The number as a thermometer for the challenge
The most telling part of the IEA report isn’t just how much consumption will grow, but how it grows. During the training of the most advanced AI models, electricity demand can spike by 400% within a fraction of a second. Conventional grids simply weren’t built to absorb swings like that. The answer lies in industrial-scale battery storage, smarter grids and — somewhat paradoxically — more AI itself: the IEA estimates that if these same algorithms were applied to energy management at scale, potential savings by 2035 could equal Indonesia’s entire annual electricity consumption.
This is the data-driven sustainability we’ve been discussing for months, now playing out at a scale that leaves no room for error.
Where engineering comes in: the building decides too
This is where the debate stops being macroeconomic and becomes very close to home. Spanish regulation already requires new data centres to target a PUE (power usage effectiveness ratio) below 1.2, and 90% of projects currently in design prioritise liquid cooling and high-density architectures to get there. That number — PUE — isn’t report jargon: it’s HVAC systems, electrical distribution, BMS and MEP design thought through from the first sketch.
Every watt saved in cooling is a watt that never needs to be generated. And that space — between the server and the power plant — is exactly where we work: installations that aren’t an afterthought to the building, but the reason the building performs, saves and lasts.
A closing thought
The conversation around AI tends to focus on algorithms and models. But its most tangible footprint is measured in kilowatts, grids, cooling systems and engineering decisions that rarely make headlines. In this context, innovation isn’t just about going digital — it’s about designing the physical infrastructure that makes everything else possible, and sustainable.
How is your sector preparing for this new wave of energy demand? We’d love to hear your take in the comments.