Anthropic Secures 1.7 Gigawatts of Gas-Fired Generators in $3.4 Billion Data Center Deal

Anthropic has committed approximately $3.4 billion to secure 1.7 gigawatts of heavy-duty gas-fired generators for its data centers. The transaction represents the largest known single purchase of power-generation equipment by an artificial intelligence laboratory. By acquiring the hardware directly, the company is effectively bypassing traditional utility providers.[1][3]

The agreement, finalized in late September 2026, centers on equipment commitments held by San Antonio-based supplier OnePWR Solutions. The Financial Times first reported the arrangement, noting that the underlying hardware stems from orders placed with heavy equipment manufacturer Caterpillar Inc. two years ago.[1][4]

The deal structure avoids a corporate buyout. Anthropic did not acquire OnePWR Solutions itself, and no employees from the Texas supplier will transfer to the AI startup. Instead, the transaction transfers ownership of the Caterpillar orders while securing the engineering and construction capabilities required to install the units.[1][4]

OnePWR had originally placed the orders with Caterpillar anticipating the surge in data center demand. However, the supplier required additional financing to complete the massive transaction. This created the opening for Anthropic to step in and assume control of the hardware pipeline.[1][4]

The arrangement marks a strategic pivot for the AI developer. According to sources familiar with the matter, Anthropic initially planned to operate as a standard customer, buying electricity from OnePWR. The lab ultimately opted to take over the equipment orders directly, assuming control of the physical infrastructure.[1][5]

Bypassing the Grid Bottleneck

Artificial intelligence developers have historically relied on public utilities and third-party data center operators to supply electricity. However, surging demand for gas turbines has stretched manufacturing lead times years into the future, while grid interconnect queues routinely delay new facility launches.[1][4]

By taking over pre-existing Caterpillar orders rather than joining the back of the manufacturing queue, Anthropic secures equipment scheduled for delivery in the near term and over the next two years. The move locks in hardware priced before the recent surge in turbine demand.[1][4]

The financial metrics of the transaction underscore the premium placed on immediate availability. At $3.4 billion for 1.7 gigawatts of capacity, the purchase works out to approximately $2.0 million per megawatt, or roughly $2,000 per kilowatt, inclusive of the associated installation and engineering capabilities.[1][4]

The $2,000 per kilowatt price tag reflects the premium buyers are willing to pay to bypass the utility queue. Standard gas-fired generation capacity typically costs significantly less to deploy, but the immediate availability of the Caterpillar units justifies the markup for a lab racing to deploy its next model.[4]

Anthropic is not the only frontier lab using its balance sheet to cut out energy intermediaries. Earlier this year, Elon Musk’s xAI spent $2 billion to acquire mobile gas turbines to rapidly power its own data centers, while Google’s parent company Alphabet acquired energy developer Intersect in late 2025.[4]

"The transaction is the largest known single purchase of power-generation equipment by an AI lab, underscoring how the race for electricity is pushing tech companies deeper into the energy industry," the Financial Times noted, highlighting the sector's aggressive upstream integration.[1]

The Infrastructure Execution Test

Securing the turbines solves only the procurement phase of the power challenge. The inclusion of engineering, procurement, and construction capabilities in the OnePWR deal indicates that Anthropic must now manage heavy physical infrastructure deployment alongside its core foundation model training.[1][4]

Execution of the installation plan over the next 24 months will test the operational bandwidth of the AI startup. Without absorbing OnePWR's staff, Anthropic will rely heavily on contracted construction crews to integrate the 1.7 gigawatts of capacity into its data center footprint.[1][4]

The specific deployment locations for the gas-fired generators remain unconfirmed. As tech companies increasingly turn to fossil fuels to meet their immediate energy needs, gas-fired data center expansion faces mounting local regulatory scrutiny and environmental opposition in several key jurisdictions.[4]

AI

The physical buildout also faces secondary supply chain constraints. While Anthropic has secured the primary generators, the broader electrical infrastructure required to connect them—including heavy-duty transformers and specialized switchgear—remains in severe global shortage.[4]

The pivot to natural gas highlights a growing tension within the technology sector. While major AI developers maintain ambitious long-term climate targets, the immediate necessity of powering gigawatt-scale training clusters is forcing a reliance on carbon-intensive generation when renewable grid connections cannot be secured fast enough.[4]

Financing the AI Power Play

The $3.4 billion infrastructure commitment arrives as Anthropic prepares for a highly anticipated initial public offering. Financial markets expect the listing to launch as early as November 2026, potentially ranking among the largest technology debuts in history.[4]

The San Francisco-based company is reportedly aiming to raise between $80 billion and $100 billion through the public offering. Investment bankers have modeled a target valuation ranging from $1.25 trillion to $1.75 trillion, reflecting the massive capital requirements of frontier AI development.[4]

Anthropic currently holds close to $150 billion in cash and liquid investments. Against that balance sheet, the $3.4 billion generator purchase accounts for roughly 2.3 percent of its liquid reserves, demonstrating the scale at which the top-tier AI labs now operate.[4]

Despite these reserves, financing massive physical infrastructure commitments presents distinct debt market challenges. The capital-intensive nature of building proprietary power grids requires sustained borrowing, which alters the financial profile of a software-focused research laboratory.[4]

Investment analysts have warned that Anthropic could receive a junk credit rating rather than an investment-grade classification. That assessment reflects the inherent risks of the company's long-term, capital-heavy infrastructure buildout ahead of its public market debut.[4]

The Compute Arms Race

Anthropic’s aggressive move into power generation is directly tied to the escalating compute requirements for its future Claude models. The company has disclosed long-term computing commitments extending through 2045 that total nearly $900 billion.[4]

AI

The lab has historically relied on deep-pocketed partners to secure power, silicon, and data center land. Anthropic remains a major customer of SpaceX, with its monthly payments to lease computing capacity expected to reach $3.1 billion by the end of 2026.[4]

As foundation models scale, the primary bottleneck has shifted from acquiring graphics processing units to securing the raw electricity required to run them. A 1.7-gigawatt power reserve provides Anthropic with the physical megawatts needed when public grids cannot accommodate its training schedule.[1][4]

"AI developers typically rely on utilities and specialist power companies to provide power, but with demand for gas turbines and generators soaring... AI companies are using their financial muscle to cut out intermediaries," the Financial Times reported.[1]

By taking direct control of its energy supply chain, Anthropic ensures that its next-generation model development will not be stalled by utility interconnect queues. The OnePWR deal transforms turbine delivery into a balance-sheet asset directly tied to the lab's product roadmap.[1][4]

The transaction sets a new baseline for infrastructure investment in the artificial intelligence sector. As the race for artificial general intelligence accelerates, the ability to independently generate gigawatts of electricity has become as critical as the algorithmic research itself.[1][4]