Elsevier model improves carbon forecasting 26% for cross-border power grids
Researchers at Elsevier have developed a node-aware Graph Neural Network model that improves carbon intensity forecasting in cross-border power grids by over 26%. This modeling approach helps optimize computational demand, potentially reducing the carbon footprint of intensive workloads such as large language model training by up to 90.89%.
Key Takeaways
- Model improves carbon intensity forecasting by 26.46% over single-region models and 20.34% over one-hop neighbor grid models.
- Node-aware embedding mechanism captures physical carbon network rules across 28 European member countries.
- Integrated Long Short-Term Memory (LSTM) submodel tracks temporal dependencies across hourly, daily, and weekly periodic patterns.
- Implementation could reduce the carbon footprint of large language model (LLM) training and inference by 90.89% and 89.03% respectively.
- New CISRNN super-resolution layer allows for higher-resolution carbon intensity output to match diverse downstream application frequencies.
Why It Matters
Accurate carbon forecasting is shifting from a corporate social responsibility metric to a technical necessity as streaming and AI workloads consolidate in massive regional data centers. This model addresses a critical measurement gap where low-carbon countries like Switzerland import high-carbon power from neighbors, a reality that existing localized models often ignore. For the streaming ecosystem, which accounts for up to 80% of internet traffic, integrating this transparency allows CDNs and cloud encoders to shift heavy rendering or model training to periods of verified clean energy supply. Watch for large-scale cloud providers like AWS or Google to integrate similar grid-aware scheduling into their carbon-neutral optimization dashboards by late 2026.
Additional Context
The demand for high-fidelity carbon tracking arrives as the industry faces a surge in environmental scrutiny. Per Allianz Trade (July 2026), global data center CO2 emissions reached 315 million tonnes in 2025, a 57% increase over previous IEA estimates. This rise is largely attributed to AI workloads, which are projected to consume 40% of data center power by 2030. In the United States alone, data center capacity is expected to more than quadruple from 40 gigawatts in 2025 to 169 gigawatts by 2030, according to research published in Energy & Fuels (July 2026). This expansion places intense pressure on power grids that are already struggling with interoperability and clean energy transition targets. In the European Union, regulatory requirements are tightening the window for voluntary reporting. The Carbon Border Adjustment Mechanism (CBAM) enters its definitive regime in 2026, requiring importers to declare embedded emissions with increasing granularity. Simultaneously, the Energy Efficiency Directive now mandates that large data center operators report energy performance and water footprints into a centralized European database as of 2024. Per a report from the European Commission (July 2025), these measures aim to address the projected doubling of data center electricity consumption to 945 TWh by 2030. For streaming providers, these regulations mean that regional offsets are no longer sufficient; they must prove the real-time carbon intensity of the specific grids powering their delivery infrastructure.
Read full article at sciencedirect.com
Get this in your inbox → Subscribe
Enjoy our coverage?
Add StreamingMeme as a preferred source on Google to see more of our streaming news at the top of your Search results.
Add as preferred source