RADICAL REALIST

The AI revolution comes wrapped in clean aesthetics and optimistic narratives. We love to tell ourselves that AI is the ultimate tool for sustainability – a silver bullet of sort – that will optimize our grids, model our forests, and ‘debug’ our climate.

But behind the green gloss lies a darker reality. As AI expands, so does its footprint. Not just its energy use, but its exclusion, opacity, as well as unchecked waste. The uncomfortable truth? AI will not save the planet, at least not on its current trajectory.

It could, but not without radical rethinking.

The business model is the problem

Let’s talk numbers. According to the International Energy Agency, the AI industry could consume as much electricity as Japan by 2030 (945 TWh annually). In April, Google’s ex-CEO Eric Schmidt warned that AI might one day devour 99% of our global electricity consumption. Perhaps an hyperbole, but the concern is real.

The hidden irony is that AI itself does not emit any CO₂, it’s the energy sources powering it that do. This is sometimes framed as an infrastructure challenge, but what if it is a values challenge? These costs don’t come from protecting coral reefs but from building ever-larger models to auto-complete emails or sell products. The carbon is real. The benefit is often trivial. We might have confused “scale” with “significance.”

The AI waste that no one tracks

Behind every celebrated AI breakthrough lies a digital graveyard of models: trained, tuned, discarded. For every successful large language model, researchers may train dozens or hundreds of variants, each consuming massive computational resources before being abandoned. This is part of the unaccounted-for emissions of the digital economy that never appear on corporate sustainability reports.

This is one area where “move fast and break things” is reckless, because the thing we risk breaking is the planet.

Consider the resource asymmetry: A community solar project requires environmental impact assessments and public hearings, while AI companies can silently spin up thousands of GPUs (Graphics Processing Units) in data centers with no direct public disclosure requirements for their emissions, water usage, or electronic waste footprint.

In a world that meticulously counts carbon from plastic bags, we somehow inexplicably ignore the Terawatts behind new AI systems. Digital waste remains often treated as weightless. Until we hold AI accountable for its full lifecycle impact, we are not solving climate problems, we are outsourcing them.

Sustainable AI but for whom

Let’s be clear, most “green AI” is built for those who already have power, literally and figuratively. Stable grids, cloud access, high-speed bandwidth. The tech press celebrates pilots in smart cities, not rural drought zones.

Meanwhile, in large parts of the Global South, AI is a nonstarter, not for lack of vision, but for lack of infrastructure. No energy, no AI.

And yet, we design for scale, not scarcity. If we can’t make AI run on patchy networks, in local languages, on low-power devices, it isn’t climate tech – it’s greenwashed inequality.

Frugal AI might be the only one that scales

The future of AI is not bigger. It is smarter and smaller. Frugal AI flips the paradigm. It starts from the limits – energy, bandwidth, income – and innovates within them. Think multilingual voice models that run on basic phones. Edge computing in disconnected regions. Models that optimize for tokens per watt, not petaflops per second. This is not compromise, it is the cutting edge.

The next generation of innovation will be measured not by what it can do in theory, but what it can sustain in practice – at scale, under pressure, for all.

Regulation is coming

Current AI regulation is fixated on privacy, bias, and safety. Why aren’t we mandating environmental disclosures for AI development? We regulate what we fear and incentivize what we value. So what does our silence on AI’s environmental impact say about our priorities?

This is one area where “move fast and break things” is reckless, because the thing we risk breaking is the planet. And yet, how do we regulate without killing the innovation we desperately need? As a risk expert, I am used to evaluating trade-offs. But this is one of the first times I do not know where the line lies. That uncertainty is not a weakness, it is a sign we need collective clarity, not just faster code.

What a sustainable AI future could look like

It is not complicated, but it is radical. A responsible AI future would mean:

• Models are selected for fitness, not flash
• Digital tools run on decentralized energy
• Frugal infrastructure becomes the default
• Every model discloses its environmental and social TCO (Total cost of ownership)
• Communities participate in shaping how AI is deployed, and for whom

This is not a utopia; it would be a correction.

The bottom line: critical optimism in action

This perspective may sound sharp but it is meant to be. What we need is not blind techno-optimism but critical optimism, the kind that asks harder questions while embracing genuine possibilities.

Because AI’s potential for good is real. From optimizing grid efficiency to detecting biodiversity loss through bioacoustics monitoring, AI can be a powerful ally for planetary health. These are not mirages, they are happening now. But they are happening at the margins, not at scale.

The challenge is not to abandon AI, but to enhance it: frugal by design, inclusive by default, accountable by standard. Promise alone is not progress. So let’s stop, reflect, and build the AI the planet actually needs. One that works for everyone without demanding more than Earth can give. It is not too late. But it may be later than we think.

 Gwendoline Grollier is a Partner at T3 (Risk & Regulation)