UK's National Grid Bets $1.75B on U.S. AI Power Startup Joulent and Its 2.67-Gigawatt Texas Plant

There is a new kind of infrastructure deal reshaping the American energy landscape, and it does not originate from an AI lab or a hyperscaler. On July 1, 2026, Britain’s National Grid announced it would invest $1.75 billion for a 35% stake in Joulent LLC, a purpose-built energy company developing dedicated power generation for AI data centers that cannot wait for conventional utility interconnections.

The investment is the largest single bet any traditional electric utility has made on AI power infrastructure, and it frames the sector’s energy demand not as a problem to be managed but as a structured financial opportunity spanning decades.

The Project at the Center

Joulent’s flagship development, called Project Kilby, is a 2.67-gigawatt gas-fired power generation campus in Reeves County, West Texas. To put that number in context: 2.67 GW is roughly equivalent to the entire electricity output of a mid-sized coal plant, delivered to a single customer campus. Joulent is developing Kilby in a 50/50 partnership with Chevron Corporation, through its subsidiary Energy Forge, with GE Vernova supplying the turbines.

The buyer of that power is Microsoft, which is building a data center campus at the site and has signed a 20-year power purchase agreement (PPA) with Joulent. Microsoft President Noelle Walsh, commenting on the deal, described the arrangement as addressing a coordination gap between AI advancement and energy infrastructure: “AI and cloud are advancing at a pace that requires closer coordination between energy and infrastructure.”

Project Kilby is targeting first power delivery by 2028. By that point, it will serve as the primary electricity source for what is expected to be one of the largest AI computing campuses on the continent.

Why National Grid?

The more unexpected dimension of this deal is the identity of the investor. National Grid — a company whose primary business is operating the electricity transmission network in England and Wales, along with significant U.S. assets in New York and New England — is not an obvious candidate to bankroll a West Texas power plant. The company’s existing American footprint is concentrated in the northeastern states, thousands of miles from Reeves County.

National Grid CEO Zoë Yujnovich described the investment as “disciplined, partner-led investment in contracted critical infrastructure for the AI-driven large load economy.” The key word in that framing is “contracted”: the 20-year PPA with Microsoft means Joulent’s revenue is backed by a counterparty with a AAA-equivalent credit profile, producing a cash flow profile that resembles a long-duration infrastructure bond more than a speculative energy venture.

For National Grid’s shareholders, the investment thesis is that AI-driven power demand has created a new category of energy infrastructure — large, contracted, long-duration — that fits squarely within the financial profile of what an electric utility can and should own. The fact that Joulent is building the plant in Texas rather than National Grid’s traditional service territory doesn’t change the financial structure of the asset.

National Grid also brings something more than capital. Yujnovich noted that the company’s expertise in high-voltage networks, infrastructure project execution, and electrical equipment supply chains is part of what Joulent is buying with the partnership. Building a 2.67 GW campus involves solving problems that a newly founded power developer, however well-funded, does not have embedded institutional knowledge to address efficiently.

The Broader Interconnection Problem

Joulent’s founding premise — and the investment thesis behind this deal — is that traditional utility interconnection timelines are incompatible with the pace of AI infrastructure deployment. In most of the United States, connecting a large new power load to the grid can take five to ten years, navigating a queue of competing interconnection requests, transmission upgrade requirements, and regulatory reviews.

AI data centers are being designed and contracted on timelines measured in months. Microsoft, Amazon, Google, and Meta have announced hundreds of billions of dollars in AI infrastructure commitments over the past 18 months, but the actual power to run those campuses has emerged as the primary constraint on delivery.

Joulent’s approach is to bypass the interconnection problem by co-locating generation with load. Rather than drawing power from the grid, Project Kilby generates electricity on-site, feeding directly into the Microsoft campus without requiring a high-voltage transmission connection subject to public queue and regulatory review. The state of Texas facilitates this model more readily than most other jurisdictions, because the state’s ERCOT grid operates largely independently of federal oversight and has a relatively permissive framework for private power arrangements.

The Chevron partnership matters here as well. West Texas is one of the most prolific natural gas producing regions on earth. Joulent, through Energy Forge, can access gas supplies at prices and on timelines that a company without Chevron’s upstream relationships could not. The turbines from GE Vernova are gas-fired, but Joulent’s integrated architecture also includes battery storage and renewable generation capacity — a hedge against eventual carbon pricing and a signal that the company is not building a stranded-asset bet on unabated natural gas.

Capital Racing to Meet Demand

The Joulent deal is not an isolated event. It is the latest in a series of major capital commitments flowing toward AI power infrastructure as investors recognize that the bottleneck in the AI build-out is not silicon or software — it is electrons.

Microsoft itself has announced power supply agreements aggregating more than 10 GW globally over the past year. Amazon has committed to purchasing renewable energy at a scale that, if delivered, would make it the world’s largest single buyer of clean power. Google has revived nuclear power purchase agreements for the first time in decades, signing PPAs for advanced small modular reactors.

National Grid, for its part, has been explicit about the scale of what it sees coming. The company has said it expects to connect more than 10 GW of data center demand across its U.K. and U.S. service territories within five years — a figure that would dwarf the current data center load on its network.

Chris James, Joulent’s founder and CEO, summarized the strategic context in straightforward terms: “This investment strengthens Joulent’s ability to deliver reliable, large-scale power on the timelines AI infrastructure and advanced industry now requires.”

What This Signals to the Industry

The National Grid-Joulent deal is a landmark in the institutionalization of AI power infrastructure as an asset class. When a regulated utility with a $40+ billion market capitalization commits $1.75 billion to a startup with a single development project, it sends a signal that the underlying demand trajectory is credible enough to anchor long-duration balance sheet commitments.

For AI companies and their investors, the more consequential question is how many more Project Kilbys need to be built — and how quickly. Current industry estimates suggest that the aggregate AI power demand announced by U.S. hyperscalers alone, if fully delivered, would require adding the equivalent of several large states’ current electrical generation capacity within a decade.

The United States has not built infrastructure at that pace since the mid-twentieth century. Whether the combination of private capital, utility expertise, and tech company PPAs can substitute for the public infrastructure programs of that era remains to be seen. What the National Grid-Joulent deal confirms is that at least some of the largest traditional infrastructure operators believe it can.

Final investment decision on the Joulent stake is expected before the end of 2026. Project Kilby’s 2028 power delivery target is, by the standards of energy infrastructure, aggressive — but by the standards of AI compute demand, merely essential.