Claiming "Green" While Relying on Natural Gas
It has emerged that Microsoft is partnering with energy giant Chevron to advance one of the largest natural gas-powered data center projects in the United States. According to TechCrunch, the two companies have signed a 20-year Power Purchase Agreement (PPA), under which electricity will be supplied directly from a newly built natural gas power plant to the data center.
The main reason this news has sent ripples through the industry is that Microsoft is a company that has publicly committed to becoming carbon negative (removing more CO2 than it emits) by 2030. Natural gas is considered relatively clean among fossil fuels, but it still releases CO2 when burned. A long-term contract spanning 20 years means that carbon emissions will effectively be locked in for at least that period.
Why Natural Gas, and Why Now?
The backdrop is the "power wall" created by the generative AI boom. Training and inference workloads for large language models (LLMs) consume electricity on an entirely different scale compared to conventional cloud workloads. Renewable energy alone cannot keep pace with this surging demand. Solar and wind are intermittent energy sources subject to weather conditions, and data centers that operate 24 hours a day, 365 days a year require stable "baseload power" — sources capable of delivering a constant supply at all times.
Nuclear power has attracted attention as an ideal solution, but building new plants takes decades. Microsoft is also in negotiations to restart existing reactors slated for decommissioning, but that alone cannot meet demand. As a result, natural gas has emerged as the practical near- to mid-term answer.
Nvidia Strikes Back with "Water-Free" Data Center Design
As a response to these power and environmental challenges, Nvidia is taking a different approach. According to The Verge, the company claims that by making entire data centers liquid-cooled in the reference design specifications for its next-generation "Rubin" GPU architecture, it has "reduced water consumption to near zero and significantly cut power consumption."
Conventional air-cooled data centers consume vast amounts of cooling water. Particularly when built in arid regions, they have repeatedly faced backlash from local residents and environmental groups for straining regional water resources. Nvidia's liquid cooling approach dramatically improves cooling efficiency by capturing heat directly from the chips, eliminating water evaporation losses in the process.
However, as The Verge also points out, this design only addresses problems in the "operational phase." Challenges remain regarding the resources consumed and carbon emitted during the construction of data centers, as well as the environmental impact of manufacturing large volumes of AI chips.
Greenwashing, or a Pragmatic Transitional Response?
The Microsoft-Chevron deal has already drawn criticism from environmental groups and investors. At the same time, some voices within the tech industry offer a defense, calling it "a realistic bridge until renewable energy can catch up."
To this writer, who has observed cloud infrastructure in Silicon Valley for many years, the essence of this issue is not a simple "environment vs. technology" conflict. It is the "visualization of a gap" — the gap between the speed at which AI capabilities are being deployed into society, and the speed at which the energy infrastructure to support them is being developed.
With Groq confirming a $650M funding round (reported by TechCrunch the same day) and the AI chip market as a whole in full swing, power demand for data centers will only accelerate further. Microsoft's choice should not be seen as an extreme outlier, but rather as a symbol of a structural problem facing the entire industry.
Conclusion
Microsoft and Chevron's 20-year contract has brought into the open the trade-off that lies between the expansion of AI infrastructure and environmental goals. While Nvidia advocates for improved "operational efficiency" through liquid cooling, the question of the power source itself remains unresolved. In 2026, the question of "where energy comes from" is becoming just as significant a business challenge for AI companies as any technical problem they face.