You probably don't think about gas generators when you ask Claude to write an email or help with your work. But Anthropic apparently has to.
The company behind Claude has reportedly agreed to spend $3.4 billion on gas-powered generators to help power its AI data centers.
According to the Financial Times, the deal covers equipment capable of generating 1.7 gigawatts of electricity.
That's a huge amount of money to spend on power equipment, even for an AI company. And it raises an obvious question: Why doesn't Anthropic just get its electricity from the power grid like everyone else?
Well, AI companies are building data centers so quickly that electricity suppliers are struggling to keep up. Anthropic reportedly wants to get ahead of the problem, even if it means spending billions on generators that burn natural gas.
And that's where things get complicated.
Anthropic reportedly spent $3.4 billion to skip the waiting line
According to the Financial Times' November 8 report, Anthropic reached an agreement with OnePWR Solutions, a Texas-based company that supplies power equipment.
OnePWR had already ordered heavy-duty gas generators from Caterpillar, with some orders reportedly placed years ago.
Anthropic agreed to take over those orders rather than start from scratch.
There's a reason for that. Demand for power equipment has grown so much that companies can face years-long waits for new generators.
Anthropic originally planned to buy electricity from OnePWR, according to the FT. But the arrangement reportedly changed, and Anthropic decided to secure the equipment itself.
The $3.4 billion agreement also covers engineering and construction services needed to install the generators. Anthropic reportedly didn't buy OnePWR or take over its employees.
And to be clear, Reuters couldn't independently confirm the transaction. Neither Anthropic nor OnePWR responded to its requests for comment.
The generators aren't necessarily ready to use, either. Their eventual locations and operating schedules haven't been confirmed.
Still, taking over existing orders could save Anthropic time if the company can get the equipment installed sooner than it could obtain new generators.
Why can't Anthropic just plug its data centers into the grid?
Every time someone sends Claude a prompt, computers inside a data center process the request. Those machines need electricity, and so do the cooling systems that stop them from overheating.
Now multiply that by millions of users and the enormous computing systems required to train newer AI models.
You can see why electricity is becoming a headache for AI companies.
According to the International Energy Agency, data centers worldwide consumed around 485 terawatt-hours of electricity in 2025. By 2030, that could reach 950 terawatt-hours.
AI-focused data centers are expected to account for much of the increase.
The trouble is that building new electricity infrastructure takes time. Power companies may need to install transmission lines or expand generating capacity before they can supply a large data center.
Sometimes, the computers are ready long before the electricity is.
For a company like Anthropic, that could mean spending billions on advanced AI chips, only to have some of them sit unused while it waits for a power connection.
And with companies competing to release more capable AI models, waiting isn't exactly appealing.
Anthropic's reported generator deal could give it more control over when electricity becomes available for its facilities. Of course, the company would still need somewhere to install the equipment and a reliable natural gas supply to operate it.
There's a catch with using gas generators
Gas generators have an advantage that AI companies appreciate: they can produce electricity whenever it's needed, as long as fuel is available.
That's helpful for data centers that need to run around the clock.
But there's an environmental problem.
Natural gas produces carbon dioxide when burned. Gas-powered equipment can also release air pollutants, while leaks during fuel production and transportation can add methane emissions.
If Anthropic eventually runs these generators for long periods, they'll contribute to the company's environmental footprint.
And AI's electricity-related emissions are already expected to increase.
In its October 2026 report, the IEA estimated that annual emissions associated with electricity used by data centers could rise from about 150 million tonnes of CO2 to 350 million tonnes by 2035.
Part of that increase comes from companies turning to on-site natural gas generation to power their facilities.
Now, those numbers cover the global data center industry. We still don't know how frequently Anthropic would run its reported generators, so it's too early to estimate their emissions.
But here's something worth thinking about.
If Anthropic spends $3.4 billion securing gas-powered equipment today, will it still want to use those generators years from now, even if cleaner electricity becomes easier to access?
After all, companies generally want to get their money's worth from expensive equipment.
Anthropic could eventually use less gas if other electricity sources become available. How long it relies on the generators will determine their environmental impact.
Other AI companies are trying to solve the same problem
Anthropic isn't the only AI company concerned about where its electricity will come from.
Meta has been investing in nuclear energy and other power sources to support its expanding data centers. As The Neuron previously covered, the company has ambitious plans to build enormous facilities for AI.
Google and NVIDIA are also exploring ways to make data centers less demanding on electricity grids.
In September, The Neuron reported on an industry alliance that wants data centers to temporarily reduce electricity consumption when the grid is under pressure.
Some companies are taking another approach.
The IEA describes a growing interest in "bring your own power," where data center developers arrange their own electricity generation instead of waiting for existing grids to accommodate them.
It sounds like a straightforward solution until you consider everything involved.
Generators still need fuel and permits. Someone also has to install them and deal with the equipment's emissions.
Then there's the question of who pays for improvements to the surrounding electricity infrastructure.
Communities near new data centers may worry about higher electricity bills or additional pollution, especially when projects require substantial upgrades to local power systems.
AI companies will have to answer those concerns as they build more facilities.
So, is electricity becoming AI's next big expense?
Anthropic's reported deal offers a pretty good indication of where the AI industry's money is going.
Companies are already spending enormous amounts on chips and data centers. Now, some are prepared to spend billions more to make sure those facilities have enough electricity.
It's understandable why Anthropic would want to secure its power supply. Delays could leave costly computing equipment unused while competitors continue expanding.
But the environmental consequences are harder to ignore.
Anthropic may be able to get its data centers running sooner with gas-powered equipment. The unanswered question is how long it will depend on fossil fuels to keep them running.
