Microsoft and Chevron Signal a New AI Reality: Big Tech Needs Big Energy

A 20-year natural gas deal in West Texas is not an outlier. It is the clearest sign yet that the AI buildout is pulling Silicon Valley into a direct partnership with the oil and gas industry — and that reliable power, not software, is now the bottleneck.

For the better part of two decades, the biggest technology companies in America sold themselves as the industry that would carry the economy past fossil fuels. Carbon-negative pledges. Renewable energy credits. Corporate sustainability reports with wind turbines on the cover.

Then came artificial intelligence, and with it a power appetite that no press release can satisfy.

In June, Chevron and Microsoft signed a 20-year power agreement for Project Kilby, a natural gas generation facility in Reeves County, West Texas, built to feed a Microsoft AI data center. Chevron says the project is designed to deliver roughly 2.67 gigawatts of capacity, built in phases, with power expected to begin flowing in 2028 pending final investment approval. CNBC pegged the consumption at the equivalent of about two million homes. A majority of the electricity will come from large gas turbines, and the facility is designed to run off-grid rather than wait in line for a utility interconnection.

That last detail is the story. Microsoft did not ask a utility for more power. It went to an oil company and bought a power plant's worth of output for two decades.

Why the AI boom broke the old model

Data centers are not ordinary customers. They want electricity every hour of every day, at industrial scale, with no tolerance for interruption. A campus that idles because the grid is strained is a campus burning capital on depreciating hardware.

The traditional route — request service, join the interconnection queue, wait for transmission to get built — now takes years. In an industry where competitors are spending hundreds of billions of dollars a year on capacity, years is an unacceptable answer.

So the hyperscalers stopped waiting. The emerging strategy is what the industry calls "behind-the-meter" generation: build the power plant next to the data center, own the fuel supply relationship, and skip the grid entirely at the start. Bessemer Venture Partners counted roughly 50 gigawatts of behind-the-meter gas generation projects announced in 2025 alone, and called it the dominant strategy for building new AI data centers.

Natural gas won that race for unglamorous reasons. The turbines can be ordered and installed on a commercial timeline. The fuel is already in the ground and in the pipelines. And in the Permian Basin, the gas is often being produced a short distance from where the electrons are needed.

This is a trend, not a one-off

Project Kilby is the cleanest example of tech-meets-oil, but it is not alone. The pattern is showing up across the industry in three distinct shapes.

The oil major as power supplier. Chevron and Microsoft is the template — an energy producer moving from selling fuel into commodity markets to building and operating dedicated generation for a single technology customer.

The developer-utility partnership. NextEra Energy and Google announced plans to co-develop three gigawatt-scale data center campuses in the U.S., each paired with directly connected generation. NextEra also disclosed agreements with Meta to finance roughly 2.5 gigawatts of new capacity.

Fossil generation with a carbon hedge. NextEra and ExxonMobil are pursuing a 1.2-gigawatt natural gas facility paired with carbon capture, anchoring a data center campus built to attract a hyperscale tenant. Exxon has said it is in advanced talks with power providers and tech companies on exactly this structure.

Google, Meta, Amazon and Microsoft have all pledged aggressive climate targets. All four are now associated with gas-backed data center projects. Microsoft has a separate gas project in West Virginia. Google has acknowledged a planned gas power partnership in Texas.

Meanwhile, the sector's own accounting is drifting. AI has been blamed in part for a 2.4% uptick in U.S. fossil fuel emissions last year, according to the Rhodium Group. Microsoft still targets carbon-negative operations by 2030 but now describes the effort as a marathon rather than a sprint. Google's 2030 clean power goal has been relabeled a "moonshot."

Nobody has formally abandoned anything. The targets have simply been demoted below the thing the market rewards right now, which is compute capacity delivered on schedule.

The honest energy argument

Here is the part that deserves to be said plainly, because most coverage dances around it.

Natural gas is not the permanent answer for the AI economy. It is the bridge — and right now it is the only bridge wide enough to carry the traffic.

The United States cannot build enough nuclear generation, high-voltage transmission, or long-duration storage in the next five years to serve the load the AI industry is bringing online. Advanced reactors are the right long-run answer for firm, low-carbon power at data center scale, and the interest from tech buyers is real. But reactors are measured in a decade of permitting, construction and commissioning. Turbines are measured in quarters.

That is not an argument against nuclear. It is an argument about sequencing. The choice on the table was never gas versus reactors. It was gas now versus a stalled buildout while the reactors get built. Microsoft and Chevron answered that question with a 20-year contract.

Whether gas is the right call in any specific location still depends on fuel supply, pipeline access, water availability, air permits, capital costs and grid conditions. Those tradeoffs are real and they should be argued locally, on the numbers. What is no longer arguable is the direction of the national trend. RBC Capital Markets forecasts U.S. data center gas demand reaching roughly 6.1 billion cubic feet per day by 2030.

The Permian is part of the AI power map

New Mexico does not have to host Project Kilby to be affected by it. Reeves County sits in the Permian Basin, a regional energy system that runs directly into southeastern New Mexico. The same producing formation, the same pipeline network, the same labor pool, the same service companies.

The strategic implication is what matters. AI investment is beginning to follow firm energy supply rather than tax incentives or corporate branding. Sites with gas production, pipeline capacity, available land and workable permitting become candidates. Sites without them do not, no matter how attractive the incentive package.

New Mexico has a local version of this already in the proposed Project Jupiter campus in the southern part of the state, where developers have pursued gas-backed on-site power and a Texas energy company has sought authorization for a pipeline connected to the project. The specifics remain unsettled. The pattern does not: data center developers are securing power first and arranging the rest of the project around it.

That reframes the state's position. The question is not whether New Mexico can influence a national scramble for electricity. It is whether New Mexico ends up with the jobs, tax base, industrial investment and durable infrastructure that come with being a power hub — or simply supplies the gas, the land and the water while the returns book somewhere else.

The bottom line

The AI race gets narrated in the language of chips, models and capital expenditure. Its next phase will be decided by turbines, pipelines, substations and fuel contracts.

Microsoft did not sign a 20-year deal with an oil company because the sustainability report changed. It signed because electrons have to show up every hour, and Chevron can make that happen before anyone else can. The companies that deliver dependable power are becoming as strategically important to artificial intelligence as the companies designing the silicon.

The green branding will survive. It just no longer gets to set the schedule.