Anthropic buys 1.7 GW of gas generators for AI data centers

By Nathan Brooks

Anthropic has agreed to spend $3.4 billion on 1.7 gigawatts (GW) of gas-fired generators and related installation work for its AI data centers. The late-September agreement, reported by Reuters, transfers existing Caterpillar generator orders from Texas-based OnePWR Solutions to Anthropic. It also covers the engineering and construction needed to install the equipment.

Anthropic is not buying OnePWR itself. The underlying Caterpillar orders were placed roughly two years ago and required additional financing. By taking them over, Anthropic avoids starting from scratch and waiting for a new large-order production slot.

Global turbine orders have reportedly exceeded 100 GW, while estimated manufacturing capacity stands at 60 GW to 70 GW. Average delivery times in the United States have reached six years, and some prospective orders stretch into 2030 or 2031. Large gas-turbine orders are piling up as AI companies compete for power that can arrive sooner than conventional grid connections.

What does taking over OnePWR’s orders change?

Taking over OnePWR’s orders is closer to buying a reserved production slot than placing a new order at the end of the queue. Anthropic gains equipment already committed by the supplier rather than turbines that still need a manufacturer slot. The transaction also gives it installation capability, not merely a collection of generators.

That distinction matters because securing hardware without engineering and construction would leave another schedule risk standing at the gate. The deal covers the full installation scope, reducing the chance that Anthropic must coordinate those tasks separately from the generator commitment.

This is not an Anthropic-only maneuver. xAI has already used converted General Electric jet engines to bring large computing facilities online without waiting for conventional grid connections. In Mississippi, the company installed eight such units at a data center. Each can produce up to 50 MW, giving the installation a potential combined output of 400 MW.

The approach carries an environmental challenge that cannot be scheduled away. The Southern Environmental Law Center has challenged the emissions permits for the Mississippi site. Shakeel Hashim, founder of ReadTransformer, tweeted: “This is pretty grim—these kinds of generators are terrible for the environment.”

How are AI companies changing turbine supply?

Musk’s companies have moved further into the equipment chain. In August, it emerged that they were bringing turbine-blade and vine manufacturing in-house. Musk identified component production as a major constraint on generator deliveries, and some turbine blades reportedly take 60 to 90 weeks to manufacture.

SpaceX acquired APR Energy in July, adding a fleet of trailer-mounted gas and diesel generators capable of producing more than 1 GW. The acquisition gives Musk’s businesses access to a large mobile generation fleet, while in-house component work targets the production delay behind larger systems. Together, these moves shift the problem from securing generators to controlling how their parts get made.

OpenAI took a different route for its first Stargate data center in Abilene, Texas. In July, Crusoe, the facility developer, acquired 29 GE Vernova LM2500XPRESS gas turbines rated at 34 MW each. The units could produce up to 986 MW in total. Anthropic, by contrast, took over Caterpillar orders for existing equipment instead of buying a newly assembled turbine fleet.

Can used equipment close the gap?

The power shortage has pushed AI companies beyond new generator orders, producing an upsurge in demand for used and refurbished turbines. Retired jet engines converted into generators have also become more attractive as developers seek capacity without waiting years for new production slots.

Those options do not eliminate the supply problem. Buyers must still assess maintenance requirements, remaining service life and whether a unit can support continuous operations. For Anthropic, acquiring OnePWR’s Caterpillar commitments could sidestep some of those questions because the equipment and installation scope are already defined. It does not remove the environmental cost of gas generation or the manufacturing constraints affecting turbine components.

The $3.4 billion commitment shows the scale of capital AI companies are putting behind rapid power access in a market where reported orders exceed estimated factory capacity. With turbine slots scarce and some delivery dates reaching 2031, AI data-center developers are likely to compete for both new and refurbished equipment, while component lead times determine which power projects come online first.