Two Unrelated Headlines. One Uncomfortable Truth About AI's Future.

Two news stories landed recently. On the surface, they appear unrelated. But connect them — and a third story emerges, one that will reshape an entire industry.

India just recorded its highest-ever total energy consumption. And El Niño is projected to peak next year, bringing with it drought, heat stress, and destabilised power infrastructure across large parts of the world.

These are not separate stories. Their collision will place AI — the most resource-hungry technology ever built — under enormous pressure.

The Chain Most People Haven't Connected

Data centres already consume 1–2% of global electricity, and that figure is accelerating sharply with generative AI. Cooling alone accounts for roughly 40% of a data centre's energy use — and most cooling systems are water-intensive.

El Niño doesn't just mean heat. It means scarcity — of water, of stable grids, of predictable supply chains for the energy sector. India, which is building out its AI ambitions at scale, is doing so at exactly the wrong climatological moment. And India is not alone.

The infrastructure powering AI may itself become the resource crisis it was meant to solve.

A Natural Selection Event for AI Architectures

What we are watching is evolutionary pressure — quiet, systemic, and largely invisible until it isn't. The AI models that survive the next decade won't simply be the most capable. They will be the ones that can operate within tighter resource envelopes.

This isn't a prediction. It's physics meeting policy meeting climate. The pressure pushes hard in one direction:

• Smaller, specialised models over giant general-purpose ones

• Edge AI — processing on-device rather than in power-hungry cloud infrastructure

• Neuromorphic and analog computing — chips that approach the brain's extraordinary efficiency

• Quantized and pruned models that trade marginal capability for dramatic efficiency gains

• Cooling architectures that decouple data centres from water-stressed regions

The Irony Worth Sitting With

Google, Microsoft, and others have quietly walked back their net-zero pledges as AI power demands blew past every projection. That is politically and reputationally unsustainable — but it also reflects a genuine reckoning no one had planned for.

AI was sold, in part, on a promise of efficiency: smarter grids, better weather prediction, optimised resource allocation. The industry may now need those very capabilities to solve the problem it has helped create.

Resource scarcity may ultimately do what regulation hasn't managed to: force a rethink of the scale-is-everything paradigm that has defined AI development since 2020.

The Future Belongs to Low-Power Models

The next wave of competitive advantage in AI won't come from who trains the biggest model. It will come from who builds the most capable model per watt — per litre of cooling water — per dollar of constrained infrastructure.

The future of AI will belong to low-power models. Not because the industry will choose it — but because physics and climate will demand it.

Two unrelated news stories. One very connected future.