Microsoft, Chevron Plan Huge Gas Data Center

Natural gas turbines solve that problem the old-fashioned way. They spin up on command, run continuously, and deliver a firm block of capacity that a developer can point at a single customer. By co-locating the plant next to the data center, Microsoft sidesteps the years-long queue to connect new load to the public grid, which has become one of the biggest bottlenecks in AI infrastructure today.

That convenience comes at a cost to Microsoft public image. The company pledged to eliminate its carbon emissions by 2030. A brand-new fossil-fuel plant running for two decades pushes hard against that promise, and Microsoft has spent years positioning itself as a sustainability leader.

How a co-located gas data center actually works

The mechanics are worth understanding because this model is spreading fast across the industry.

• Dedicated generation: Instead of buying power from a utility, the data center sits beside its own power plant. The electricity flows directly to the servers.
• Turbine mix: Large GE Vernova turbines carry the bulk of the 2.67-gigawatt load, while Solar Turbines units handle the remainder, likely for flexibility and faster ramping.
• Long-term contract: The 20-year power purchase agreement gives Chevron the financial certainty to build, and gives Microsoft a fixed, predictable power supply insulated from grid volatility.

One detail stands out. Solar Turbines, despite the green-sounding name, is a gas turbine maker, and the same brand appears in xAI unpermitted power plant near Memphis. The naming coincidence underlines how aggressively AI companies are bolting their own gas generation directly onto their compute.

Environmental groups are already sounding the alarm. The Environmental Integrity Project estimates Project Kilby could release more than 13 million tons of carbon dioxide, around 3,200 tons of criteria air pollutants, and roughly 278,000 pounds of hazardous air pollutants over its operating life.

The numbers behind the deal

Here is how the key figures break down.

For scale, 2.67 gigawatts is roughly the output of a large conventional power station, and here it is earmarked for a single corporate customer.

What happens next over the coming days

Expect the conversation to move quickly on three fronts in the next 24 to 72 hours.

First, scrutiny of Microsoft 2030 carbon pledge. Reporters and climate analysts will press the company on how a 20-year gas commitment squares with a goal that is now under four years away. Watch for a Microsoft response leaning on carbon capture, offsets, or future fuel-switching to soften the contradiction.

Second, the permitting fight. The pollution estimates from the Environmental Integrity Project are projections, not approved limits, and Texas air permits for a plant this size invite public comment and likely legal challenges. The xAI Memphis comparison, where a plant ran without full permits, will keep coming up.

Third, the competitive read-through. If Microsoft can self-supply 2.67 gigawatts of dedicated power, rivals racing to scale AI will feel pressure to strike similar deals. Look for analysts to frame this as a template, and for other hyperscalers and energy firms to signal their own co-located generation plans.

The bigger picture for AI and energy

This deal crystallizes a tension the AI boom can no longer hide. The technology promised efficiency and intelligence, but at scale it demands staggering amounts of always-on electricity, and right now the fastest way to get that electricity is to burn gas.

Microsoft is betting that the strategic value of guaranteed power outweighs the reputational hit. Whether that bet ages well depends on how fast cleaner firm power, like advanced geothermal or nuclear, can actually be deployed. For now, the AI gold rush is being powered by fossil fuels, and Project Kilby is the clearest proof yet.

Source: TechCrunch