Electricity-price driver decomposition — load growth is not the national driver

One-line summary: CGEP research (drawing on Lawrence Berkeley National Lab + Brattle Group data) finds US retail electricity price increases are driven first by natural-gas/fuel prices and second by transmission & distribution capex (aging assets, storm/wildfire recovery, equipment inflation) — NOT by load growth, which in supply-rich states has actually lowered prices. The affordability story is regional, not national.

The insight

The prevailing narrative blames AI data centers for rising electricity bills. The CGEP decomposition says the causality is mostly elsewhere:

1. #1 driver: fuel prices. Gas turbines set the marginal price in most US markets, so gas-price swings pass through to all retail electricity.
2. #2 driver: T&D capex — aged-infrastructure replacement, hurricane/wildfire recovery, line burying, and equipment inflation (see aging-grid-replacement-to-td-capex-supercycle).
3. Load growth cuts both ways. Where excess generation and transmission exist (Nebraska, North Dakota, New Mexico), new load lowered prices by spreading fixed costs; where supply is constrained (Virginia/PJM), anticipated load is inflationary because marginal cost exceeds average cost.
4. The problem is regional. 23 states saw real price declines 2019–2025; 27 saw increases, with different drivers (wildfire in CA, gas volatility in the Northeast, capacity-market dynamics in PJM).

Evidence

• doug-arent in 2026-06-30-podcast-columbia-energy-exchange-doug-arent-and-robin-millican-on-what-s-really: "what really is driving prices dominantly, it's fuel price... when natural gas prices fluctuate, they set the price for all electricity which is sold and those price increases pass through to customers. So that's the number one driver. The second driver is in fact increased distribution costs, transmission costs."
• doug-arent in 2026-06-30-podcast-columbia-energy-exchange-doug-arent-and-robin-millican-on-what-s-really: "It's surprising that load growth does not stand out as a national driver of price increases... There are a few states where it actually led to price decreases." (Nebraska, North Dakota, New Mexico — "adding more load growth to a fixed capital base spreads those costs over larger sales.")
• robin-millican in 2026-06-30-podcast-columbia-energy-exchange-doug-arent-and-robin-millican-on-what-s-really: "if you look at the period from 2019 to 2025, you had 23 states that saw price declines in real terms... while 27 states saw [increases]... the drivers differ by region, where you have wildfire costs... California... Fuel volatility in the Northeast... and capacity market dynamics in the PJM interconnection."
• robin-millican in 2026-06-30-podcast-columbia-energy-exchange-doug-arent-and-robin-millican-on-what-s-really: "demand in and of itself is not a problem if supply can efficiently rise to meet demand. And demand is actually sometimes an asset... if you can spread fixed costs over a larger customer base."
• doug-arent in 2026-06-30-podcast-columbia-energy-exchange-doug-arent-and-robin-millican-on-what-s-really: "in 43 states, consumers saw electricity prices increase just last year [2024→2025]... residential prices rise faster than consumer and industrial prices." (The political moment is real even though the causal attribution to data centers is weak.)
• alice-yake in 2026-07-07-podcast-columbia-energy-exchange-alice-yake-on-planning-for-a-reliable-cleaner-grid (independent second source — ex-Xcel chief planning officer — on the age-not-load driver): "the average age of that infrastructure across the US is around 55 years... we're looking to replace a significant portion of the infrastructure... even without data centers, even without load growth... we have a lot of investment that has to happen... As we do that energy prices are going to increase. I've been telling people for a year now... you should not be expecting your energy costs to go down, the question really becomes what are they going up to?"
• alice-yake in same (the depreciation front-loading mechanic — why the replacement wave resets bills upward): "When you are early in the life cycle and you have a brand new asset, the cost of that asset is at the highest. It's not depreciated yet. So your bill is more significantly impacted... But at the asset it's end of life, the opposite is true. You've largely depreciated that asset... right now... average age of 55 years... We have a low cost old system." (Straight-line depreciation means the current cheap "old system" flips to the high end of the cost curve as it is replaced — a bill-driver mechanically independent of load growth.)
• alice-yake in same (the data-center load-factor cost channel, beyond raw MW): "Load factor for data centers is much higher than a typical customer... you increase the overall system load factor... you have to call on assets more frequently. When you run a generator more, you typically have higher impact on the maintenance of that asset as well as the shortened life. And so there are cost impacts associated with that."
• alice-yake in same (why new large loads no longer just spread fixed cost — the system is full): "Our system is largely full across the US. So it's not like you can add them... and they pay for an existing system and just make it cheaper for everyone. We have to add a lot of infrastructure." (Sharpens Millican's "demand is an asset if supply can rise" caveat — the spread-fixed-cost benefit assumes headroom that no longer exists in much of the US.)

Trading implications

• De-risks the T&D picks-and-shovels leg from AI-demand risk: if the #2 bill driver is aging-asset replacement and hardening (not data centers), the grid-equipment capex wave survives even if the AI buildout disappoints — see aging-grid-replacement-to-td-capex-supercycle.
• Regionalizes the utility trade: PJM supply-constraint theses (pjm-capacity-prices-to-nuclear-premium) are confirmed as a special case, not generalized nationally. Supply-rich regions (ERCOT, MISO-west states) can absorb data-center load without a price/political backlash — favoring builders who site there.
• Political-backlash risk is a rates/regulatory risk for DC-adjacent utilities: 43 states saw increases and residential rates rise fastest; the cross-subsidization debate (large-load tariffs) is where that pressure lands.

Contradictions / tensions

• The wiki's AI-power chains (e.g., ai-capex-to-power-and-materials-cascade, pjm-capacity-prices-to-nuclear-premium) sometimes read as "data centers drive power prices nationally." This decomposition refines, not falsifies, those chains: data-center load is a genuine forcing function in supply-constrained regions (PJM, Virginia) and for forward capacity prices, but it is not the dominant driver of realized national retail bills. Both are recorded; not silently reconciled.
• All-In (2026-05-08 ingest, on the cascade page) claimed "electricity costs are going down in Texas where data centers are built, up in NY/CA which build nothing" — directionally consistent with the supply-availability framing here.

Sources

• 2026-06-30-podcast-columbia-energy-exchange-doug-arent-and-robin-millican-on-what-s-really
• 2026-07-07-podcast-columbia-energy-exchange-alice-yake-on-planning-for-a-reliable-cleaner-grid — independent CGEP-adjacent second source (Alice Yake): age-not-load driver, depreciation front-loading, load-factor cost channel.

Related

• aging-grid-replacement-to-td-capex-supercycle
• phantom-data-center-load
• grid-enhancing-technologies-latent-capacity
• pjm-capacity-prices-to-nuclear-premium
• ai-capex-to-power-and-materials-cascade
• doug-arent
• robin-millican
• alice-yake