I didn’t pay much attention to AI electricity consumption until a departmental meeting at work in December 2025. I’m not close to cloud infrastructure, but somebody had asked a question about strategies for shifting cloud computing out of Western Europe and up to Northern Europe. In the response, I heard something that sounded strange to me: the particular cloud provider was running out of electricity in Western Europe. Really? Is it possible to run out of electricity in Western Europe? Apparently, it is.
There’s a headline doing the rounds that makes it sound as if Microsoft has started stockpiling barrels of crude. It hasn’t. To clear up the confusion straight away: Microsoft is not buying physical oil to burn or refine. What it has done is sign a 20-year power purchase agreement with Chevron, the oil and gas major, in a development the two companies are calling Project Kilby. Chevron’s wholly owned subsidiary, Energy Forge One, will develop a co-located power facility in West Texas that delivers dedicated electricity to a Microsoft-operated data center. The plant draws on natural gas from Chevron’s existing Permian Basin production, and at full build-out Project Kilby is designed to generate approximately 2.67 gigawatts, roughly the electricity needs of two million homes, rolled out in phases.
That’s the literal substance of the deal. The more interesting part is what it signals. I’d argue this is one of the clearest data points yet on how the AI build-out is actually being fuelled, and it cuts against a lot of the tidy narratives we’ve been told. Four things stand out to me.
1. Bypassing a grid that can’t keep up
AI infrastructure is staggeringly power-hungry, and the usual route, buying from the local utility, has hit a wall. Connection queues for large loads now stretch years, and the shared transmission system simply isn’t expanding fast enough to absorb a fleet of new data centers. Microsoft’s answer here is to step around the problem entirely. The plant is built to run independently of the Texas grid, supplying Microsoft directly rather than drawing on ERCOT, and that structure sidesteps the multi-year interconnection queues now slowing large data center projects. This is the “behind-the-meter” play: generate your own power on site, on your own timeline. Power is expected to begin flowing in late 2028, with development extending through the 2030s. When the public grid becomes the bottleneck, the deepest-pocketed players just build their own.
2. The practical reality versus the climate pledge
This is the part I find hardest to look away from. Microsoft has committed to becoming carbon negative by 2030, and for years its strategy leaned on wind and solar. But AI workloads need continuous, 24/7 baseload power, the kind intermittent renewables can’t guarantee without enormous and expensive battery backup. So here we are: a fossil-fuel deal that locks in two decades of gas generation at a time when Microsoft and its peers face pressure to align AI growth with climate targets. I don’t read this as hypocrisy so much as revealed preference. When the immediate physical demands of the AI race collide with a green timeline, the timeline is what bends. That’s worth being honest about rather than pretending otherwise.
3. Turning “waste” gas into AI gold
Here’s the bit that makes the economics sing for Chevron. Drilling for crude in the Permian throws off a vast volume of associated natural gas as a byproduct, and there’s often nowhere to put it. Volumes in the Permian routinely outrun what regional pipelines can carry, forcing operators to flare the excess, a dynamic that depresses local gas prices. Flaring is, quite literally, burning money into the sky. Project Kilby flips that problem on its head. Chevron said that dynamic gives the project a competitive cost advantage, turning a regional supply headache into dispatchable electricity sold straight to a tech giant. It’s part of Chevron’s effort to build revenue streams less directly tied to oil and gas commodity prices. Cheap stranded gas in, premium AI power out.
4. A new model for tech and oil
Step back and you see a genuinely symbiotic arrangement forming. Microsoft secures firm, reliable power to stay in the AI race. Chevron secures a multi-billion-dollar, two-decade revenue stream, cash flow it expects to be independent of oil and gas price cycles, targeting mid-teen returns, and, not incidentally, a closer relationship with the AI tools it increasingly uses to optimise its own exploration and drilling. The trade-off is emissions, and we should expect more of these dedicated gas-for-data-center deals in the Permian and other gas-rich basins as AI load outpaces what congested grids can add.
So the takeaway isn’t that Microsoft went shopping for oil. It’s that the AI boom has quietly become an energy story, and the people best positioned to supply firm power at speed are the incumbents of the carbon economy. The grid couldn’t deliver, the renewables couldn’t guarantee round-the-clock, and Big Oil could. Project Kilby is what that compromise looks like in concrete and steel, and I suspect it’s the first of many.
The Nigerian Angle
There is an obvious Nigeria angle here. Nigeria is one of the countries where this story should sting the most. We flare gas and then complain about lack of electricity. We burn value into the sky while homes run petrol generators, factories run diesel, and businesses price unreliable power into everything they do.
What Chevron and Microsoft have shown is not simply that gas is useful. We already know that. What they have shown is the power of matching stranded gas with a bankable customer. In Texas, the customer is Microsoft. In Nigeria, it could be a data centre, an industrial park, a cement plant, a fertilizer plant, a refinery, a port, a manufacturing cluster, or even a group of telecom towers. The principle is the same: capture gas that would otherwise be wasted, generate power close to source, and sell it directly to someone who can pay.
That is the real missing link in Nigeria. We often discuss gas flaring as an environmental scandal, which it is. But it is also an economic scandal. It is energy poverty sitting beside wasted energy. It is not just that Nigeria lacks electricity; it is that we have failed to organise fuel, infrastructure, finance, regulation and credible demand into a working system.
Project Kilby should therefore not be read in Nigeria as an American curiosity. It should be read as a challenge. If Big Tech can make Big Oil turn stranded gas into AI power in West Texas, why can Nigeria not turn flared gas into industrial power in the Niger Delta, Port Harcourt, Warri or other productive clusters? The answer is not physics. The answer is institutions, execution and bankability.
Sources
- Chevron — Chevron Signs 20-Year Power Agreement with Microsoft for West Texas Data Center
- CNBC — Chevron to fuel massive Microsoft data center in Texas using natural gas
- World Oil — Chevron signs 20-year Microsoft power deal for West Texas AI project
- mGrid — Chevron, Microsoft Sign 20-Year Deal for 2.67 GW Off-Grid Gas Plant
oyebode.com 