ANTALYA, November 17, 2026 — A Harvard‑led study reveals that 403 U.S. hyperscale data centres emitted 25.7 million tonnes of CO₂ between May 2024 and April 2025, matching Denmark’s entire annual output. The findings spotlight an AI‑driven surge in electricity demand and a grid under strain.
Methodology and Emissions Snapshot
Researchers at Harvard University modeled electricity use and carbon output for the 403 largest U.S. facilities, relying on publicly available data and peer‑reviewed techniques. Their calculations show the centres consumed 82 terawatt‑hours of power—enough for nearly seven million homes—and released 25.7 million metric tonnes of CO₂, three times the level recorded in 2018, the last peer‑reviewed baseline.
Regional Hotspots and Fossil Fuel Dependence
Virginia, Oregon, Ohio and Iowa alone account for more than half of the electricity studied, underscoring a geographic concentration of AI workloads. More than half of the power feeding these hyperscale sites comes from fossil fuels, with natural gas as the primary source and coal contributing about 15 % of the mix.
Grid Operators’ Playbook for AI‑Heavy Loads
Grid managers are scrambling to balance reliability with the new AI surge. They are deploying a mix of short‑term measures—such as demand‑response programs that temporarily curtail non‑critical loads—and longer‑term strategies, including accelerated procurement of renewable contracts and upgrades to transmission capacity in the affected states. Some operators are experimenting with real‑time carbon‑intensity pricing to incentivise data‑centre operators to shift workloads to greener periods.
Political Backlash and Regulatory Moves
The emissions shock has fueled a political wave. In the November 2026 U.S. midterms, candidates advocating moratoria on new data‑centre construction and tighter environmental rules won significant victories. Meanwhile, at COP31 in Antalya, negotiators launched a voluntary pledge urging companies and nations to power data centres with clean energy and to disclose power, water and emissions data. The United States, notably, skipped the summit.
Industry Investment Trends and Future Outlook
Despite the backlash, hyperscalers are pouring money into AI infrastructure. Combined capital expenditure among Alphabet, Amazon, Meta and Microsoft topped US$250 billion in 2024 and is projected to exceed US$600 billion by 2026, according to Bloomberg. Global data‑centre electricity demand is set to climb from 485 TWh in 2025 to 950 TWh by 2030, highlighting the scale of the challenge.
Local Resistance and Project Controversies
Community opposition is rising. In Goochland County, Virginia, residents rallied against a proposed 1,200‑megawatt Google campus on 1,000 acres, citing transparency, health and infrastructure concerns. The county’s Technology Overlay District allows such projects to proceed “by right,” bypassing separate public hearings, a point that has intensified local dissent.
Clean‑Energy Pledge and Transparency Initiatives
At COP31, the voluntary pledge called for transparent reporting on power, water and emissions for all data‑centre operations. UN climate chief Simon Stiell warned against accepting higher bills and increased pollution today in exchange for future emissions‑saving promises, emphasizing that current carbon spikes cannot be justified by speculative gains.
Next milestone: Hyperscale Data Inc. aims to energise the first 10 MW of its Michigan AI data centre in December 2026, funded partly by proceeds from a recent Bitcoin sale.
Sources
- Data centre industry in India
- Hyperscale vs Edge: The New Rules of AI Infrastructure
- Goochland, Henrico residents pack meeting over potential Google data center
- Hyperscale Data Sells Remaining Bitcoin Holdings for $16 Million to Reduce Debt and Fund AI Data Center
- Nov 17, 2026 | US hyperscale data centres emit as much as Denmark: study
- U.S. data centres now emit more carbon than an entire country | CBC
Photo: Chandapiwa Malema (CC BY-SA 4.0)





