AI: Five charts that put data-centre energy use into context (www.carbonbrief.org)

Carbon Brief’s five‑chart synthesis of IEA data reframes the AI‑energy debate: data centres today account for just over 1% of global electricity use and ~0.5% of CO2 emissions, but rapid AI‑driven growth could make them a standout sector where emissions rise rather than fall. Under the IEA’s central scenario data‑centre electricity demand more than doubles to ~945 TWh by 2030 (a ~530 TWh increase), equivalent to current Japan‑level consumption; that rise would drive roughly 8% of projected global electricity demand growth to 2030 (12% in a faster‑growth pathway). AI’s share of that load — currently estimated at 5–15% of data‑centre consumption — could grow to 35–50% by 2030, and the sector is highly geographically concentrated (≈50% of data‑centre power in the US, 25% China, 15% Europe; e.g., Dublin data centres already use ~79% of local power). Technical implications are twofold: grid stress and the carbon mix. Today ~60% of data‑centre power is fossil‑fuel based, 27% renewables and 15% nuclear; the IEA projects a switch to ~60% clean power by 2035 but also expects gas generation for data centres to more than double (120 → 293 TWh). That mix means many regions may need new fossil capacity in the near term even as renewables expand, raising concerns about grid capacity, electricity bills and trade‑offs with electrification goals. Crucially, large uncertainties remain — sparse reporting, efficiency gains from better chips and algorithms, and how many proposed facilities will be built — so outcomes depend heavily on policy, procurement choices and technology trends.