AI for Planet: Communities vs Corporations for Greener Futures

From community pushback against data centres to debates over AI’s environmental impact, to the real-world consequences of tech growth – this week’s newsletter looks at the urgent need for transparency, equity, and smarter policy. Here’s the 5 things that matter most today:

1. Community pushback forces withdrawal of Missouri data centre project

With the US Great Lakes now under threat from AI development, communities are mounting a growing fightback. According to DataCenterWatch, $64 billion worth of U.S. data centre projects have been blocked or delayed in the past two years due to opposition from residents and activist groups. The latest example is in St. Charles, Missouri, where developers CRG were forced to withdraw plans for the 440-acre “Project Cumulus” data centre after widespread concern over environmental pollution, competition for resources, and a lack of trust and transparency around the project (GovTech). Local mayor Dan Borgmeyer, who initially supported the plan, reversed his stance, warning: “You can go a long time without data, you can only go 3 days without water” (MSN). Although CRG intends to resubmit a proposal, the case underscores a larger question: who ultimately controls data centre development — corporations, regulators, or local communities — and who bears the costs in noise, pollution, environmental strain, and rising bills?

2. Whose AI is Greener? Energy & Water claims spark debate over transparency

Amid mounting controversy over AI’s resource demands, Google’s disclosure on the energy and water use of its Gemini AI was billed as a step toward transparency — but are the numbers reliable? Google reported that a median text prompt consumes just 0.24 watt-hours of electricity, 0.03 grams of CO₂, and 0.26 millilitres of water. Yet critics argue that the company calculated emissions using a market-based method that factors in renewable energy purchases made elsewhere, rather than the fossil-heavy grids that actually power its data centres — a choice that effectively cuts reported emissions by two-thirds. With thanks to @ShaoleiRen for highlighting further concerns on the methodology: the disclosure was company-led, omitted key metrics such as training costs and query volumes, and was not peer reviewed. By contrast, Mistral’s July lifecycle assessment of its Large 2 model was independently peer-reviewed and more rigorous, though it too failed to disclose query volumes. Without transparent, standardised reporting across the industry, the true ecological footprint of large language models will remain opaque. Pressure is growing for tech giants to work with independent bodies to establish reporting standards that cover the full lifecycle – training, deployment, and query volumes – so the true environmental cost of AI is clear. Work is already underway at ISO to define standardization needs for the application of AI in the financial services industry, a process that could help shape broader accountability frameworks.

3. AI’s water footprint in Africa: A Global South perspective

A Carnegie Mellon Africa study found that AI’s water footprint across 41 African countries is shaped by both climate and energy systems. Rainforest and steppe regions such as Congo and Ethiopia consumed water at or above the global average, driven by high cooling needs and reliance on hydro- and thermoelectric power. By contrast, other African states used less water overall than the U.S., reflecting lower fossil fuel reliance and reduced energy intensity. Infrastructure inefficiencies – particularly leaking pipelines – also played a significant role, a warning relevant to all countries in the Global North. The study shows how AI’s sustainability impacts are highly context-dependent, varying with energy mix, climate, and infrastructure. The growing differences between models were striking: generating a 10-page report required just 0.7 litres of water with Llama-3-70B, compared to 60 litres with GPT-4 (CMU Africa). The findings highlight the urgent need for location-specific policy and infrastructure upgrades to ensure AI’s global growth does not deepen local sustainability crises.

4. Keep it Simple – Physics based AI models outperform Deep Learning for climate prediction

MIT researchers found that simpler, physics-based AI models can outperform deep learning models in predicting climate impacts – while using far less energy (MIT News, Aug 26, 2025). Large language models (LLMs), which consume substantially more energy and water to train, struggled with the complexity of Earth systems. By contrast, simpler methods such as linear pattern scaling (LPS) proved highly accurate at gauging climate impacts. This matters because climate emulators, which inform policy modelling, depend on accurate inputs. Errors in AI-driven models risk flawed policies. The lesson is clear: better predictions may come from simpler, more accurate approaches rather than defaulting to the most complex models.

5. AI and blockchain set to revolutionise the Blue Economy

AI and blockchain are increasingly being applied across the Blue Economy – the sustainable use of ocean resources in industries such as fisheries, shipping, and marine energy. Blockchain can strengthen transparency in seafood supply chains by tracking sources, helping to combat illegal and unregulated fishing, an industry linked to forced labour and abuse in Southeast Asia (Springer, 2024; EJF, 2019). AI, meanwhile, can optimise shipping routes to cut fuel use, forecast the spread of pollutants, and integrate diverse marine datasets to support sustainable resource management (Springer, 2024). These applications show how digital technologies can enhance sustainability and governance in ocean-based industries as well as those land based – an essential but often overlooked component of the global climate system.

Final Thought

This fortnight’s newsletter highlights new considerations but recurring themes in the AI sustainability debate. Community pushback in Missouri echoes earlier cases like Teesside in the UK, where governments weighed AI expansion against alternative socio-ecological priorities. Transparency remains a sticking point: reports from Google and Mistral offer progress but still fall short of providing a full picture of AI’s true impact. Location continues to matter — climate, infrastructure, and energy mix shape whether data centres ease or exacerbate ecological strain, raising the question of a “Goldilocks Zone” for sustainable siting. Meanwhile, MIT’s findings remind us that the most complex tools are not always the most efficient, with model choice directly tied to environmental footprint. And beyond land, AI and blockchain hold promise for strengthening the Blue Economy, vital to global livelihoods and climate regulation.

All these threads underscore a central lesson: sustainable AI requires context, accountability, and a willingness to prioritise ecological realities over unchecked technological growth.

Gus Healey