AI Data Center Power: Why Used Turbines Matter

A supply queue collides with a fast buildout

The mismatch is pretty simple: demand for big power equipment is moving faster than manufacturers can produce it. Reuters has reported that global gas turbine orders have topped 100 gigawatts, while annual manufacturing capacity is estimated at 60 to 70 gigawatts. Siemens Energy, GE Vernova, and Mitsubishi Heavy Industries have all said orders are running ahead of production.

And that gap shows up in delivery timelines. Wood Mackenzie estimates the average wait for a new gas turbine in the US at six years, with some discussions stretching to 2030 and 2031. For an AI project that needs power long before that, ordering "new" equipment can be incompatible with the schedule.

To be clear, this isn't only an AI problem. Utilities and industrial customers are also competing for turbines. But data centers can demand unusually large, continuous loads. Thousands of processors, networking gear, and cooling systems all draw power while the site is operating. In practice, compute matters only when the electricity is actually there - at the right scale, and with enough reliability.

Grid power is not an instant alternative

A common thought is to connect to the public grid. In reality, new transmission lines, substations, and grid connections can still take years. The turbine queue and the grid-capacity queue are separate bottlenecks, but either one can block a site.

That's why on-site generation is attractive. A developer might build with generators on or near the campus while waiting for a larger grid connection, or use them as a longer-term supply source. Still, on-site power isn't a magic escape hatch. You need fuel infrastructure, emissions controls, construction work, and operating permits. The project has to be designed with those requirements in mind from the start.

The sheer size of the electricity demand makes timing matter. The International Energy Agency expects global data-center consumption to rise from 485 terawatt-hours in 2025 to 950 terawatt-hours in 2030, and it projects consumption at AI-focused facilities to triple over the same period. Forecasts aren't guarantees, but they help explain why power procurement has moved from a late-stage detail to a core part of planning.

The power plan is part of the product plan

For an AI company, a delayed power connection can delay access to computing capacity. That affects when a facility can serve customers, support model training, or start generating revenue. It can also force the company to pay for land, construction, or hardware long before the site can operate.

The takeaway is to treat electricity like a critical-path dependency, not a background assumption. The key question isn't only "How many megawatts do we need?" It's "When can we actually get those megawatts, under what operating conditions, and what has to be built to make them dependable?"

Aeroderivative turbines adapt aircraft technology

An aeroderivative turbine is a stationary power generator that uses technology adapted from aircraft jet engines. Instead of powering an aircraft, the equipment is configured to drive a generator. For developers, it can be another option when traditional large turbines are backlogged.

The phrase can be misleading. It can sound like someone bought engines directly from retired planes and bolted them into a power plant. That's not necessarily what's happening. A new or purpose-built generator can use aircraft-engine-derived technology without using an engine removed from a specific aircraft. The design origin and the ownership/maintenance history are different things, and they matter when you're evaluating what you're actually buying.

That distinction shows up in recent deals. In April, AI infrastructure firm Crusoe ordered 13 units from Missouri-based ProEnergy, each rated at 50 megawatts, for a total of 650 megawatts. The units use jet-engine-derived technology and are intended for future US data centers - but that order doesn't mean Crusoe bought 13 retired aircraft engines.

Also, a headline capacity number doesn't equal available electricity. The equipment still has to be delivered, installed, connected to fuel and electrical systems, tested, and permitted. "Rated output" describes what the units can produce under specified conditions - not what a campus will receive in every real operating scenario.

Refurbished equipment is a separate route

Some buyers are also looking at older turbines - and even retired aircraft engines - that can be rebuilt for stationary power. That's different from ordering a new aeroderivative unit. Refurbished equipment has an operating history, and the condition depends on the specific machine: how it was maintained, whether parts are available, and how much work is needed to convert or recommission it.

The second-hand market can add capacity, but it's not an endless supply of interchangeable machines. Suitable equipment is limited, and competition can push prices up. EthosEnergy CEO Ana Amicarella told Reuters the secondary market was very active, with asking prices climbing as more buyers went after the same equipment.

For a developer, the label isn't enough. Two machines with similar rated output can differ a lot in age, performance, maintenance needs, and readiness for the site you're planning to build. A buyer should know what's being sold, what refurbishment has already been completed, and what extra work remains before the generator can operate.

Different equipment can fit different project needs

The technology choice should follow the operating plan. A generator meant to cover a temporary gap while grid service comes online has different requirements than equipment expected to supply a major share of a campus's power for years. The same goes for backup-only needs versus continuous on-site generation.

Gas turbines aren't the only game in town. Finnish manufacturer Wärtsilä told Reuters it booked around 3 gigawatts of orders from US data centers since spring 2025 for generators powered by gas engines rather than turbines. That doesn't prove one approach is universally better. It does show developers are considering multiple equipment types as they try to secure power on a workable timeline.

So the comparison isn't just "turbine vs. engine." It's the whole solution: deliverable capacity, expected availability, fuel requirements, site constraints, emissions obligations, maintenance support, and how long it takes to reach commercial operation.

The headline price can look compelling

Reuters cited equipment brokers and energy analysts estimating that used turbines in good condition cost roughly $1.5 million to $1.7 million per megawatt. New units were estimated at about $2 million to $2.5 million per megawatt. On those numbers, a used machine can look significantly cheaper - and possibly available sooner.

One broker described a sale of two smaller turbines with a combined capacity of 29 megawatts for $12.5 million. The broker said equivalent new equipment would have cost about $35 million and taken five years to arrive. That's exactly why buyers look beyond new equipment. But it shouldn't be treated as a universal apples-to-apples comparison. Condition, specs, location, and remaining work all affect the real economics.

And the purchase price is only one piece of the budget. Buyers also need to factor in inspection, transportation, refurbishment, installation, interconnection, fuel infrastructure, emissions controls, and ongoing maintenance. If a machine needs a lot of work, the apparent discount shrinks fast. If it reaches service substantially earlier, then the value of that time may matter more than the equipment price difference.

Time has economic value - but only if the rest of the project is ready

Earlier delivery can be valuable if the data center is already in a position to operate. Getting power on sooner can help the company bring computing capacity online earlier and avoid some of the costs tied to waiting.

But a turbine that arrives early doesn't automatically create value if construction is incomplete, fuel isn't available, or air permits are still unresolved.

That's why project teams should compare complete, date-based scenarios instead of only comparing machine prices. For each option, estimate when the equipment can be installed and when it can legally and reliably produce power. Include the work that has to happen before and after delivery, and then test how likely delays would ripple through the full schedule.

A new turbine might cost more and take longer to arrive, but it can come with a clearer equipment history or a more predictable support arrangement, depending on the contract. A used machine might be cheaper and earlier, but it can require more diligence and bring more uncertainty about condition. Either way, neither option is automatically "lower risk."

Supply is scarce, not magically unlimited

A secondary market only helps if suitable equipment is available. As more data-center developers chase the same machines, prices rise and the best-maintained units become harder to secure. Refurbishment capacity and specialist labor can also act as constraints. Used equipment is a response to the shortage - not a way to remove manufacturing limits or infrastructure bottlenecks.

It can also be misleading to confuse planned capacity with operating power. Cleanview identified about 90 gigawatts of proposed on-site generation at US data centers in its mid-2026 analysis, but only about 2 gigawatts was operating at that time. That gap is a reminder: proposed capacity still has to clear design, equipment, fuel, permitting, and construction milestones.

For investors, customers, and local communities, the difference between planned, ordered, installed, and operating matters. A company can have a credible power strategy without having power online yet. The dates and dependencies behind the strategy are more informative than a single headline megawatt number.

Start with the job the equipment must do

Before evaluating a used turbine, define what you need it to do in the power system. Is it supposed to provide continuous primary power? Cover a temporary bridge until grid service is ready? Provide backup during outages? Those use cases come with different expectations for operating hours, maintenance, fuel supply, and emissions compliance.

Then match the generator to the campus load profile. Rated output isn't the same as usable power under your site's actual conditions. The project team should validate expected output and operating limits with qualified engineering and equipment specialists - not just rely on a seller's headline rating.

A practical first-pass checklist includes:

• Required power capacity and the date it must be available
• Expected operating hours and the generator's role in the wider system
• Fuel source, delivery capacity, and backup arrangements
• Site space, noise constraints, and electrical connection requirements
• Emissions standards and the permits needed for the planned operating profile
• Maintenance needs, parts availability, and qualified service support
• Testing, installation, and commissioning milestones

This process helps surface mismatches early. For example, a machine might be purchasable but not realistic for the required operating schedule, or it might be hard to permit for the specific site.

Verify condition, costs and delivery claims

Used equipment needs more than a quick price check. Buyers should confirm the machine's operating and maintenance history, the scope of refurbishment, what parts were replaced, and what work remains. They also need to clarify who takes responsibility if inspection turns up problems - and whether service support and replacement parts will actually be available when needed.

The delivery schedule deserves the same level of scrutiny. Ask what "delivery" means in the contract: arrival at a port, delivery to the site, installation, or tested operation. Those are different milestones. A credible timeline should cover engineering, transport, site preparation, construction, fuel connection, testing, and commissioning - not just the date the machine leaves the seller.

Build a comparison based on total installed cost and realistic time to operation. Add contingency for equipment repairs and possible schedule slip. Then compare the turbine option with alternatives like a new unit, gas-engine generation, or a grid connection using the same assumptions. If each option is assessed under a different definition of "capacity" or "availability," the result won't support a solid decision.

Treat permits and community impact as design inputs

Putting a power plant beside a data center changes how the site relates to nearby residents and businesses. Fuel use, air emissions, and noise aren't paperwork you solve later - they're operating realities. The US Environmental Protection Agency says stationary turbines and engines used by data centers fall under federal air-pollution standards. What rules and permits apply depends on the equipment and project details, so developers need site-specific environmental and legal review.

Permits affect both the operating plan and the schedule. If a project expects to run on-site generators continuously, it shouldn't assume the same conditions apply as they would for limited backup use. The intended hours and configuration need to be reflected in the design and permitting strategy. If planned operations don't match what's permitted, the value of the equipment can evaporate.

Community concerns are increasingly part of project risk as well. OpenAI recently promised to cover electricity and grid costs at future campuses it owns and operates after voters backed tighter restrictions on data-center construction. That won't fix every local concern, but it shows the broader direction: developers are being pushed to account for how their power demand and infrastructure affect the places where they build.