Climate Impact:
Renewable-Powered Data Centers
If data centers are built to run on the surplus and curtailed output of solar and wind farms, how much of their computing emissions are avoided? On a current-policies (IEA STEPS) grid baseline, this approach cuts the lifecycle emissions intensity of data center computing by about 87%, from 219 to 28 kt CO2e per TWh-eq, or roughly 191 kt CO2e avoided for every TWh-eq of computing. The reduction comes from siting compute behind the inverter at a renewable plant's DC bus-bar so that otherwise-wasted generation runs the workload instead of grid electricity. Across an addressable market of about 1,348 TWh-eq of data center computing projected by 2035, an illustrative 1% capture represents roughly 2,575 kt CO2e (about 2.6 Mt) of avoided emissions per year.
191
kt CO2e / TWh-eq
1348
TWh-eq (2035)
~2,575
kt CO2e at 1% capture
Model Dashboard
Core metrics at a glance. Forecast year 2035 unless noted.
Unit Impact (Avoided)
191
kt CO2e / TWh-eq
87% reduction vs baseline
Baseline Intensity
219
kt CO2e / TWh-eq
Data center computing (STEPS)
Solution Intensity
28
kt CO2e / TWh-eq
Data center computing with renewable energy
Addressable Market (2035)
1348
TWh-eq
Avoided Emissions (1% Capture)
~2,575
kt CO2e (2035)
At 1% market capture
* Avoided emissions shown assume 1% market capture.
Baseline vs. Solution - Lifecycle Intensity
Baseline
Data center computing (STEPS)
219 kt CO2e / TWh-eq
Solution
Data center computing with renewable energy
28 kt CO2e / TWh-eq
191 kt CO2e avoided / TWh-eq
87% reduction in lifecycle emissions intensity
Projecting to Market Scale
The avoided-emissions total is the product of three numbers: the 191 kt CO2e per TWh-eq reduction in computing intensity, the size of the data center computing market, and the share of that market served by renewable-powered compute. The intensity reduction is a property of the technology, so the headline figure scales with the market size and the capture assumption.
Unit Impact
191
kt CO2e / TWh-eq
×
1348
TWh-eq (2035)
×
1%
market capture
=
~2,575
kt CO2e
The addressable market is data center computing measured in TWh-eq of electricity-equivalent, a metric the sector forecasts more reliably than computing-performance units such as FLOPS. The model uses a conservative trajectory in which data center energy use, including AI computing, grows through the late 2020s and levels off after 2030, reaching about 1,348 TWh-eq by 2035.
At an illustrative 1% capture of that market, renewable-powered data centers would avoid roughly 2,575 kt CO2e per year (about 2.6 Mt). Capture is shown as an assumption rather than a projection: the realistic ceiling is set by how much surplus and curtailed renewable generation is available near demand and by how much computing can be scheduled to use it.
Want to explore the full model?
Customize assumptions and download the audit-ready datasheet.
Key Findings
- 1
Renewable-powered computing cuts emissions intensity by about 87%
Running data center workloads on surplus and curtailed solar and wind generation lowers the lifecycle emissions intensity of computing from 219 to 28 kt CO2e per TWh-eq versus a current-policies grid baseline, an avoided 191 kt CO2e per TWh-eq. Because data center electricity demand, including AI computing, is rising quickly, the intensity at which that compute runs is a growing driver of real-world emissions.
- 2
The mechanism turns otherwise-wasted generation into compute
A conversion system installed behind the inverter at a renewable plant's DC bus-bar lets the data center draw directly from excess DC generation without relying on the AC grid. The system detects when generation exceeds demand, from clipping, curtailment, or inverter outages, and shifts that surplus into deferrable workloads such as AI training or simulation. This displaces conventional grid-based computing rather than adding new grid load.
- 3
About 2,575 kt CO2e per year avoided at 1% capture of a 1,348 TWh-eq market
Data center computing is projected to reach about 1,348 TWh-eq by 2035 on a conservative uptake path that levels off after 2030 and already includes AI computing. At an illustrative 1% market capture, the 191 kt CO2e per TWh-eq reduction translates to roughly 2,575 kt CO2e (about 2.6 Mt) of avoided emissions per year. The total scales linearly with how much of that computing is shifted onto surplus renewable generation.
- 4
The benefit depends on siting and workload flexibility
The avoided emissions are not unconditional. The solution still carries a 28 kt CO2e per TWh-eq lifecycle intensity, and the gain depends on co-locating compute with renewable generation that would otherwise be curtailed and on running workloads that can be deferred to match supply. Where surplus generation is scarce or workloads cannot flex, the realized reduction is smaller, and the 1% capture rate is an illustrative assumption rather than a forecast.
Methodology & Data Provenance
This model uses the Koi avoided emissions methodology: the difference in lifecycle GHG intensity between a baseline and a solution, multiplied by the addressable market to estimate total avoidable emissions.
Baseline: Emissions from data center computing are primarily driven by operational energy consumption, mainly electricity for computing power and cooling in data centers. This analysis excludes the material production of hardware (such as GPUs) and end-of-life emissions, as they are assumed to be negligible relative to emissions from energy use. The GHG emissions intensity is presented in functional units of TWh-eq of computing, as data center operations are often measured by their electricity consumption. The emissions intensity is calculated from the global grid intensity in the IEA Reference Technology Scenario (RTS). The STEPS reflects the outcome of current energy-related policies.
Solution: The GHG intensity of data center computing powered by renewable energy. Based on the GHG intensity of conventional data center computing, and using an average GHG intensity of wind and solar PV based electricity generation for the operation phase.
Market: Annual data center computing estimated from forecast demand for data center energy consumption. The market is represented in units of TWh-eq computing rather than a computing performance metric (e.g. FLOPS) as data center operations are typically measured and forecast by their electricity consumption. This estimate represents a conservative scenario of data center uptake, with data center energy consumption leveling out after 2030. Data center electricity includes AI computing.
Data Quality Assessment
Reviewed and confirmed by domain experts with primary-source verification.
Reviewed and confirmed by domain experts with primary-source verification.
Reviewed and confirmed by domain experts with primary-source verification.
Reviewed and confirmed by domain experts with primary-source verification.
Companies working to create this future
References & Resources
- IEA (2024), World Energy Outlook 2024 (free dataset)
- NREL (2012), Life Cycle Greenhouse Gas Emissions from Solar Photovoltaics (LCA Harmonization)
- NREL (2012), Wind LCA Harmonization
- IEA (2024), Electricity 2024, IEA, Paris
- Goldman Sachs (2024), AI is poised to drive 160% increase in data center power demand
- Recharge News (2024), AI data centres could need more power than whole of US by 2050
- Koi Data & Methodology Overview
- Full Model Datasheet (Koi platform)
Published by Rho Impact. Data sourced from the Koi Data Lake. Last updated June 2026.
Open This Model in Koi
Customize assumptions, adjust time horizons, and download the full audit-ready datasheet for this model. Free access available via the CRANE Tier.