Data centers are turning to old jet engines as the wait for power grows

The Grid is Broken and Builders are Getting Desperate

If you want to understand the current state of the AI arms race, don't look at the sleek software demos coming out of San Francisco. Look at the industrial graveyards where old jet engines are being scrapped and sold to data center developers. We have reached a point where the demand for compute has completely outpaced the physical capacity of our civilization to power it. The result is a pivot toward "aeroderivative" gas turbines—essentially, jet engines that have been modified to sit on the ground and burn fuel to keep GPUs running.

For years, the bottleneck for builders was the chips. We couldn't get enough H100s. Now, the chips are shipping, but there is nowhere to plug them in. In major data center hubs, the wait time to get a new facility connected to the local utility grid is now measured in years, sometimes even a decade. If you are a founder or an infrastructure lead, a five-year wait is a death sentence. By the time you get power, your architecture will be obsolete. So, people are doing what builders always do when they hit a wall: they are hacking the system with whatever hardware is available.

Why Jet Engines?

The tech industry spent the last decade talking about carbon neutrality and renewable energy credits. That was a luxury of the low-interest-rate, low-compute-demand era. Today, the priority has shifted from being green to being online. Aeroderivative turbines are popular because they are modular and relatively quick to install compared to building a custom power plant or waiting for a utility substation upgrade.

These engines are designed to be light and powerful. In an airplane, that is a feature. In a data center, it is a compromise. These turbines are significantly less efficient than a combined-cycle power plant. They burn more fuel for every kilowatt produced, and they emit significantly more carbon. They are also incredibly loud and require intense maintenance. But they have one advantage that trumps everything else: you can buy them now.

The Hidden Cost of Inefficiency

Building your own power plant using 20th-century aviation tech is not cheap. The fuel costs alone are staggering. When a data center operator decides to go this route, they are effectively betting that the margin on their AI services will be high enough to cover the massive premium of self-generation. This is a clear signal that the industry believes the AI boom is not just a bubble, but a structural shift that justifies extreme capital expenditure.

However, there is a secondary cost that builders aren't talking about as much: the regulatory and social friction. Data centers are already facing localized pushback over water usage and noise. Rolling in a fleet of jet engines to run 24/7 in a suburban corridor is a recipe for a zoning nightmare. If you are building in this space, you need to realize that your technical debt is now becoming social debt.

The Infrastructure Mismatch

We are currently seeing a massive mismatch between software velocity and hardware reality. We can iterate on a large language model in weeks. We can build an app in days. But we cannot build a high-voltage transmission line in less than a decade. The permitting, the land rights, and the physical construction of the electrical grid are the ultimate throttle on the AI industry.

This is why we see companies like Microsoft and Amazon eyeing nuclear power and small modular reactors (SMRs). But SMRs are still largely theoretical in terms of mass deployment. The jet engine solution is the bridge. It is the ugly, loud, expensive bridge that gets a company through the next three to five years while they wait for better infrastructure to materialize.

What This Means for Builders

If you are a founder building at the application layer, this might seem like a problem for the big clouds to solve. It isn't. The cost of this inefficient power will eventually be passed down to you. When the cost of a token is tied to the price of fuel being burned in a repurposed jet engine, your margins are at the mercy of global energy markets.

We are moving away from the era of "infinite" cheap cloud compute. For the last 15 years, we assumed that compute would always get cheaper and more abundant. That trend is hitting a physical wall. Builders who optimize for inference efficiency—getting more out of fewer cycles—will have a massive competitive advantage over those who assume they can just throw more compute at the problem.

The grid doesn't care about your roadmap. If you can't find the power, your software doesn't exist.

The Skeptical Takeaway

There is a certain irony in using fossil-fuel-burning engines from the 1980s and 90s to power the "intelligence" of the 2020s. It exposes the fragility of the tech narrative. We talk about the future, but we are currently cannibalizing the past just to keep the lights on. This isn't a sustainable path, but in a gold rush, nobody cares about sustainability; they care about who gets to the mine first.

Expect to see more of this "patchwork" infrastructure. We will see data centers built on barges, data centers co-located with natural gas wells, and more recycled industrial machinery being put back into service. The builders who win won't just be the ones with the best code; they will be the ones who figured out how to secure their own energy supply chain without waiting for permission from the local power company.

Final Thoughts for the Founder

• Diversify your providers: Don't rely on a single region or a single cloud if they are struggling with power constraints.
• Monitor the energy markets: Infrastructure is no longer just about fiber and latency; it is about juice.
• Optimize early: Efficiency is no longer a localized performance gain; it is a hedge against rising infrastructure costs.

The jet engine trend is a symptom of a systemic failure in our energy policy, but for the builder, it is a reminder that reality always wins. You can't code your way out of a power shortage.

Read the original at TechCrunch AI →