Table of Contents
Wärtsilä’s Sean Hughes said AI workloads are introducing new operational requirements for power infrastructure, with transient demand patterns emerging as an increasingly important design consideration
In sum – what to know:
Transient loads – AI training creates highly erratic power demand that can challenge both grids and conventional infrastructure.
Hybrid systems – Hughes said batteries and complementary technologies can work alongside engines to manage rapid fluctuations.
Grid evolution – He expects off-grid and hybrid approaches to remain important as utilities and hyperscalers pursue different collaboration models.
Wärtsilä’s Sean Hughes said AI workloads are introducing new operational requirements for power infrastructure, with transient demand patterns emerging as an increasingly important design consideration.
“Well, as you could probably imagine, the first five questions come in a variety of different ways, but they’re all surrounding, when can you get me engines?… The next question is price… but the new one is relatively new… we’re talking about transient loads,” the executive recently told RCRTech‘s Rack to Ran program.
He described transient loads as a defining characteristic of AI training facilities. “Essentially that is the power that’s being demanded from the data center, in the case of an AI training data center, is extremely erratic to the point where even if it’s grid connected, the grid can often not handle these extreme variations in demand for power,” he said.
According to Hughes, these conditions require multiple technologies working together. “So what that means for an onsite power solution is that you need to establish a microgrid of complementary technologies. That could be batteries, it could be synchronous condensers, it could be flywheels.”
He said Wärtsilä recommends batteries as the complementary technology and emphasized the importance of integrating them with generation assets.
“And so what happens at the end of the day is that the engines will take the base load power supply and the batteries will take the peaks of the demand profile… that can be really extreme, can change by, say, 80% in milliseconds, not seconds,” he said.
Looking further ahead, Hughes expects on-site power to remain an important part of AI infrastructure. “In my opinion, this onsite power approach has many, many years to run, if not a decade or more.”
At the same time, he said utilities will continue to play a role and hybrid models may become more common. “So we will continue to see totally off-grid power supplies and we might see a lot more hybrids whereby part of the power is provided by the grid, part of it’s by the onsite power station and you get the best of both worlds,” Hughes added.
The full interview with Sean Hughes is available here.
