What happened
The Financial Times reports that Anthropic bought commitments for 1.7 gigawatts of heavy-duty generators from OnePWR Solutions, a San Antonio-based equipment supplier, for $3.4 billion in late September. The transaction is described as the largest known single purchase of power-generation equipment by an AI lab. Anthropic did not buy OnePWR itself.
The equipment had a history before Anthropic entered the deal. OnePWR ordered Caterpillar generators two years earlier but needed financing to complete the transaction. Anthropic initially planned to become a OnePWR customer, then chose instead to take over the equipment orders. The FT says the arrangement included engineering and construction capabilities needed to install the generators. None of OnePWR’s employees transferred to Anthropic, according to a person familiar with the matter. Both companies declined to comment.
The timing matters to the practical value of the order: the engines had been ordered years before prices rose, and the FT says they are available in the near term and over the next two years. Demand for gas turbines and generators is surging, while manufacturing lead times can stretch years into the future. AI developers have typically relied on utilities and specialist power companies, but some are using their financial resources to bypass intermediaries and take more direct control of production or equipment.
The purchase comes as Anthropic seeks greater control of the infrastructure behind its operations. The FT says the company disclosed on Monday that it had nearly $900 billion in computing commitments through 2045, relying heavily on partners for power, chips, data centres and land, and paying a premium to secure capacity and meet demand. Buying generators directly gives Anthropic more control over electricity for its facilities. The FT also reports that Anthropic’s monthly payments to lease computing capacity from SpaceX are expected to reach $3.1 billion by year-end.
The deal sits in a wider industry push. The FT reports that xAI, owned by SpaceX, bought mobile gas turbines earlier in the year for $2 billion, while Alphabet bought energy developer Intersect in late 2025. Anthropic is also preparing to go public as early as this month, while seeking to finance more than $895 billion in computing infrastructure spending through 2045. The FT notes a potential obstacle: Blackstone warned investors that it might receive a junk credit rating rather than the investment-grade rating previously expected.
What the archive adds
New Runtime's earlier coverage reported on October 26 that Anthropic was offering Blackstone more than 10% yield on part of financing for a $60 billion deal intended to support about 1.5 GW of data-centre development. The generator purchase is a new fact in the same broader financing and infrastructure story, not a restatement. Earlier coverage of Anthropic financing
Trend signal
This adds to the case that AI infrastructure strategy is expanding beyond compute procurement into control of power supply and the assets needed to deliver it. The shift is not directly captured by a tracked pattern, but it reinforces the practical importance of infrastructure constraints in operating AI systems. This does not yet fit a pattern New Runtime tracks.
Connections to other events
The Financial Times also reports that xAI bought $2 billion in mobile gas turbines earlier in the year, and that Alphabet bought energy developer Intersect in late 2025. These are distinct approaches to the same constraint: a lab can secure power equipment directly, while a parent company can acquire an energy developer. Anthropic’s transaction is notable for taking over pre-existing generator orders without acquiring the supplier.
Why it matters
For teams building AI products, this deal is a reminder that model availability and GPU capacity are only part of the infrastructure picture. At large scale, electricity supply, delivery timelines and the ability to install generation can determine whether computing capacity is available when demand arrives. Anthropic’s approach buys more control, but it also entails a substantial capital commitment and dependence on gas-fired generation.
Most builders will continue to use cloud platforms rather than source power equipment. Still, infrastructure planning should include the constraints upstream of cloud compute: energy availability, data-centre buildout and the lead times behind capacity commitments. This is especially relevant for organizations planning sustained high-volume inference or training. The reported scale is not a template for smaller teams; it is evidence that the largest AI companies are treating energy as a strategic input rather than an interchangeable utility purchase.
Limits and uncertainty
The FT attributes details about OnePWR’s financing need and employee transfers to a person familiar with the matter; the companies declined to comment. The reported purchase is a commitment for equipment, not confirmation that all 1.7 GW is installed, operational or connected to Anthropic facilities. The article does not establish the locations, operating schedule, fuel supply, emissions profile, or final cost of installation. Anthropic’s future infrastructure spending and expected SpaceX payments are projections or commitments, not completed expenditures.
Reader takeaway
Anthropic’s $3.4 billion generator commitment shows that power can be a direct bottleneck to AI capacity, not merely a bill paid after compute is secured. The deal is unusually large and does not imply that ordinary product teams should own generation, but it makes energy access and delivery schedules part of the infrastructure equation at frontier scale.