Flexible AI Data Centres May Help Cut Grid Costs
Artificial intelligence is driving one of the fastest jumps in electricity demand the grid has seen in years. Global electricity consumption from data centres surged in 2025, and it's set to grow quickly from here, with demand almost doubling by 2030. For AI-focused facilities specifically, power use is expected to climb even faster, tripling over the same period. That growth is forcing a question grid planners haven't had to ask at this scale before: how do you connect all these new data centres without triggering expensive spikes in power demand?
New research published in iScience suggests a surprisingly simple answer, and it isn't "use less electricity." It's "use it more flexibly."
The idea: shift the work, not cut it
Most data centres run below full capacity most of the time. That spare capacity matters, because a lot of AI computing work isn't time-critical. Training an AI model, for example, doesn't need to finish at a specific moment; it just needs to finish. So instead of running flat-out whenever demand happens to spike, that kind of workload can be paused or slowed during peak hours and picked back up a few hours later, once the pressure on the grid eases.
Researchers modelled this idea across three real power systems in the US (Texas's grid, the Mid-Atlantic region, and the Western US grid) to see what would happen if data centres routinely shifted flexible workloads away from peak demand. The result was consistent: system-wide electricity costs went down. By smoothing out the sharp peaks in demand, flexible data centres let grid operators get more out of the power plants and transmission lines already in place, rather than needing to build new capacity just to cover a handful of peak hours a year.
The catch: it doesn't automatically mean cleaner electricity
Here's the part worth paying attention to. Shifting demand around doesn't guarantee lower emissions; it depends entirely on what's generating the electricity at the time.
In regions with plenty of wind and solar power, flexible data centres are a clear win for the climate: they can soak up renewable electricity that might otherwise go to waste when generation is high, instead of drawing power when fossil fuels are filling the gap.
But in regions still dominated by coal and gas, the same flexibility can cut the other way. Smoothing out demand peaks can let fossil fuel plants run more steadily rather than switching on and off, which is efficient for the grid, but can mean those plants are used more overall. Without cleaner generation available to shift toward, that steadier fossil fuel use could increase emissions rather than reduce them.
In other words: flexibility helps the grid either way, but it only helps the climate where clean power is already there to take advantage of.
A sign of where things are heading
This isn't just a theoretical exercise. A separate study, Power-Flexible AI Data Centers: A New Paradigm for Grid-Responsive Compute, has already shown the concept working in practice: software controlling a real AI data centre's power draw in real time, shifting workloads and adjusting demand in response to grid conditions, without disrupting performance on priority tasks.
Together, the research points to a shift in how we think about data centres: not as a fixed source of new demand the grid simply has to absorb, but as active participants that can help balance the system they're part of.
The direction is promising: lower system costs, better use of existing infrastructure, and a genuine route to easier renewable integration. But how much of that promise turns into lower emissions will keep coming back to the same question the whole energy transition depends on: how quickly clean generation keeps growing.