It sounds like an unlikely pairing: an oil giant building a power plant for a tech company. But that’s exactly what’s happening — and it’s a stark illustration of just how much electricity the AI boom is consuming.
Chevron Corp. (CVX) recently announced a 20-year power purchase agreement with Microsoft Corp. (MSFT) to build a large natural gas-fired plant in West Texas, dedicated exclusively to powering Microsoft’s AI data center. Dubbed Project Kilby, it’s one of the largest co-located gas power and data center developments in the U.S.
A full-scale power plant built for a single company’s data center. Just how hungry are AI facilities for electricity?
When fully built out, Project Kilby will deliver approximately 2.67 gigawatts of capacity.
For perspective: that’s enough to power roughly 2 million homes. And it’s all earmarked for one Microsoft data center campus.
The project will be built in phases, using a modular approach to add capacity incrementally as Microsoft’s data center needs grow. Chevron expects to make a formal Final Investment Decision by the end of this year, with the project carrying a price tag of around $7 billion. First power delivery is targeted for 2028, with full buildout continuing into the 2030s.
GE Vernova (GEV) will supply most of the gas turbines and associated electrical infrastructure, with additional capacity from Solar Turbines, a Caterpillar Inc. (CAT) subsidiary.
This isn’t an isolated case. AI data center power demand is growing at a staggering pace.
According to BloombergNEF, U.S. data center power consumption is on track to double to 77 gigawatts by 2030. Let that sink in: the entire country’s data center electricity use is set to double within roughly four years, and AI is the primary driver.
Training and running large language models requires massive computing power, and computing power runs on electricity. As AI applications move from research labs into production, the demand curve only steepens. That’s why Microsoft, Google and Amazon are all racing to “lock in” power supplies — securing energy sources ahead of time to underpin their AI growth strategies.
If the power grid already exists, why would Microsoft need Chevron to build a dedicated plant?
The answer comes down to reliability and grid impact. The advanced compute requirements of AI and cloud operations demand extremely stable, dispatchable electricity. Co-locating a new, large-scale generation facility with the data center delivers reliable power directly to Microsoft while mitigating the impact on the regional grid that consumers depend on.
Put differently: AI’s appetite has grown so large that it’s starting to need its own kitchen.
For Chevron, the deal represents more than just a single power plant — it’s an entirely new business platform.
Chevron will supply the facility with locally produced natural gas from its vast Permian Basin resource base. By converting that gas into electricity under a 20-year contract, Chevron generates more predictable cash flow that’s partially insulated from the commodity price volatility of its traditional oil and gas operations — supporting resilient shareholder returns, including continued dividend growth. The company expects mid-teen returns on its investment.
And Project Kilby is likely just the beginning. When Chevron formed its partnership with Engine No. 1 early last year, the target was to deliver up to 4 gigawatts of projects — with room to expand beyond that. Given the massive power needs of AI data centers, there should be no shortage of opportunities.
So back to the original question: just how much power do AI data centers need?
Enough that oil companies are building dedicated power plants to feed them. Enough that America’s total data center electricity use is set to double in a decade. Enough that tech giants are now treating “energy security” as a core strategic priority.
Project Kilby is a microcosm of a bigger shift. As AI compute demand keeps exploding, the link between energy and computing will only grow tighter. Looked at that way, the next phase of the AI race might not be won with better models or faster chips — it might be won by whoever can secure the most, cheapest, most reliable electricity.
