What AI teaches us about power

On April 28, 2025, the Iberian Peninsula went dark.

For more than 10 hours, nearly 60 million people across Spain and Portugal faced a sudden and sweeping blackoutthat paralyzed cities, stalled trains, silenced digital communications, and flooded emergency rooms with complications from power-dependent devices. According to Spanish Prime Minister Pedro Sánchez, the grid collapsed with staggering speed: 15 gigawatts lost in just five seconds, the equivalent of disconnecting half the country’s energy supply in a blink.

The cause? Still undetermined. Grid operator REE pointed to a fault in the France-Spain interconnection. Portuguese authorities floated the possibility of heat-driven atmospheric anomalies. While sabotage hasn’t been ruled out, no cyberattack has been confirmed.

A fragile system in a high-demand era

We thought this incident was the perfect occasion to pause and re-examine the limits of our power infrastructure.

Because it wasn’t just a regional failure. It was a symptom of aging grids under strain, of climate-driven volatility, and perhaps most urgently, of the rising power demand from our digital age.

One of the most silent yet voracious consumers of electricity today is artificial intelligence. Behind every AI query, vision model, or large language engine are data centers running millions of computations per second, day and night, without pause.

Globally, data centers now consume 1–1.5% of electricity and projections show this could quintuple by 2030 due to AI and crypto proliferation. What’s more alarming is that AI-dedicated infrastructure consumes up to 8x more energy than conventional compute systems, and the efficiency curve is slowing (5).

The burden and the backbone of AI

AI isn’t just guzzling power but also changing the physics of how energy is used.

AI workloads operate with ultra-low inertia, meaning they don't cushion shocks to the grid. They also cause sharp power surges, steep idle-to-peak fluctuations, and heavy cooling requirements. In short, AI has created a new electrical personality, one that existing grids aren’t ready for.

According to the IMF, this could drive higher electricity prices in the near term, and contribute significantly to fossil-fueled emissions as many data centers still rely on traditional power sources (3).

But here’s where the story shifts.

AI — for all its demands — also holds the keys to resilience.

Advanced AI is already being used to detect faults in milliseconds, predict grid stress before it hits, and optimize load balancing. The International Energy Agency estimates that with the right application of AI-enabled sensors and management tools, we could unlock up to 175 gigawatts of transmission capacity, without building a single new line (7).

What happened in Iberia is a snapshot of what every connected economy could face if we fail to prepare: massive cascading disruptions, triggered by fragile links and accelerated by digital dependencies.

At Ultrai, we believe the path forward isn’t to slow down progress, it’s to build smarter systems that anticipate and adapt. That means investing in renewable-backed AI infrastructure, innovating in heat reuse, advancing energy-aware models, and designing with a systems view of sustainability.

Because the truth is: AI is here to stay. But our grid, as it stands, can’t stay the same.

And if we want AI to power the future and, not pull the plug on it, the time to act is now.

Sources

1. "Spain and Portugal blackout: electricity loss, possible causes and impact" by The Guardian

2. "Avería eléctrica como causa más probable del apagón" by The Huffington Post

3. "AI, Data Centers and Energy Report" by Institut Français des Relations Internationales (IFRI)

4. "AI is poised to drive a 160% increase in power demand" by Goldman Sachs

5. "Generative AI power consumption and the need for sustainable data centers" by Deloitte

6. "The Unseen AI Disruptions for Power Grids: LLM-Induced Transients" by Yuzhuo Li, Mariam Mughees, Yize Chen, Yunwei Ryan Li

7. "AI’s dual impact on electricity demand and grid optimization" by IEA