Technology

Why an AI Company Is Spending $3.4 Billion on Gas Generators

Martin HollowayPublished 3d ago3 min readBased on 4 sources
Reading level
Why an AI Company Is Spending $3.4 Billion on Gas Generators
source:anthropic.com

Anthropic is spending $3.4 billion on gas-fired generators to power its AI data centers, according to a Financial Times report published on November 8, 2026. Reuters carried the FT account the same day.

The machines are made by Caterpillar. Together they can produce up to 1.7 gigawatts of electricity, which is a very large amount of power.

Another company, OnePWR, had ordered the same generators about two years ago. It could not find the financing to complete the purchase. Anthropic did not take over that order as originally planned. Instead, it bought the generators directly from Caterpillar, according to TechCrunch, which reported the change on November 9, 2026.

Buying directly is what lets Anthropic move ahead of a line. Tom's Hardware reports that the deal lets the company skip a six-year waitlist for this equipment.

Here is the order of events. A large purchase was arranged but sat for about two years without the money to finish it. A buyer with the funds then stepped in, and the waiting list for this equipment did the rest.

For people who plan data centers, the bottleneck is no longer only the specialized chips that train and run AI models, or the networking that links them. Power matters too. Firm power, meaning electricity that can be switched on whenever it is needed, and the factory slots that deliver the equipment to produce it now sit near the front of the line. Whoever holds a paid-for place in that line sets the pace for everyone behind them.

In my view, the missing financing matters more than the headline number. A generator order is a promise of future power. Like a restaurant reservation, it holds a place only if the deposit is paid. The OnePWR order shows that demand alone did not carry the project to completion. That is a useful signal for operators, developers and equipment makers who are weighing how much of the AI buildout depends on who can pay for hardware with long delivery times.

The broader context here is that AI computing is now as much a question of electricity as of chips. Gas-fired power at this scale raises questions about permits, fuel supply and emissions, and a single purchase does not settle any of them. What the deal does establish is narrower: a leading AI developer has taken direct ownership of a large block of on-site generating capacity after a supplier's order stalled. Whether other companies follow will depend on how the next wave of data-center projects is financed and where they are built.

The practical lesson for anyone building or buying computing capacity, as I read it, is that a turbine order is only as solid as the money behind it. Companies that can commit capital early may decide which projects get power on schedule. That makes equipment queues and financing a large part of the path to more AI computing, alongside the model training that gets most of the attention.