Amazon's planned Texas data centre permitted to emit 33 million tons of CO2 a year, report says
An on-site gas plant sidesteps grid queues but puts emissions accounting at the centre of the data centre build-out.
Amazon is planning a large data centre campus in Pecos County, Texas, that would be served by its own on-site natural gas power plant, and the air permit for that plant would allow it to release 33 million tons of carbon dioxide a year, according to reporting by The New York Times cited by TechCrunch on 8 August. At that level the plant would be the single largest source of climate emissions in the United States.
TechCrunch reports that Amazon's carbon emissions rose 16 per cent in the previous year, with AI development named as a significant contributor. The company has a standing pledge, made when it co-founded The Climate Pledge, to eliminate its carbon emissions by 2040. A company spokesperson said the Pecos County site would be powered by new on-site generation that would not raise electricity costs for Texas families. Another representative acknowledged that the world looks different now than when the pledge was created, while stating that the company's commitment has not changed.
The article places the project against a broader backdrop. Data centre construction faces growing political opposition in several parts of the country, with concerns about residential electricity prices among the most common objections. Large technology companies have increasingly backed the construction of natural gas plants to power AI workloads, in part because such plants can be built and controlled by the developer rather than waiting on the wider grid. TechCrunch did not give a construction timeline, turbine count or capacity figure for the Pecos County facility, and this article reports only what that piece contains.
Why it matters
The energy story of AI infrastructure has shifted from how much power is needed to whose power it is and who bears the cost. Behind-the-meter generation lets a developer avoid grid interconnection queues, but it also concentrates emissions and regulatory attention on a single site. Corporate climate pledges made before the current build-out are now being measured against gigawatt-scale gas plants, and the gap between the two will shape permitting debates for years. Expect emissions accounting, not just megawatts, to become part of every data centre announcement.
The energy story of AI infrastructure has shifted from how much power is needed to whose power it is and who bears the cost.
Grid-connected campus
Utility supplies power; needs interconnection agreement and grid capacity; emissions counted as purchased electricity (Scope 2)
Behind-the-meter campus
Developer builds on-site generation; avoids grid queue; needs fuel, water and air permit; emissions counted as on-site fuel burn (Scope 1)
What you can learn from this
Behind-the-meter generation bypasses the grid, not the physics. A behind-the-meter plant sits on the same side of the utility meter as the load, so the electricity never crosses the public grid and does not need a formal interconnection agreement. That removes years of queue time, which is why developers are drawn to it. It does not remove the need for fuel supply, water, gas pipelines or an air-quality permit, all of which become the developer's responsibility instead of the utility's.
Air permits set ceilings, not forecasts. An emissions permit states the maximum a facility is allowed to release under its approved design and operating hours. Actual output depends on how many turbines run, how often and at what load, so a permit figure is best read as the outer bound. Regulators require the ceiling up front because once a plant is built its potential to emit is fixed by its hardware, and neighbours need to know the worst case.
Gas turbines dominate because they are dispatchable and fast to build. A simple-cycle gas turbine can be started and stopped on demand and can ramp quickly, which suits a load that must run at full tilt around the clock. Solar and wind are cheaper per unit of energy but intermittent, so pairing them with a constant load requires large storage. Developers under time pressure choose the option that guarantees capacity on a known date, and that choice carries a carbon cost.
Corporate emissions are measured in scopes, and on-site generation moves them. Under the widely used Greenhouse Gas Protocol framework, Scope 1 covers fuel burned on your own site, Scope 2 covers purchased electricity and Scope 3 covers your supply chain. Buying grid power and matching it with renewable certificates affects Scope 2; burning gas on-site is Scope 1, which cannot be addressed with certificates in the same way. That accounting difference is one reason on-site gas is harder to reconcile with a net-zero target.
A data centre's load is nearly flat, which is unlike most electricity demand. Homes and offices peak in the evening and drop overnight; a server hall running training or inference draws close to its maximum every hour. Generators serving a flat load run at a high capacity factor, which is efficient for the plant but means annual emissions sit close to the theoretical maximum. This is why a gigawatt of data centre load implies far more energy over a year than a gigawatt of residential peak.
We teach this
How to use this in practice
Do the energy arithmetic for a hypothetical campus. Take a 1 GW IT load, multiply by 8,760 hours and you get 8.76 TWh a year; then apply a PUE of 1.2 to include cooling and losses. Write the calculation in a spreadsheet with the inputs as editable cells. Done means you can change the load or PUE and see the annual energy and, using a published emissions factor for gas generation, an approximate CO2 figure update automatically. Note the source of the emissions factor you used.
Draw the power path for a behind-the-meter site. On paper or in a diagramming tool, sketch gas supply, turbines, switchgear, the site meter, the data halls and, as a dotted line, an optional grid tie. Label which permits or agreements each link needs, and add a note on where the emissions are counted at each stage so the diagram doubles as a Scope 1 versus Scope 2 map. Done means a one-page diagram that a colleague can read and that distinguishes what the developer controls from what the utility controls.
Check the carbon reporting your own cloud usage produces. Log in to your cloud provider's console and locate its carbon footprint or sustainability report for your account; the major providers each offer one. Export a month of data and identify which region carries most of your emissions. Done means a short note listing your top region, its reported emissions and the provider's stated methodology, so you know whether the figures are location-based or market-based.
Read one real air permit. State environmental agencies publish permit applications and draft determinations for power plants. Find one for any gas plant in your region, locate the section that lists allowable annual emissions and the assumed operating hours, and note the difference between potential and expected emissions. Done means you can explain to someone else what a permitted tonnage figure does and does not tell you, and you have saved the document link and the two figures in the same spreadsheet as your energy calculation.
Sources
Our reporting is an original summary; full coverage is at the links above.
Don't just read about it — build it.
Square 1 teaches the skills behind the headlines, with every line of your work graded by AI. Find your starting point in 3 minutes.
Get your free skill report