Anthropic Pays $3.4 Billion for 1.7GW of Gas Generators to Power AI Data Centers

Anthropic has agreed to pay approximately $3.4 billion for 1.7 gigawatts of Caterpillar gas‑fired generators, the Financial Times reported via CNBC. The deal, signed in late September with San Antonio‑based equipment supplier OnePWR Solutions, is the largest known single purchase of power‑generation equipment by an AI lab.

OnePWR had placed the Caterpillar orders about two years ago but needed capital to complete the purchase. Anthropic was originally planning to buy power from OnePWR as a customer. Instead, it took over the entire queue position — generators plus the engineering and construction capabilities required to install them, according to AI Weekly. Anthropic did not acquire OnePWR itself, and none of the supplier's employees transferred to the AI startup.

The deal solves a prosaic but urgent problem. Demand for gas turbines and large generators has outrun factory capacity as AI developers race to build data centers, and fresh orders now face long manufacturing waits, AI Weekly noted. Paying $3.4 billion to inherit a two‑year‑old order is faster — likely by years — than placing a new one and waiting at the back of the queue.

At roughly $2,000 per kilowatt, the pricing bundles equipment and construction, making precise per‑unit comparisons difficult. But the strategic logic is clear: for an AI company racing to train and serve ever‑larger models, time spent waiting for power is time competitors use to pull ahead.

The generator deal fits within a far larger spending picture. Anthropic's SEC filing for its upcoming IPO disclosed $895.4 billion in total infrastructure commitments, though it did not itemize this specific generator transaction, MLQ.ai reported. The IPO document itself flagged electricity access as the binding constraint on data center expansion, stating that "access to electricity limits data center expansion in several markets and new power sources and grid connections can take years to develop."

Reuters separately reported in September that Anthropic expects to spend more than $500 billion over the next 10 years building AI infrastructure, CNBC noted. Strong demand for the Claude family of AI models is driving the spending.

The deal lands as Anthropic prepares for what could be one of the largest tech IPOs in history. Investors have recently been pricing the company in a $1.2 trillion to $1.3 trillion range, down from $2 trillion talk earlier in the autumn, according to AI Weekly. The IPO is expected before Thanksgiving, with Class A shares listing on Nasdaq under the ticker ANT, led by Goldman Sachs and Morgan Stanley.

Piper Sandler, which served as OnePWR's exclusive financial adviser, described the September transaction as the sale of its "power‑generation platform" to a "frontier AI lab" without naming the buyer. The deal has drawn attention from energy‑sector analysts who note that Anthropic must fund grid upgrades and cover consumer electricity price increases — obligations that extend well beyond the $3.4 billion purchase price, per The Energy Mag.

Every Claude API call, every prompt in Claude Code, and every Claude‑powered application runs on physical infrastructure that is becoming staggeringly expensive to build and power. The $3.4 billion generator deal is a single line item in a spending trajectory measured in hundreds of billions. For builders, the practical takeaway is straightforward: AI model access is not getting cheaper at the infrastructure level. The companies providing these models are making decade‑scale bets on physical hardware — and those costs will eventually be reflected in API pricing, rate limits, and service availability.

The race to secure power is now as competitive as the race to build better models. When the Financial Times calls this "the largest known single purchase of power‑generation equipment by an AI lab," per the Financial Times, it is describing not just one deal but an industry‑wide condition where electricity itself has become the scarcest resource in AI development.