National Grid Invests $1.75 Billion in Joulent, an AI Energy Firm

National Grid announced on July 1 a $1.75 billion investment in Joulent, a Houston startup that builds dedicated power plants for artificial intelligence data centers. The transaction, executed through its commercial arm National Grid Ventures, gives it a 35% stake in the firm.

Joulent emerged from stealth on June 22 after three years of incubation by Engine No. 1, the investment firm that gained prominence in 2021 by securing three seats on ExxonMobil's board. Its model, called Across-the-Meter, involves building natural gas generation plants, supplemented with batteries and solar power, directly adjacent to data centers and signing long-term power purchase agreements with operators. The strategy addresses a concrete bottleneck: interconnection queues at operators such as ERCOT (Electric Reliability Council of Texas) can stretch for years, while data center electricity demand grew 17% in 2025, compared to 3% for global demand, according to Reuters.

The anchor project is Project Kilby, a 2.67 GW plant in west Texas developed as a 50-50 joint venture with Chevron through the subsidiary Energy Forge One. The facility, which will use GE Vernova turbines already secured, will supply a data center campus operated by Microsoft under a 20-year PPA (power purchase agreement). First energy delivery is scheduled for 2028. The final investment decision is expected by end of 2026, and the firm projects positive free cash flow by the early 2030s. Brian Boland, Joulent's chief financial officer, described the model to Latitude Media as a return to the dedicated plant scheme that once powered refineries and steel mills in the twentieth century, now applied to the AI era.

The investment is incremental to National Grid's £70 billion capital plan through 2031. JP Morgan analysts estimate returns will exceed the 9-10% return on equity the company earns from its regulated networks. The move signals that electric infrastructure capital is willing to bypass the traditional interconnection queue when compute demand requires it.

This article was written with artificial intelligence assistance from verified sources and reviewed by a human editor before publication.