Microsoft and Chevron sign a 20-year agreement; a 2.67-gigawatt AI power project breaks ground in Texas.

As demand for AI computing power continues to surge, collaboration between tech giants and energy majors is deepening. According to Bloomberg, on June 22 local time, Chevron announced it had signed a 20-year long-term power supply agreement with Microsoft to provide natural gas–fueled electricity for Microsoft’s planned large-scale data center campus in West Texas. The project’s total installed capacity is expected to reach 2.67 gigawatts (GW), sufficient to meet the electricity needs of over 530,000 households—and potentially ranking among the largest AI data center–dedicated power projects in the United States. Against a backdrop of increasingly strained grid resources, technology companies are increasingly adopting the ‘self-built power generation + data center’ model to secure long-term, stable energy supplies—responding to the ever-rising computational demands of the AI era. This trend not only reflects AI infrastructure’s enormous appetite for energy but also signals a deeper symbiotic relationship forming between traditional energy enterprises and the technology industry. The AI Computing Race Fuels an 'Energy Scramble' Microsoft is accelerating the global expansion of its data center footprint to meet the explosive growth in compute demand driven by generative AI. Microsoft previously stated it plans to more than double its data center capacity over the next two years to maintain its competitive edge against Google and Amazon in the AI arena. This rapid data center expansion has made electricity one of the most critical resources across the AI value chain. Bloomberg forecasts that by 2030, total U.S. data center electricity consumption will double from current levels to 77 GW. Faced with escalating electricity demand, multiple U.S. regions have already begun experiencing grid capacity constraints and rising electricity prices—sparking concerns among regulators and consumers alike. Against this backdrop, Microsoft has opted to partner directly with energy producers to secure long-term, stable, and predictable power sources. A 2.67-GW Natural Gas Power Plant Dedicated to Data Centers Under the agreement, Chevron will jointly develop the 'Project Kilby' power generation initiative with investment firm Engine No.1. Located near Pecos, Texas, the project is scheduled to begin operations in 2028 and gradually scale up to 2.67 GW. It will deploy gas turbines from GE Vernova and draw on abundant natural gas resources from the Permian Basin. Unlike conventional models, this power plant will supply electricity exclusively to Microsoft’s data center campus—without connecting to the public grid or relying on local utility companies for transmission and distribution. Jeff Gustavson, Chevron’s Head of New Energy, stated: 'Consumers are already feeling the impact of rising electricity demand—and this project was designed from the outset to avoid placing additional strain on the local grid.'

A new challenge for AI data center power consumption: power fluctuations are causing frequent damage to critical equipment, putting trillion-dollar investments to the test.

AI’s enormous electricity demand is widely known—but a more hidden, thornier problem is now surfacing: extreme power fluctuations in data centers are damaging critical equipment, undermining facility reliability, and injecting instability risks into the broader grid. According to a Bloomberg report dated August 6, core systems in AI computing facilities—including batteries, generators, and cooling units—are failing or wearing out prematurely due to abnormal electrical surges. At xAI’s Colossus supercomputing facility in Memphis, Tennessee, gas turbines have developed cracks; similarly, smaller data centers across the UK have reported turbine cracking. Some newly installed voltage-stabilizing batteries require replacement within weeks—or at most months—under high-intensity operation. The market impact of this issue is already emerging. A source familiar with data center financing told reporters that actual uptime at certain facilities has dropped to roughly 80%—far below the design expectation of year-round, uninterrupted operation. If unresolved, this problem could directly impact investors in these projects within the next 12 to 24 months. Meanwhile, the North American Electric Reliability Corporation (NERC) issued a rare Level-3 alert earlier this year, urging large data centers to immediately address grid stability risks. Root Cause of Power Fluctuations: 'Millisecond-Level' Shocks from GPU Clusters AI data centers exhibit fundamentally different electricity consumption patterns compared to traditional ones. While conventional facilities draw relatively steady power, AI workloads—especially during model training—trigger hundreds of thousands of GPUs to start and stop synchronously, causing electricity demand to surge or plummet dramatically within milliseconds. Shannon Miller, Founder and President of Mainspring Energy Inc., vividly described the phenomenon: 'In a 1-gigawatt data center—the size of Boston—half the load may flicker on and off every few seconds.' Some AI campuses currently under planning in Texas and the U.S. Midwest exceed 5 GW in scale, with average power draws approaching that of New York City as a whole. Drew Baglino, former Tesla executive and Founder of Heron Power Electronics Co., noted that instantaneous power demand at AI data centers can spike more than 50% above design capacity—'a 1-GW facility might momentarily draw 1.5 GW.' His company is developing power fluctuation management hardware for NVIDIA’s next-generation servers, slated for release in 2027. Amber Villegas-Williamson, Chief Consultant (UK) at Uptime Institute, likened the stress to aggressive driving: 'Like constantly revving an engine to redline—much more wear-and-tear than cruising at a steady speed.' Equipment Damage Is Already Real: From Cracks to Arc Flash According to interviews conducted by Bloomberg with more than thirty power professionals across the U.S. and Europe, the physical stress AI data centers impose on equipment is already evident. Multiple insiders reported that crankshafts in small natural-gas internal combustion engines used in data centers have fractured. At xAI’s Colossus facility, gas turbines developed cracks; operators subsequently installed batteries to smooth out power fluctuations and reduce turbine loading.

Chevron Keeps Microsoft Data Center Timeline Despite Texas Pause

1 day ago

Chevron Corp. maintains its timeline to supply natural gas for a Microsoft data center in West Texas, aiming for first power in 2028 despite Governor Abbott’s permit suspension. CFO Eimear Bonner confirmed that permitting and financing remain on track.

bloomberg

2026-06-26

Chevron eyes more deals to power US data centers

2026-10-30

Chevron Keeps Microsoft Data Center Timeline Despite Texas Pause

Chevron Keeps Microsoft Data Center Timeline Despite Texas Pause

FAQ