AI’s power hunger is rewriting climate promises
Amazon’s GW Ranch project is a massive natural-gas power plant built beside an AI-focused data center in West Texas, designed to secure dedicated electricity for artificial intelligence workloads while exposing the scale of emissions, water use, and regulatory gaps hidden behind glossy corporate climate pledges.
The core problem is stark: AI data center emissions are now big enough to justify building an entire fossil fuel power plant for a single site. Amazon is developing the GW Ranch gas facility next to its new data center in Pecos County, planning to use most of the electricity on-site rather than sending it to the wider grid. This is not a side project; it is AI infrastructure baked straight into fossil fuel expansion. The choice signals that when forced to pick between more compute and less carbon, hyperscalers are choosing compute first and tidying up the narrative later.
According to one report, GW Ranch has received a permit allowing up to 33 million tons of carbon dioxide emissions a year from 35 gas turbines rated at around 7.65 gigawatts of capacity, illustrating how a single AI data center can rival the dirtiest power plants on Earth.
A 7.65GW gas plant built for AI, not homes
On paper, GW Ranch looks like classic heavy industry: 35 natural gas turbines and around 7.65GW of capacity, with a state permit to emit up to 33 million tons of CO₂ annually. In practice, this is not being framed as a public utility. Amazon has bought the property and plans to buy the electricity, with much of the output expected to go straight into its data center racks rather than nearby homes and businesses.
The permitted emissions ceiling is stunning. That allowance exceeds the limit for the largest coal plant in the United States and would make GW Ranch the single biggest power-plant source of CO₂ in the country if fully used. Supporters can argue that this is only a maximum, not a guarantee of actual emissions. But permits are political documents: they reveal what regulators are willing to tolerate. Granting a 33‑million‑ton budget to serve one AI hub shows how lightly environmental guardrails are being applied to digital infrastructure.
If a company can lock in that level of fossil fuel capacity for one campus, it sets a precedent: compute-first facilities with their own fossil fuel power plants become not an exception, but a template.
Climate pledges versus the reality of AI data center emissions
Publicly, Amazon insists nothing has changed about its climate ambitions. The company helped found a coalition promising net‑zero emissions by 2040 and has repeated that it remains committed to cutting carbon across its operations. Privately, its build-out choices tell a different story. Emissions have climbed in recent years, and the company itself links that rise to rapid AI and data center expansion.
This is the contradiction at the heart of current AI infrastructure sustainability. On one hand, Amazon promotes renewable projects and future carbon reductions. On the other, it is backing what could be the largest single power-plant CO₂ source in the country to feed one site. When spokespersons say “the world looks different now” while insisting the pledge is intact, they are admitting that AI growth is outpacing the company’s willingness or ability to match it with clean energy.
If net‑zero goals must bend every time AI compute demands a faster build, then those goals are not guiding investment; they are retrofitted marketing language for decisions made on other grounds.
Beyond carbon: water, health, and local environmental costs
Carbon is only part of the data center environmental impact story. Amazon reports that its global data centers withdrew about 2.5 billion gallons of water in 2025, even while claiming slight efficiency gains and increased use of treated wastewater. That volume could supply around 13 million people with two liters per day for a year, a figure that should end any illusion that cloud services are immaterial.
Locally, environmental groups warn that burning gas at this scale near communities will affect both public health and the environment. The plant’s emissions permit is silent on the broader AI data center emissions chain: methane from gas production, air pollutants beyond CO₂, and the long-term lock‑in of fossil infrastructure. Meanwhile, data centers still demand water for cooling, even when companies say they rely on outside air most of the time.
Amazon has pledged that by the end of the decade it will return more water to communities than its operations take. That is a welcome commitment—but it does nothing to soften the front‑loaded damage if the company keeps pairing new AI campuses with mega‑scale fossil fuel projects.
A regulatory vacuum that invites more fossil-fuelled AI
GW Ranch’s permit underscores a regulatory gap around AI infrastructure. Regulators were willing to authorise up to 33 million tons of CO₂ a year for one gas plant serving one AI data center, with no broader framework for aligning such builds with climate targets. That is not oversight; it is a blank cheque.
Industry analysts expect “an explosion of off‑grid gas projects” as hyperscalers tire of waiting for strained grids and opt for their own fossil fuel power plants instead. Gas is becoming the default: fast to build, politically favoured, and easy to justify as “reliable” for AI workloads. Others are experimenting with nuclear deals or ambitious solar concepts, but these remain exceptions, not the rule.
The lesson from Amazon’s Texas project is clear. Without rules that tie data center approvals to real emissions caps and clean energy supply, AI’s growth will pull climate policy backwards. If society wants AI without a wave of new fossil infrastructure, lawmakers will need to regulate AI data center emissions as tightly as any heavy industry—and stop letting climate pledges stand in for enforceable limits.
