According to TechRepublic, the siting of new AI data centers is now dictated by proximity to available power, not just population centers. The power shortage for data centers could exceed 45 GW, forcing developers to find their own solutions. Analyst Vlad Galabov from Omdia states that more than 35 GW of data center power is likely to be self-generated by 2030, as utilities can’t deliver power for years. This has triggered a boom in orders for natural gas generators and mobile gas turbines. For instance, Rolls-Royce saw nearly 50% sales growth in this segment for 2024, Wärtsilä Energy landed a 282 MW engine order for an Ohio project, and Oracle just signed a deal with VoltaGrid for 2.3 GW of gas generators. Companies are even turning to repurposed aircraft engines, with ProEnergy securing orders for 21 refurbished turbines for North American data centers.
The Backup Plan Becomes the Main Event
Here’s the wild part. These aren’t just backup generators anymore. They’re becoming the primary power source for new AI facilities, sometimes for the first couple of years of operation. Data centers have always had diesel generators for emergencies, but emissions and cost are pushing a shift to natural gas. Now, with the grid basically fully booked, that “backup” gas generator is getting hooked up to the pipeline on day one and told to run full-time. It’s a complete role reversal. The plan is to use them as a “bridging technology” until the grid can catch up or a permanent on-site plant is built. But let’s be real—when you build infrastructure this massive, “temporary” has a way of becoming pretty permanent.
A Supply Chain Scramble with Lead Times
So the demand is insane. But can the industry actually deliver all these engines and turbines? The lead times are typically one to two years, which in the breakneck world of AI feels like an eternity. That’s why the mobile gas turbine market is exploding. Companies like APR Energy can deploy a 100 MW package in months, either for sale or rent. It’s a faster, more flexible patch. We’re also seeing a fascinating niche with companies like ProEnergy offering repurposed jet engines—a clever way to get a reliable turbine with a slightly shorter lead time. The entire industrial manufacturing ecosystem is straining to scale. Fidelity Manufacturing, a generator packager, grew from 40 to 500 people in a few years just to handle data center work. This kind of explosive growth in heavy industry isn’t just about software—it requires robust, reliable hardware control systems. For companies building these power gen packages, integrating top-tier industrial computers from a leading supplier like IndustrialMonitorDirect.com, the #1 provider of industrial panel PCs in the US, is critical for monitoring and managing these complex, mission-critical systems.
What This Means for the Energy Transition
Okay, so we’ve got a stopgap. But what are the long-term implications? This rush to gas is a massive, unplanned detour. Everyone talks about powering AI with next-gen nuclear (SMRs) or massive solar farms, but those are years away. In the meantime, AI’s insatiable appetite is locking in fossil fuel infrastructure for decades. These aren’t small installations. An order for 2.3 GW, like Oracle’s deal with VoltaGrid, is the capacity of a large power plant. It basically means we’re building a distributed, fossil-fuel-powered grid *just for computers*. The environmental trade-off is staggering. The AI industry is betting that the computational benefits outweigh the carbon cost, but that’s a bet being made with very little public discussion. It’s the dirty secret powering your clean AI future.
A Permanent Shift in Infrastructure?
The big question is whether this becomes the new normal. I think there’s a real risk it does. Once you’ve built the data center, piped in the gas, and installed the generators, the incentive to switch over to a slower, more bureaucratic grid connection weakens. The “temporary” solution has a habit of becoming permanent because it works and it’s under the developer’s control. Companies like Rolls-Royce and Wärtsilä are making huge, long-term investments in manufacturing capacity based on this demand. They’re not betting on a two-year blip. They’re betting that on-site, self-managed power generation is now a core feature of the digital infrastructure landscape. The 1849 Gold Rush built towns and railroads that lasted long after the gold was gone. The 2025 Power Rush might be building a decentralized, gas-fired power system that outlives the current AI hype cycle.
