AI is guzzling energy for slop content (www.theguardian.com)

At COP30 in Belém, a coalition of UN bodies, NGOs and the Brazilian government launched the AI Climate Institute to promote “AI for good” in developing countries — training governments and organizations to apply AI for emissions reduction and climate resilience. Proponents say tools ranging from more efficient public‑transit routing and precision agriculture to grid optimization and faster, ML‑assisted weather forecasting can cut waste, better schedule renewable output and improve disaster mapping without every location running heavyweight numerical weather models. A London School of Economics analysis cited at the talks estimates AI could reduce 3.2–5.4 billion tonnes of CO2 over the next decade, even after accounting for AI’s own energy use. Skeptics counter that generative AI and large model deployment are already driving a rapid expansion of power‑ and water‑hungry data centers. Academic work warns US AI growth could add tens of millions of tons of CO2 by 2030, and industry studies suggest the same optimization tech could accelerate oil extraction (potentially unlocking ~1 trillion barrels), negating climate gains. The debate therefore centers on trade‑offs: AI’s capacity to monitor and optimize physical systems versus its substantial compute footprint, governance gaps, and commercial incentives that may favor profit over planetary benefit. The Institute represents a push to tilt deployment toward public‑interest uses, but outcomes will depend on energy sourcing, regulation and who controls the tools.