Latest top stories
Technology

AI’s Growing Energy Crisis: Why COP30 Is Sounding the Alarm

13 November 2025

When delegates step into the humid Belém air for COP30, they’ll be greeted by the usual line-up of climate emergencies: phaseouts, finance, adaptation, the Amazon. But humming just beneath the headlines is a new, fast-growing concern—one that barely registered at climate summits a year ago.

It isn’t a petrostate or an oil giant. It’s something far more abstract, seductive, and omnipresent:

Artificial intelligence—and its exploding energy consumption.

AI, long hailed as a potential climate saviour, is quietly becoming a climate problem. Its electricity demand is soaring so fast that European utilities, policymakers and data-centre operators are scrambling to catch up. For the first time, a global climate summit is acknowledging what many engineers already know: if AI infrastructure continues expanding unchecked, it could undermine the world’s climate goals.

Europe, squeezed between climate commitments and competitive pressure to build AI capacity, has become the world’s laboratory for what some are calling Green AI Infrastructure—AI data centres engineered to minimise carbon emissions, manage electricity demand and survive the next decade of compute growth.

Whether this experiment succeeds will shape not only Europe’s digital future but also the global emissions trajectory of AI itself.

The AI Boom Meets a Harsh Energy Reality

In just two years, generative AI has gone from experimental novelty to everyday utility. Business processes, creative workflows, manufacturing planning, retail systems—everything is being re-engineered around AI models running on vast clusters of power-hungry chips.

But all this intelligence comes with a hefty electricity bill.

  • The International Energy Agency warns that global data centres could consume as much electricity as Japan by 2030, with AI-focused facilities driving a significant share of the growth.

  • Europe’s data-centre consumption could more than double to over 150 TWh by 2030, according to industry forecasts.

  • Training a single large AI model is estimated to emit hundreds of tonnes of CO₂e, depending on energy source.

  • Some European utilities say they are now receiving grid-connection requests “the size of small towns,” reflecting the growth of AI factories.

The transformation has been so sudden that many governments didn’t see it coming. Ireland has slowed data-centre connections around Dublin. The Netherlands tightened national rules after early moratoriums in Amsterdam. Belgium is debating similar measures. None of these actions were part of long-term planning—they’re crisis management.

Why Traditional Data Centres Can’t Handle AI

The data centres built over the last decade were designed for cloud storage, streaming and business applications—not AI’s extreme compute density.

Two big shifts explain why technology outpaced infrastructure:

1. GPUs Have Replaced CPUs

AI models rely on accelerators like GPUs and TPUs, which draw far more power than traditional CPUs. A typical enterprise data-centre rack consumes 6–15 kW. An AI rack can demand 50–120 kW or more.

2. AI Needs Tight Clusters, Not Distributed Compute

To train a frontier-scale model, thousands of accelerators must sit in close proximity, generating concentrated heat that air cooling simply can’t handle.

Older data centres can’t be retrofitted fast enough. Air cooling becomes inadequate. Power distribution maxes out. And grid-connection queues stretch into years.

This pressure is driving the rise of AI factories—ultra-dense, electricity-heavy, liquid-cooled facilities built from the ground up to host GPU clusters.

The new AI Factory in Munich, where Polarise is responsible for the specialised infrastructure, represents one of Europe’s early attempts to build an AI-first, energy-aware facility. It’s compact, purpose-built, and engineered around high-density compute from day one.

Across Europe, similar projects are emerging. Together they form the first wave of what the industry is calling sustainable AI infrastructure.

At COP30, AI Steps Into the Climate Spotlight

AI wasn’t a headline topic at COP27 or COP28. It appeared on panels and in side discussions, but climate negotiators were still focused on fossil fuel politics and finance gaps.

By COP30, the tone has changed.

Negotiators are circulating briefing papers on three emerging issues:

1. AI’s carbon footprint is invisible in corporate reporting

Companies disclose cloud emissions, but rarely show which portion comes from AI. This makes AI’s real climate impact hard to measure—and easier to ignore.

2. Grids are already feeling the strain

Utilities in Ireland, the Netherlands, Belgium and parts of Germany say new AI clusters could require grid reinforcement or even new generation capacity.

3. AI’s energy demand is accelerating faster than policy can respond

Electricity systems evolve over years. AI deployment evolves over weeks. The mismatch is becoming unsustainable.

Europe’s Patchwork of Green AI Innovations

While global climate governance is still catching up, Europe is already experimenting with solutions—out of necessity more than idealism.

Nordics & Iceland: Renewable-Powered GPU Farms

Some of Europe’s lowest-carbon AI clusters sit in the Nordics and Iceland. Abundant hydro, wind and geothermal energy allows data centres to run on near-100% renewables with exceptionally low PUE ratios (1.05–1.2). Cool climates reduce the need for mechanical cooling entirely.

Germany: Urban High-Density AI Zones

Germany is building AI factories close to its industrial base. Facilities like the one in Munich prioritise:

  • high-voltage grid connections

  • advanced liquid cooling

  • proximity to automotive and manufacturing clients

Netherlands: Europe’s Regulatory Prototype

After rapid early growth, the Netherlands introduced stringent data-centre planning rules, including renewable-purchase requirements and, in some areas, mandatory waste-heat reuse.

Iberia: Solar-Powered Compute Regions

Spain and Portugal are rapidly positioning themselves as Europe's solar-backed AI hubs, with gigawatt-scale renewable projects directly supporting data-centre development.

Step by step, Europe is sketching out the contours of a future where AI capacity expands—but doesn’t blow up climate commitments.

Five Technologies That Could Make AI Sustainable

1. Liquid and Immersion Cooling

Replacing air cooling with liquid systems can cut cooling energy use by tens of percent and reduce overall facility power consumption by 10–25%.

2. On-Site Renewables & Battery Storage

AI factories increasingly deploy private solar fields, battery energy storage systems (BESS), and hydrogen-ready backup power to stabilise grid load.

3. Carbon-Aware Workload Shifting

Software can now route AI training or inference to data centres with the cleanest available electricity mix, hour by hour.

4. Waste-Heat Recovery

Cities such as Stockholm and Odense already channel data-centre heat into district heating networks. AI facilities, with their high thermodynamic output, offer even greater potential.

5. Water-Efficient Cooling & Hardware Circularity

Closed-loop liquid cooling and GPU refurbishment programmes are emerging as key answers to rising water scarcity and e-waste.

The Fork in the Road: AI as Climate Hero or Climate Burden?

AI could dramatically accelerate climate solutions—improving grid forecasting, modelling extreme weather, verifying carbon removal and slashing industrial emissions.

But if powered by carbon-heavy grids and inefficient infrastructure, it could:

  • trigger fresh fossil-fuel investment,

  • overload fragile electricity systems,

  • inflate corporate emissions, and

  • weaken national climate targets.

The technology isn’t inherently good or bad—it’s energetic. Whether it helps or harms the climate depends entirely on how its infrastructure evolves.

A New Frontier for Climate Policy

As COP30 unfolds, one thing is becoming clear: AI is no longer a sidebar in climate policy—it’s a central actor.

Europe’s early experiments—liquid-cooled AI hubs, strict planning rules, solar-powered compute zones—may soon form the blueprint for the rest of the world. But time is tight, and the growth of AI isn’t slowing.

For now, the race to build Green AI Infrastructure has begun. And how quickly we succeed may define whether AI becomes the engine of the net-zero transition—or its undoing.