The AI revolution is happening at lightning speed. But few people realise the huge strain these powerful data centres are placing on the world’s electricity grids. Ayșe Coskun’s TED Talk sheds light on how the very technology causing these problems could actually be the solution.
Key Takeaways
- AI data centres use a massive amount of electricity, sometimes as much as whole cities.
- They’re often blamed for rising energy bills and grid instability.
- But these centres aren’t just energy guzzlers—they can also help stabilise power grids if managed smartly.
- Making data centres more flexible with their power use could fast-track new projects and save communities money.
- AI can manage these systems in real time, like a conductor leading an orchestra.
The Growing Problem of AI’s Energy Appetite
Whenever someone mentions ‘AI data centres,’ most folks immediately think of one thing: energy hogs. And—let’s be honest—they’re not wrong. Training advanced models, like GPT-4, can eat up as much electricity as thousands of average homes. In Ireland, nearly a fifth of the country’s whole electricity use now goes straight to these centres.
This isn’t just an abstract number. In places like Virginia’s “Data Centre Alley,” there’s been a real impact. Locals have noticed electricity bills shoot up by 20% in just a few years. Utilities are racing to supply these new facilities, and it’s getting difficult to keep up.
| Region | Impact |
|---|---|
| United States | Data centres require city-sized power loads |
| Ireland | Data centres use 20% of the country’s electricity |
| Virginia | Residents saw 20% higher bills |
Rethinking What Data Centres Can Do
But that’s only half the story. Here’s the twist: unlike our fridges or hospitals, which need a steady, constant supply, data centres often run tasks that are totally flexible. Some jobs can be delayed or slowed down without anyone caring. For example, if you’re running a giant research project or fine-tuning an AI model, you’ll probably be fine if it takes a little longer, right?
This opens up a whole new way to use these centres. Rather than just gulping down electricity, they can act like the ‘muscles’ of the grid—flexing up or down, soaking up excess power when the sun is shining or dialling back during evening peaks.
Old Way vs. New Way
- Old way: Data centres run all out, chewing through electricity non-stop.
- New way: Data centres adjust in real time, matching the supply available and helping smooth out peaks and valleys in grid demand.
How Flexibility Changes Everything
There’s a big technical shift here. Instead of always trying to compute as fast as possible, engineers are now asking: How do we meet our computing needs while also helping the grid?
What does this mean?
- Capping Power Use: Only draw as much as the grid can spare.
- Shifting Workloads: Delay less urgent jobs to times when there’s more supply.
- Acting as Virtual Batteries: Soak up excess solar power, then slow down when everyone else needs energy.
This is already moving out of the lab and into the real world. Coskun talks about early prototypes seeming impossible—most people thought it would never work. But after years of stubbornness (her word!), it’s now actually running in live data centres.
Why Timing Matters Most
The grid’s biggest challenge isn’t always generating more electricity—it’s matching supply and demand hour by hour. Solar power peaks at noon, but everyone uses most electricity in the evening. Wind is even more unpredictable.
Nuclear energy is slow and expensive to build. Batteries are important but costly to scale. Meanwhile, new data centres can wait up to seven years just to connect to the grid. Yet, AI moves fast, changing every six months. Clearly, something’s got to give.
The Future: AI Conducting the Orchestra
So, what’s the catch? Managing all this is complicated. Prices jump around by the hour. Workloads come and go. Every country, every utility has different rules. No one person can handle all that.
Here’s where AI comes in again. The same tech that’s creating the need for these massive data centres can also help run them smartly. AI learns, adapts, and makes real-time decisions—shifting workloads, managing power draws, even working across whole networks of data centres.
Imagine these server farms as an orchestra—a mess of sound on their own, but with an AI conductor, everything works together, smooth and in tune. The result? Cleaner, cheaper, more reliable computing—and power grids that don’t break under the strain.
Why This Matters Now
For Coskun, this has been a long journey from rough sketches to working systems. She believes this is only the beginning. If we’re bold enough to see these ‘energy hogs’ as part of the answer, not just the problem, we might find that technology can help power a brighter, cleaner future.
Here’s the real question we need to ask: Not just ‘How much energy does AI use?’ but ‘How much flexibility and clean power can AI help us unlock?’