JouleCloudinferenceresearchdocsplaygroundmodelspricingcalculatorstatus
Research · electricity price and emissions

Cheaper power aligned with lower emissions.

Across seven U.S. electricity markets, all 13 market-by-year comparisons showed the same result: the cheapest 10% of hours had lower average emissions than the most expensive 10%. Flexible AI can turn that relationship into lower cost today and lower carbon over time.

See the evidenceMethods and sources
33% lowerTypical emissions gap between the cheapest and most expensive hours
13 of 13Every market-by-year comparison showed the same pattern
113,270Hourly price-and-emissions records compared
7 marketsU.S. wholesale markets included in the analysis
Evidence 01 · 13 market-by-year comparisons

As power gets more expensive, average emissions rise with it.

The same relationship appeared in every market we studied.

Estimated average emissions · g/kWhNext-day wholesale price · $/MWhMiddle half of comparisons
$0$25$50$75$100150g200g250g300g350g400gCHEAPEST 10%MOST EXPENSIVE 10%HOURS, FROM CHEAPEST TO MOST EXPENSIVE →
Median price$15 → $87/MWh
Median average emissions230 → 376 g/kWh
A typical market-by-year comparison showed a 33% emissions gap between the cheapest and most expensive hours. The shaded band shows the middle half of the results.
Evidence 02 · bigger opportunities

Moving demand away from peak hours can cut even more carbon.

Power plants that switch on for peak demand are often less efficient. Moving AI work to another hour can avoid using them.

Coal steam962 g
Gas turbine · peak-demand benchmark582 g
Combined-cycle gas · more efficient399 g
Solar PV0 g
Wind0 g
Nuclear0 g
Full-lifecycle emissions · NREL medians

Manufacturing, construction, operations, and retirement included.

13 Wind43 Solar PV13 Nuclearg CO2e/kWh
Why peak-demand plants matterA peak-demand gas plant emits about 46% more direct CO2 per MWh than efficient combined-cycle gas.

Gas turbines often switch on for peak demand. Moving flexible AI to a different hour can avoid those higher emissions.

Fossil values are derived from EIA 2024 full-load tested heat rates × EIA fuel CO2 factors. The EIA gas-turbine benchmark represents peaker operation. Wind, solar, and nuclear have zero direct fuel-combustion CO2; lifecycle medians add manufacturing, construction, and retirement: NREL 2021.
Why the curves can move together

Low-cost wind, solar, nuclear, and hydro can lower wholesale prices.

Grid operators generally use the lowest-cost available power first while keeping the system reliable. Wind and solar have no fuel cost; nuclear and much hydropower also cost little to keep running. When these sources are plentiful, higher-cost plants run less often and wholesale prices fall.

When clean power is abundant, prices can fall to zero or below. Local grid data shows which data centers can reach that power.

Where this goes next

Price is the starting signal. Grid data makes routing even smarter.

  • Fuel mix shows when low-cost power comes from clean sources.
  • Grid stress shows when moving demand has the greatest value.
  • Regional power flows show where electricity was produced.
  • Local grid limits show which data centers can use it.
  • Emissions data measures the carbon saved by each move.
What this enables

Flexible AI can lower cost and carbon together.

Price-aware routing is live today. The next step is to add grid-emissions data so every request can seek both cheaper power and a smaller carbon footprint.

NREL · demand flexibilityBerkeley Lab · load shiftIEA · value of flexibilityIEA · Renewables 2025
What we learned

A strong signal and a clear path forward.

Established

Cheaper hours were cleaner across every comparison

Across all 13 market-by-year comparisons, the cheapest hours had lower average emissions than the most expensive hours.

Next measurement

Measure the carbon saved by every move

Connect each routing decision to live grid-emissions data so customers can see the impact.

Product direction

Route for cost and carbon

JouleCloud already combines electricity price and GPU availability. Adding carbon data makes every routing decision smarter.

How we measured it

Seven U.S. markets. Two years. 113,270 matched hours.

We matched hourly day-ahead prices with EIA emissions data from July 27, 2024 through July 26, 2026. Within each market and year, we compared the cheapest 10% of hours with the most expensive 10%, weighted every comparison equally, and used periods with at least 95% data coverage.

Study inputs

Grid emissions: EIA Hourly Electric Grid Monitor. Wholesale prices: GridStatus.io datasets for CAISO, ERCOT, ISO-NE, MISO, NYISO, PJM, and SPP.

EIA Hourly Electric Grid MonitorGridStatus.io · hourly price datasets
Interpretive context
FERC · understanding energy marketsBerkeley Lab · Wholesale Electricity PricesSeel et al. · negative wholesale pricesSensfuss et al. · merit-order effectFleschutz et al. · price response and CO2