Google backs a Texas data center being prepared for power from a large gas plant

Google has invested in Goodnight, a new data center in Texas that, according to project documents, could partly run on its own gas generation. It is another sign that AI infrastructure increasingly depends not on a "green" agenda but on fast access to power.

How the project works

Goodnight is a campus in Armstrong County, Texas, being built by infrastructure company Crusoe with Google's participation. In November, Google announced $40 billion in investments in Texas AI infrastructure, and this site is part of that package. According to the environmental permit application, the site will have six buildings: the first four will connect to the shared power grid, while the fifth and sixth are planned to run on a private gas plant located on the campus itself.

The most contentious point is the scale of emissions. The documents cite more than 4.5 million tonnes of greenhouse gases a year. That is comparable to putting more than 970,000 gasoline cars on the road. Google notes that it does not yet have a final contract for gas power supply for this particular campus. But the project's design shows the option is already being considered seriously, which means it has become part of the new logic of building large AI facilities.

Why developers choose gas

Alongside gas generation, Goodnight also has about 265 megawatts of wind capacity declared. But the main takeaway is not that Google has abandoned renewable energy; it is that renewables alone are no longer enough to bring large data centers online on the required schedule. Developers run into long grid-interconnection queues and try to get electricity directly, bypassing infrastructure bottlenecks.

"Grid growth is not keeping up with demand from AI, so gas has become a critical bridge," says

Crusoe co-founder Cully Cavness.

Here is what makes this case telling:

• more than 900 megawatts of gas capacity is under consideration at the site;
• some buildings are to receive power not from the grid but from on-site generation;
• nearly 100 gigawatts of gas projects are being prepared for data centers in the US;
• similar campuses from other players are planned with even higher emissions.

This approach is called behind-the-meter power: the facility builds its own generation next to where the electricity is consumed. For AI companies, it is a way to avoid waiting years for a grid connection and to launch clusters for model training and inference faster. The problem is that almost everywhere the "bridge" turns out to be natural gas, not storage, solar plants or other low-carbon sources, which are harder to deploy at the required scale and pace.

Climate and politics

The story is especially sensitive for Google because the company was long one of Big Tech's main symbols of climate pledges. Amid the AI boom, that is getting harder: over the past five years, Google's total emissions have risen noticeably, although the company says emissions from its data centers fell 12 percent last year. It is also pursuing alternative scenarios: another Texas site, in Haskell County, is promised to be built next to a new solar plant and an energy storage system.

The political backdrop is shifting too. The White House recently convened the largest tech companies to discuss how to add new power capacity without putting pressure on consumers. For the market, though, that is more a signal than a real mechanism. Meanwhile, US senators have begun asking AI companies and data center operators to explain why they increasingly choose gas over renewable sources for new campuses.

What this means

Demand for AI computing is growing faster than power grids and clean generation. So even companies with strict climate goals are starting to consider on-site gas plants as a fast way to bring new capacity online. For the market, this is bad news in terms of emissions, but an important marker of what the race for AI infrastructure looks like in practice.