PowerGen has long been an energy industry conference, far removed from IT concerns. This year’s January show? Turned everything on its head.
Data centers dominated. The organizers put together a Data Center Pavilion and an entire data center track. Even in sessions that were supposed to have nothing to do with data centers, the presenters would inevitably broach the topic and explain how AI’s power needs are transforming the industry.
“There has been more than $100 billion a month in new AI data center projects announced worldwide for the past year, with the majority being in North America,” said Shane Mullins, VP of Product Development at Industrial Info Resources (IIR). “These are real projects, not pie in the sky.”
October 2025 alone recorded more than $350 billion of tangible data center projects under development. This, Mullins said, is no longer driven by the compute demands of training large language models (LLMs). We have moved into the inferencing stage. It is inference by applications that now consumes massive amounts of compute resources.
In addition to generative AI applications like Gemini and ChatGPT, AI is being used in autonomous vehicles, robotics, liquefied natural gas, and more.
North America’s AI dominance
Of the roughly $3.2 trillion in ongoing data center projects worldwide being tracked by IIR, North America accounts for two-thirds.
Additionally:
- Europe accounts for $440 billion
- Latin America: $125 billion
- East Asia: $82 billion
- Southeast Asia: $81 billion
- China: $70 billion
- South Asia: $65 billion
While IIR believes these are all realistic projects, not all are scheduled to kick off in 2026 or 2027. About 4,300 of the 7,250 data center projects will begin construction in the next 18 months or so.
Where AI meets the limits of power
IIR named power as the primary constraint within the data center market. With rack sizes of up to 600 kW on the near horizon, the scramble is on to find power from anywhere. Gas turbines of all sizes have largely been gobbled up by the end of the decade.
In response, gas engine vendors were out in force at PowerGen promoting a range of designs for data centers. These companies are cobbling together multiple smaller gas and diesel engines to provide data center developers with the power they need during construction and beyond. One vendor said they received a rush order after one large AI data center realized it would be short of about 20 MW of power.
Other options cover the gamut. Batteries are in high demand. Tesla is doing plenty of business in data centers for its battery plants. Nuclear, too, is being looked to for substantial AI power. One data center developer even said he was negotiating to buy an old nuclear submarine or aging aircraft carrier to use the reactor to power a facility.
“With utilities telling data center developers that any requests for power over 500 MW will likely have to wait until the 2032/33 timeframe, the build-your-own-power concept has become a necessity for many projects,” said Britt Burt, senior vice president of Research for the Power Industry at IIR. “This is driving the hyperscalers and others to partner with power industry and construction industry specialists with the know-how to drive projects through to completion.”
Related: OpenAI says its Stargate data center project will cover its own power costs as AI infrastructure scales.
