Anthropic secures 1.7GW of gas generators for AI power

Securing the power supply

Anthropic has secured commitments for 1.7GW of heavy-duty gas-fired generators to support its growing AI needs. The company purchased these commitments from OnePWR Solutions for a total of $3.4 billion. These specific engines were originally ordered by OnePWR two years ago, before market prices experienced a significant surge.

The deal highlights a trend where AI labs are treating power as a primary procurement line. Because the public grid is often insufficient, companies are increasingly buying power as physical equipment. This allows them to create dedicated energy sources that are not solely dependent on existing municipal infrastructure.

While the Financial Times reported the deal, it did not specify exactly where these generators will be installed. However, the scale of the purchase suggests a massive expansion of the company's physical footprint. This move ensures that Anthropic can maintain the high energy loads required for large-scale model training.

A tale of two strategies

The approach taken by Anthropic in the United States differs significantly from methods used in China. According to SemiAnalysis, Envision built 420MW of wind farms for its Galaxy AI campus in Inner Mongolia. This wind capacity is about 1.75 times the 240MW load of the campus's first two phases, providing a buffer for the facility's energy needs.

By our arithmetic, Anthropic’s 1,700MW of gas-fired generators is about four times the 420MW of wind capacity built by Envision. These two examples show two distinct ways of securing power for AI campuses: the U.S. cases favor gas turbines and generators, while the Chinese case relies on sizing wind farms above the actual load. Both strategies are designed to ensure the campuses have dedicated power.

The Chinese route also involves a more integrated model of construction. SemiAnalysis notes that the company making the turbines and batteries also builds the campus itself. This contrasts with the U.S. model where companies like Anthropic buy specific equipment from established suppliers like Caterpillar to meet their energy requirements.

The equipment supply chain

The demand for power generation equipment is creating a massive boom for industrial manufacturers. Caterpillar reported $3.421 billion in Power Generation sales for the third quarter of 2026. This figure represents a 30% increase and was driven primarily by applications in the data center sector.

The scale of this demand is reflected in the massive backlog of orders currently held by major manufacturers. As of September 30, Caterpillar's firm backlog was approximately $78.1 billion. This indicates that the bottleneck for AI development is shifting toward the physical components needed to generate and distribute electricity.

Other companies are following similar paths to secure their energy future. For example, a city notice shows SpaceX proposes to construct a gas-fired power plant and super-compute facility on about 444 acres. While the specific megawatt size was not stated, the project involves a significant investment in initial phases to power its operations.

Resistance and grid limits

As AI companies rush to build out their power infrastructure, they are facing increasing pushback from various levels of government. In the United Kingdom, 315 data centers are currently queueing to connect to the grid. This represents 73GW of demand against a national peak of 45GW, meaning the queue is 162% of the peak.

Local and state governments are also taking action to manage these demands. In Texas, the governor's office directed the state's environmental regulator to halt all permits sought by data centers until an audit of the ERCOT grid is completed. This move creates a temporary pause for developers looking to expand in the region.

Public sentiment is also influencing the pace of development. In Scotland, a coalition of 75 organizations planned a march to protest the construction of AI data centers. Meanwhile, the city of Bloomington voted to ban the distribution of superintelligent AI, showing that the pushback can come from both infrastructure concerns and the technology itself.

What comes next for AI labs

For AI labs, the primary challenge is now navigating the structural bottleneck of electrons. The move by Anthropic to buy gas generators shows that companies are willing to spend billions to secure reliable power. This shift toward

What comes next for AI labs

For AI labs, the primary challenge is now navigating the structural bottleneck of electrons. The move by Anthropic to buy gas generators shows that companies are willing to spend billions to secure reliable power. This shift toward private power generation allows them to bypass the limitations of aging public grids.

The competition for equipment will likely intensify as more companies seek to secure their own energy sources. With Caterpillar's backlog already in the tens of billions, the ability to procure generators and turbines quickly will be a key competitive advantage. Labs that can secure power faster will be able to scale their models more effectively.

Observers should watch for how these private power projects interact with local regulations. As seen in Texas and the UK, the tension between the need for AI infrastructure and the limits of the existing grid will continue to create hurdles. The success of these projects will depend on balancing rapid deployment with environmental and logistical constraints.