The Power Behind the Prompt: Chevron’s Gas Plant for Microsoft AI
A 2.67-gigawatt gas plant built to feed one thing: a Microsoft AI data center in West Texas.
TL;DR — Chevron and Microsoft signed a 20-year power agreement for "Project Kilby," a ~2.67-gigawatt off-grid natural-gas plant in West Texas dedicated to a Microsoft AI data center, with a ~$7B first phase and first power expected in 2028.
Every AI answer starts as electricity. On June 22, 2026, Chevron and Microsoft revealed where a lot of it will come from.
The deal
Chevron (via subsidiary Energy Forge One) will build Project Kilby, a roughly 2.67-gigawatt off-grid natural-gas plant on 2,000+ acres in Reeves County, West Texas, under a 20-year power purchase agreement with Microsoft for a dedicated AI data center. The first phase is valued near $7 billion, with first power expected in 2028 and generation led by GE Vernova turbines. The power is dedicated and not connected to the grid.
| Detail | |
|---|---|
| Capacity | ~2.67 GW |
| Term | 20-year PPA |
| First phase value | ~$7B |
| First power | 2028 |
| Grid connection | None (off-grid, dedicated) |
What they said
"The rapid growth we’re experiencing in AI and cloud... requires energy infrastructure that can scale quickly and reliably." — Noelle Walsh, President of Cloud Operations + Innovation, Microsoft
Why it matters
- Power is the new bottleneck. AI capacity is increasingly limited by electricity, not chips.
- Oil majors pivot to electrons. Chevron becomes a direct supplier to the AI build-out.
- Off-grid is the workaround. Dedicated generation sidesteps strained public grids and slow interconnection queues.
FAQ
What is Project Kilby?
A roughly 2.67-gigawatt off-grid natural-gas power plant Chevron will build in West Texas under a 20-year agreement to power a Microsoft AI data center. The first phase is valued near $7 billion, with first power expected in 2028.
Why build a dedicated, off-grid plant?
AI data centers need large, reliable power quickly. Building dedicated generation that is not connected to the public grid avoids strained grids and long interconnection waits — letting Microsoft scale compute on its own timeline.
Sources
Image: Chevron logo by Chevron Corporation — Public domain, via Wikimedia Commons.
← Back to all posts