and Fuel Costs

LBNL's 2026 price-trends edition puts July 2026 in context. National-average nominal retail prices rose 5.3% in 2025 (2.6% after inflation), 29% from 2019 to 2025 and 39% from 2010 to 2025. The 4.4% increase in July 2026 is in line with that pace.

Source: Lawrence Berkeley National Laboratory, Retail Electricity Price Trends and Drivers, 2026 Edition (April 2026). EIA reports 2025 annual average prices of 17.30¢/kWh residential and 13.63¢ all-sector.

Residential prices rose the most, at 33%. The gap between the 5.3% nominal and 2.6% real increase in 2025 means roughly half of that year's rise reflects general inflation (NextMSC reading). LBNL also reports that aggregate residential electricity costs were 1.25% of personal expenditure in 2025. That is an economy-wide ratio and not a typical household bill.

Fuel costs belong in the story

LBNL reports that natural gas delivered to electric generators rose 36% in 2025 from the 2024 low, and it assessed higher gas prices and related wholesale prices as a driver of year-over-year retail price increases in 21 states. This is a separate channel from data centres, grids and weather, and the three-way framing of this article should be read alongside it.

Prices have accelerated since 2019

LBNL's two cumulative figures can be combined. Nominal prices rose 39% from 2010 to 2025 and 29% from 2019 to 2025. Dividing the two implies that prices rose only about 7.8% from 2010 to 2019 (1.39 / 1.29 − 1), or about 0.8% a year. From 2019 to 2025 the annualized rate was about 4.3% a year. The 2019 to 2025 pace is therefore roughly five times the earlier one (NextMSC calculation on LBNL's reported figures).

By sector, the annualized 2019 to 2025 rates are about 4.9% for residential, 3.9% for commercial and 4.1% for industrial (NextMSC calculation). The 2025 rise of 5.3% nominal sits above the 2019 to 2025 average of 4.3%, and July 2026's 4.4% is close to that average. The evidence suggests the current pace is not a one-year spike but a break from the 2010s, though the sources reviewed here do not decompose the break by cause.

What this means for the three-driver framing

Data-centre load was 1.9% of U.S. electricity in 2018 and 4.4% in 2023, so it grew during the same period in which prices accelerated. That timing is consistent with, but does not prove, a link. Fuel costs, inflation, grid spending and weather all changed over those years as well. The honest reading is that several pressures coincided, and the public data do not isolate which mattered most in any given year.

Data Centres: Current Load and 2030 Scenarios

LBNL's data-centre estimates are the best public measure of the sector's load. The 2023 and earlier figures are reported estimates. The 2030 figures are modeled scenarios and should not be read as observations.

Consumption rose from 76 TWh (1.9% of U.S. electricity) in 2018 to 176 TWh (4.4%) in 2023. That is an increase of 132% and an annual growth rate of about 18.3% (NextMSC calculation). LBNL's 2025 update projects 649 TWh (11.8%) in 2030 in its reference case, which implies 20.5% annual growth from 2023. Sensitivity cases are 578 TWh with lower installations and 664 TWh with more specialized graphics chips. The compounded uncertainty range is 521 to 843 TWh, or 9.5% to 15.3% of U.S. electricity.

EIA's September 2026 outlook forecasts retail sales of 4,135 billion kWh in 2026 and 4,211 billion kWh in 2027, an increase of 1.8% (NextMSC calculation), and cites data-centre development and manufacturing as demand drivers.

Requested load is not the same as actual load

NERC reports that ERCOT's observed average peak consumption at data-centre sites that entered service in 2022 to 2024 was 49.8% of requested megawatts. NERC applied that factor to non-crypto data-centre additions in its forecast. Forecasts built on interconnection requests can therefore overstate load, which is a reason to treat the 2030 scenarios as ranges.

What each scenario implies

The scenario table converts LBNL's 2030 cases into the growth they imply from the 2023 base of 176 TWh.

Source: Lawrence Berkeley National Laboratory, 2025 Update (June 2026). Increases and annual growth rates are NextMSC calculations from LBNL values; shares are shown only where LBNL reports them. All 2030 values are modeled.

Even the lower bound implies data-centre use rising by about 345 TWh, nearly tripling from 2023. In the reference case the average addition is about 67.6 TWh a year. For comparison, EIA's September 2026 outlook has total U.S. retail sales rising by 76 billion kWh (76 TWh) between 2026 and 2027. The two figures are not strictly comparable, because LBNL measures data-centre consumption and EIA measures retail sales, but the comparison shows how large the reference case would be against national growth. If the reference case holds, data centres alone would add roughly as much each year as EIA expects the whole country to add in 2027, which implies that either EIA assumes slower data-centre growth or other loads are expected to shrink.

Why this matters for bills

Load growth can lower or raise per-unit costs depending on whether new customers pay for the infrastructure they require. Higher load also forces earlier spending on generation and wires. The public data used here show the load but not who pays for it, which is why this article treats data centres as a structural pressure on planning and spending and not as a measured share of the 2026 price increase.

Grid Upgrade Spending and Interconnection

Utilities spend heavily on the wires that deliver power. Major utilities spent $320 billion in 2023 (in real 2023 dollars) to produce and deliver electricity, up 12% from $287 billion in 2003, according to EIA and FERC data.

Sources: U.S. EIA and FERC (utility spending, November 2024 analysis); U.S. Department of Energy (SPARK, September 24, 2026). The total is a NextMSC calculation.

The four categories shown total $42.8 billion, or 84% of the $50.9 billion distribution total (NextMSC calculation). Line transformer investment was up 23% from 2022. Distribution-level energy storage spending reached $723 million, up from $97 million in 2022. EIA notes that substation equipment helps systems withstand extreme weather, manage renewable intermittency and improve voltage control, which links grid spending to the weather driver.

DOE SPARK

DOE's September 2026 announcement covers 31 projects across 26 states with $5.25 billion in total value, of which about $1.9 billion is federal (36% of the total, NextMSC calculation). The projects would reconductor or rebuild more than 1,500 miles of lines, deploy grid-enhancing technology on nearly 21,000 miles and make more than 23 GW of additional capacity available.

Interconnection queues

Source: Lawrence Berkeley National Laboratory / GridTracker, Queued Up: 2026 Edition. The queues cover generation and storage and exclude load interconnection requests.

Utility capital spending generally reaches customers through rates approved by state regulators, which EIA identifies as the primary basis for retail charges, and it does so with a lag. The public data used here do not show how much of 2023 spending is already in 2026 rates. The queue figures show that capacity is waiting for grid access, but they do not measure how much of that delay affects customer prices.

The pace of spending

EIA and FERC report that transmission spending rose $2.7 billion (11%) in 2023 and distribution spending rose $6.5 billion. Backing those increases out gives about $25.0 billion and $44.4 billion in 2022, so distribution grew about 14.6% and combined transmission and distribution spending about 13.3% in a single year (NextMSC calculations). Distribution is 64.8% of the combined $78.6 billion, and the combined figure equals about 24.6% of the $320 billion that major utilities spend on producing and delivering electricity. Line transformer spending of $7.5 billion was up 23%, which implies about $6.1 billion in 2022, and distribution-level energy storage spending rose roughly 7.5 times from $97 million to $723 million.

These figures describe capital spending in 2023. They do not say how much has reached customer rates by 2026, which depends on state regulatory decisions. They do show that the fastest-growing categories include equipment tied to reliability and to new technology, not only to new load.

Reading the queue figures

The interconnection queue holds 1,312 GW of generation and 749 GW of storage, 2,061 GW combined (NextMSC sum). Of that, 549 GW has a draft or executed interconnection agreement but is not yet in commercial operation, about 26.6% of the combined total (NextMSC calculation). Historically, only 13% of capacity that entered queues between 2000 and 2020 reached commercial operation. The queue is therefore a measure of intent and congestion, and it should not be read as capacity that will arrive. Because it excludes load interconnection requests, it also understates the demand side of the same grid constraints that data-centre developers face.

How Cost Increases Reach Customer Bills

The path from a cost to a customer's bill is set by regulation and market structure, and the sources reviewed here describe it only in outline. EIA states that charges for retail electric service are based primarily on rates approved by state regulators, and that some states allow competitive retail suppliers to offer market-based prices. EIA also notes that it does not collect retail rates directly and instead calculates average revenue per kWh from revenue and sales.

Three channels

The practical implication is that the same national evidence can produce different local outcomes. A state that approved a rate case in 2025, a state with competitive suppliers and a state with a large data-centre cluster could each show different price paths, which is consistent with EIA's July 2026 state results, ranging from a 25.4% increase in Hawaii to a 12.4% decrease in Connecticut. The sources reviewed do not contain the state rate-case information needed to test these explanations.

Weather and Cooling Demand

Weather's effect is immediate but mostly a volume effect. NOAA's population-weighted cooling degree days for January to August 2026 show how far above normal the year has run.

Source: NOAA Climate Prediction Center, population-weighted cooling degree days, August 2026 (released September 2, 2026). Implied normal = 2026 value minus deviation; percentages are NextMSC calculations.

Virginia ran 34% above its implied normal, the largest of the four areas, while California ran 18% above. Nationally, the contiguous U.S. average temperature in 2025 was 54.6°F, 2.6°F above the 20th-century average, and the fourth-warmest year in the 131-year record, according to NOAA.

EIA's July report adds a caution. Twenty-seven states and the District of Columbia had fewer cooling degree days in July 2026 than in July 2025, and 23 states had more. The July price rise therefore cannot be pinned on a uniformly hotter July. Weather can influence average revenue indirectly, but the data here do not quantify that effect. It more plainly explains the volume side: record peaks in ERCOT (91.1 GW, about 6% above the prior 85.5 GW record) and SPP (57.9 GW).

Weather diverged sharply by state in July

EIA's July 2026 data show the weather picture was uneven. Alaska's cooling degree days fell 80% from July 2025, Vermont's 39% and New Hampshire's 36%. North Dakota's rose 75%, Montana's 67% and Wyoming's 51%. Retail sales volumes also diverged: 32 states saw an increase, led by Nebraska (+12.8%), Nevada (+10.6%) and Utah (+10.4%), while 18 states and the District of Columbia saw decreases, led by Rhode Island (−12.9%), Delaware (−9.2%) and New Jersey (−6.9%). The sources do not link these state sales changes to weather or to data centres, so they are best read as evidence that national averages conceal large local movements.

Peaks are a planning driver as much as a cost driver

ERCOT's record of 91.1 GW is about 3.8% below NERC's 2026 forecast for ERCOT summer peak total internal demand of 94.65 GW (NextMSC calculation). The two figures measure different things, an observed hourly peak load and a forecast of total internal demand, so they should not be compared as a forecast error. The point is that observed records are now close to the forecasts used for long-term planning, which is why peaks feed into grid investment decisions that reach bills years later.

PJM and ERCOT: Where Load Forecasts Concentrate

Data-centre growth is concentrated, and so are the load forecasts built on it. NERC's 2025 Long-Term Reliability Assessment shows two regions where the forecasts are large.

Source: NERC, 2025 Long-Term Reliability Assessment (January 28, 2026). Forecasts, not observations. Percentages marked NextMSC are our calculations.

In ERCOT, the 25 GW of data-centre load in the 2035 forecast equals about 42% of the forecast rise in summer peak demand (NextMSC calculation; load and peak contributions are not identical, so treat this as a rough scale). Combined with the 49.8% observed-to-requested ratio, it shows why ERCOT forecasts depend heavily on assumptions about which projects are actually built. For customers, the near-term effect of these forecasts is planning and infrastructure spending, which is the grid driver, and not a direct line item on a bill.

Original Driver Comparison

The matrix below sets the drivers side by side on how they reach bills, how fast, and how strong the verified evidence is. The mechanism and channel columns are NextMSC interpretation; the evidence column uses only reported, forecast or modeled figures from the source dataset.

Sources: EIA, LBNL, NERC, DOE, NOAA and NextMSC analysis. Evidence type follows the dataset classification.

The matrix shows that the drivers operate on different clocks. Weather acts within weeks and mainly through volume. Data centres and grid spending build over years and concentrate in a few regions. Fuel costs can move retail prices within a year. A bill reflects all of these at once, which is why attributing a single percentage to one driver would overreach the public evidence.

Implications for Stakeholders

The table links the evidence to the groups most directly affected. The implication column is NextMSC interpretation of the verified data and not a forecast.

Sources: U.S. EIA, LBNL, NERC, U.S. DOE and NextMSC analysis.

What the Data Can and Cannot Show

The following limits shape how this analysis should be used.

• No national attribution. The public datasets give price and volume by sector but no line-item split of the 4.9% residential price rise.
• Spending is not rates. Capital spending reaches bills through regulatory decisions over time, and the sources here do not show the amount currently in rates.
• Forecasts are not observations. NERC and LBNL projections are labeled forecast or modeled throughout, and the observed 49.8% ratio shows why requests can overstate load.
• Queues exclude load. The 1,312 GW and 749 GW figures cover generation and storage only.
• Proxies and preliminary data. EIA average revenue per kWh is a proxy for price, and July 2026 estimates are preliminary.

Dataset Explorer

Filter the underlying indicators, forecasts and modeled scenarios compiled for this analysis. Every record is tagged Reported, Forecast, Modeled or Calculated.

Sources: U.S. EIA, LBNL, NERC, U.S. DOE, NOAA and NextMSC analysis. Source documents and dates are shown per record. The one Calculated record is a NextMSC compound annual growth rate.

Methodology and Data Notes

This article uses Tier A sources: federal agencies, national laboratories and the North American Electric Reliability Corporation. Each record in the dataset is tagged Reported (official data), Forecast (official projection), Modeled (scenario) or Calculated (NextMSC arithmetic). Data cut-off: September 24, 2026, the release date of EIA's Electricity Monthly Update for July 2026. Planned refresh: monthly, after each EIA Electricity Monthly Update (next release October 23, 2026).

Calculations by NextMSC

These are arithmetic on cited figures: price minus sales growth (percentage points); transmission plus distribution spending ($27.7B + $50.9B = $78.6B); data-centre growth (176 / 76 − 1 = 132%; (176 / 76)^(1/5) − 1 = 18.3%); the 2023–2030 growth rate of 20.5%; implied normal cooling degree days (2026 value minus deviation); PJM and ERCOT growth percentages; the California-to-North Dakota price ratio; and the ERCOT data-centre share of forecast peak growth.

Limits

EIA average revenue per kWh is a proxy for price and July 2026 values are preliminary. The 1.25% expenditure share is economy-wide. LBNL 2030 figures are scenarios. NERC regional forecasts are U.S. assessment-area projections, and the interconnection queue figures exclude load. Bills by state or household, rate-case outcomes and wholesale-price series are outside the dataset.

Dataset composition

Source: NextMSC macro indicator dataset (see Dataset Explorer). Counts are exact.

Key terms

Average revenue per kWh is EIA's proxy for retail price: sales revenue divided by kWh sold. Retail sales are the volume of electricity sold to end users and serve as a proxy for consumption. Cooling degree days (CDD) measure how much warmer than a base temperature each day is and indicate cooling demand. Nominal values are not adjusted for inflation and real values are. TWh is a terawatt-hour, or one billion kWh, and GW is a gigawatt, or one million kW. CAGR is compound annual growth rate.

Source tiers

Every figure comes from a federal agency, a DOE national laboratory or NERC. Forecast and modeled values are labeled wherever they appear, and calculated values are marked as NextMSC arithmetic. No private research firm or news report is used as a source for any figure.

Frequently Asked Questions

Are data centres the main reason electricity bills are rising?

The public data do not support that conclusion. Data centres used 4.4% of U.S. electricity in 2023 and are concentrated in a few regions. No source reviewed here splits the national price increase by driver.

Why is price rising faster than usage?

In July 2026, all-sector average revenue per kWh rose 4.4% while sales rose 1.9%, and residential price rose 4.9% against 1.0%. The gap means higher per-unit revenue, not just higher consumption, accounts for much of the increase in that measure.

How much did weather matter?

Cooling degree days were 21% above normal from January to August 2026 and ERCOT and SPP set peak records, so weather clearly raised usage. Its effect on the per-kWh price is not quantified in the data.

Do grid upgrades show up in bills immediately?

Generally not. Utility spending reaches customers through rates approved by state regulators, so there is a lag, and the sources here do not show how much 2023 spending is already in 2026 rates.

Is the electricity price the same as my bill?

No. A bill combines the rate per kWh, how many kWh were used, and any fixed charges. EIA's average revenue per kWh is a proxy for price and does not include household usage, so it tracks the rate side of a bill.

Which states had the biggest price changes in July 2026?

EIA reports increases of 25.4% in Hawaii, 15.2% in Ohio and 14.8% in Maryland, and decreases of 12.4% in Connecticut, 8.3% in New Mexico and 6.1% in Louisiana. Among contiguous states, California had the highest average revenue at 30.71 cents per kWh and North Dakota the lowest at 9.09 cents.

How fast could data-centre demand grow?

LBNL's modeled 2030 range is 521 to 843 TWh against 176 TWh in 2023, which implies 16.8% to 25.1% annual growth. Its reference case is 649 TWh, or 20.5% a year. These are scenarios and depend on how many planned facilities are built.

What is DOE SPARK?

It is a DOE initiative announced on September 24, 2026 covering 31 projects in 26 states with $5.25 billion in total value, of which about $1.9 billion is federal. The projects focus on reconductoring and grid-enhancing technology to add more than 23 GW of capacity.

Why are prices rising faster now than in the 2010s?

LBNL's figures imply annualized growth of about 0.8% from 2010 to 2019 and 4.3% from 2019 to 2025. The sources document the change but do not decompose it by cause, although fuel costs, inflation, grid spending and weather all changed during the period.

Outlook and What to Monitor

The evidence points to a layered picture. Prices are rising faster than volumes, the rise follows a multi-year trend that includes fuel-cost increases, and the structural pressures from data centres and grid investment are large and concentrated. Weather amplifies usage in the short run. The strongest forward-looking uncertainty is the data-centre scenario range of 521 to 843 TWh in 2030, which will depend on how many requested projects are actually built.

What to monitor

Three questions for the next twelve months

• Does the price-volume gap persist? EIA's monthly data will show whether the July 2026 gap of 2.5 points for all sectors is a one-month effect or a trend.
• How much data-centre load arrives? LBNL's range is wide, and ERCOT's observed-to-requested ratio of 49.8% suggests that actual load tends to come in below requests.
• How quickly does capital spending reach rates? The $78.6 billion spent in 2023 is the latest full-year figure in the dataset, and the gap between spending and rates is the main unmeasured link in the chain.