Anthropic spends $3.4B on private power generators to bypass grid delays

Bypassing the public grid

Anthropic, the company responsible for the Claude AI model, is taking a direct approach to solving its energy needs. The company recently finalized a $3.4 billion deal to acquire 1.7 gigawatts of gas generators. This massive amount of electricity is sufficient to power more than one million homes.

The primary motivation for this move is the difficulty of connecting new data centers to the existing U.S. power grid. Reports from Hanwha Data Centers indicate that new facilities attempting to connect to the public grid might face delays of five to seven years. By creating its own power source, Anthropic can begin operations much sooner.

This strategy is becoming a trend among major AI players. For example, Elon Musk's xAI used a similar tactic for its Memphis "Colossus" site earlier this year. That company spent roughly $2 billion on mobile gas turbines to ensure its site had the necessary power to run its hardware.

The deal structure

The transaction was completed in late September through a San Antonio-based firm called OnePWR Solutions. Anthropic did not purchase the firm itself, nor did any employees change hands between the two companies. Instead, Anthropic paid for the specific equipment and the engineering work required for the installation.

OnePWR Solutions had already placed significant orders with Caterpillar for these generators. However, Anthropic stepped in and took over the entire existing order book. This arrangement ensures that the generators will be installed specifically at Anthropic's data center sites located in Texas.

Caterpillar is a major industrial supplier that manufactures diesel engines, natural gas generators, and gas turbines. Because of this deal, the company earned a significant amount of revenue from the AI sector. Industry blog FourWeekMBA estimated that Anthropic's order alone represented about 99% of Caterpillar's third-quarter power generation sales.

A new bottleneck for AI

For several years, the primary focus of the AI trade was the supply of specialized chips. Companies like Nvidia were the center of almost every conversation regarding how to build and scale large language models. However, the industry is now realizing that finding enough electricity to run those chips is the next major hurdle.

As AI models become more complex, they require more constant and massive amounts of power to train and operate. This has shifted the spotlight toward power generation firms and industrial suppliers as vital parts of the AI supply chain. These companies are now just as essential as the hardware manufacturers.

Melissa Busen, a senior vice president for electric power at Caterpillar, has noted the importance of this infrastructure. She described gas generation systems as "core infrastructure for data centers" in comments shared for a project called Monarch. This highlights how the physical reality of energy is shaping the future of digital intelligence.

The shift to private power

The move toward private power generation represents a significant change in how big tech companies plan their infrastructure. Rather than relying on public utilities, these companies are becoming their own utility providers to maintain a competitive edge. This allows them to control the timing and reliability of their energy supply.

This trend suggests that the geography of AI development will be dictated by where companies can secure reliable power. By investing in gas generators, Anthropic is securing a way to run its data centers independently of local grid limitations. This move helps ensure that their training cycles are not interrupted by infrastructure bottlenecks.

As more companies follow this path, the role of industrial giants like Caterpillar will continue to grow. The AI trade is no longer just about software and silicon; it is increasingly about the heavy machinery and fuel required to keep those systems running. This transition marks a new era where energy availability is the ultimate gatekeeper for technological growth.