Anthropic Spends $3.4B on Gas Turbines for 1.7 GW Power
AI lab purchases physical generating hardware directly from Caterpillar
Anthropic is spending $3.4 billion to acquire natural gas turbines capable of generating up to 1.7 gigawatts of electrical capacity. The transaction marks a sharp strategic pivot toward owning physical utility hardware as utility grid backlogs and severe generation constraints threaten AI model training and deployment timelines.
Key details
Under the direct agreement with equipment manufacturer Caterpillar, Anthropic will take delivery of natural gas generators originally ordered by equipment supplier OnePWR. OnePWR had reserved the manufacturing queue positions years earlier but was unable to secure private financing to finalize the procurement. Instead of waiting for third-party developers, Anthropic stepped in directly with $3.4 billion in capital to purchase the hardware.
The turbines will supply up to 1.7 gigawatts (1,700 megawatts) of dedicated behind-the-meter electric power. Historically, Anthropic has operated under an asset-light cloud leasing model. In its May S-1 regulatory filing, the company revealed extensive leasing commitments, including paying xAI approximately $1.25 billion per month through May 2029 for shared compute infrastructure. However, turbine manufacturing backlogs currently stretch into the 2030s, and regional transmission interconnection queues take up to seven years, forcing the frontier AI laboratory to secure captive power machinery.
Why this matters
Directly purchasing generating assets signals that compute availability is no longer bound solely by GPU supply, but by physical power access. By committing $3.4 billion directly to power generation equipment, Anthropic is competing directly against hyperscalers like Microsoft, Google, Meta, and Amazon for scarce electrical hardware. Without captive generation, AI labs risk sitting on billions of dollars in advanced processors that cannot be powered from overloaded public transmission grids.
Context
Across the technology sector, artificial intelligence developers are turning to off-grid fossil infrastructure to bypass utility constraints. Hyperscalers have contracted for tens of gigawatts of dedicated natural gas turbines and behind-the-meter generation plants. Projections indicate that U.S. data centers could consume more natural gas than Germany and Japan combined by 2035. This surging industrial appetite has revitalized domestic natural gas demand while complicating corporate net-zero greenhouse gas commitments.
Risks and open questions
Committing billions to gas turbine hardware exposes Anthropic to volatile fuel markets and severe environmental scrutiny. Natural gas prices are projected to rise significantly in key power regions by the end of the decade, which could drive operating costs far above original cloud leasing models. Additionally, deploying 1.7 gigawatts of dedicated gas turbines requires extensive air quality permits and local regulatory approvals, which have faced intense grassroots opposition and Clean Air Act litigation in states across the country.
What happens next
Anthropic will work to site and deploy the acquired Caterpillar turbines alongside its data center partners and emerging campus developments. The company is preparing for an initial public offering while managing over $400 billion in lifetime compute and infrastructure commitments. Regulators, local communities, and utility commissions will closely monitor permit filings to assess emissions impacts and ensure the private generation fleet complies with regional environmental standards.
Source: TechCrunch Published on AI Usage Global, author: AUG Bot
