Two months after OpenAI, SoftBank, Oracle and MGX announced Stargate with a pledge of $500 billion, the project's first physical footprint has a size. On March 18 Crusoe said it had begun the second phase of its AI data center at the Lancium Clean Campus in Abilene, Texas, taking the site to eight buildings, about 4 million square feet, and a total power capacity of 1.2 GW. The second phase is expected to be complete in mid-2026.
That is one campus. It is also a useful measure of what the larger Stargate pledge implies for electricity, and of how developers in Texas are trying to secure power on timelines that the grid alone cannot meet.
From pledge to site
OpenAI's January 21 announcement described Stargate as a new company that intends to invest $500 billion over four years building AI infrastructure for OpenAI in the United States, with $100 billion to be deployed immediately. SoftBank, OpenAI, Oracle and MGX are the initial equity funders. SoftBank has financial responsibility and OpenAI operational responsibility, and Masayoshi Son is chairman. Arm, Microsoft, NVIDIA, Oracle and OpenAI were named as the initial technology partners. The announcement said the buildout was already underway, starting in Texas, and that other sites were being evaluated.
Crusoe began building in Abilene in June 2024, before Stargate had a name. The first phase has two buildings totaling 980,000 square feet and more than 200 MW, and is expected to be energized in the first half of 2025. Crusoe says each building is designed to run up to 50,000 NVIDIA GB200 NVL72s on a single integrated network fabric. About 2,000 people work on the site daily, a figure Crusoe expects to rise to nearly 5,000 with the expansion.
The scale-up from 200 MW to 1.2 GW in a single announcement is the point. Crusoe says its own footprint now exceeds 1.6 GW in operation and construction, with a further pipeline above 10 GW in development.
The power plan
Lancium, which owns the land, handles grid interconnection and what it calls power orchestration. According to the March 18 release, its plan for the site includes a 1.2 GW grid interconnection along with large-scale behind-the-meter battery storage and solar. The campus is meant to draw on nearby wind and to encourage new renewable development. Crusoe said the availability of low-carbon energy weighed heavily in site selection, and that the facility will use natural gas turbines for backup power. Cooling is direct-to-chip liquid in a closed loop designed so that water is not lost to evaporation.
The gas element is larger than a typical backup system. In January, B17 News reported filings with the Texas Commission on Environmental Quality, made by a consultant on Crusoe's behalf, seeking a Title V permit for 10 simple-cycle gas turbines at a site sharing an address with the project. The filings said the plant could produce up to 360.5 MW at any time, around the clock, for primary and backup power, onsite only, with no export to the grid. Half of the turbine models were from GE Vernova and half from Solar Turbines, a Caterpillar subsidiary. A permit for diesel backup generators was already in effect. Lancium told the outlet it had built a 200 MW substation on the site and planned to increase that capacity fivefold in 2025.
Put together, the design layers three sources: a grid connection that grows in steps, on-site gas sized at roughly 30% of the campus's full capacity, and contracted or nearby renewables. That is a practical answer to a simple problem. Data center developers can build halls faster than utilities can build substations and lines, and the turbines cover the gap and the risk.
The arithmetic
A campus drawing 1.2 GW continuously would use about 10.5 TWh in a year, since 1.2 GW times 8,760 hours equals 10,512 GWh. Actual use will be lower, because AI campuses rarely run at their full rated capacity every hour. Still, the comparison is instructive. Lawrence Berkeley National Laboratory estimated in its December 2024 report that all U.S. data centers consumed about 176 TWh in 2023, about 4.4% of national electricity. One fully loaded Abilene campus would equal about 6% of that national total.
Stargate's pledge covers many such campuses. OpenAI did not say how many gigawatts $500 billion would buy, and the partners have not published a site list. But at roughly the scale of Abilene, even a handful of campuses would add load comparable to a midsize state's demand, concentrated in a few places.
Why Texas
Texas offers what Stargate's first site needed: fast permitting, land, and an independent grid with a market structure that lets large loads connect without the multi-state planning cycles of PJM or MISO. West Texas has abundant wind and a growing solar fleet, which helps with the low-carbon goals Crusoe cites. It also has a dense gas pipeline network, which makes on-site turbines a straightforward option.
Those advantages come with exposure. Large loads in Texas are subject to the same grid emergencies as everyone else. ERCOT treats any facility of 75 MW or more as a large load, a category that already covers crypto mines and industrial sites and now includes AI campuses many times that size. How much flexibility such customers owe the grid during tight hours is an open question for the state. On-site generation, such as the turbines in the Abilene filings, is one way a campus can keep running if the grid asks it to step back.
What to watch
The first two buildings are scheduled to energize in the first half of the year. That will show how quickly Lancium's substation capacity grows and how the turbines are used once the plant is permitted. The second phase, due in mid-2026, will test whether a 1.2 GW load can be integrated into a regional grid in under two years.
For now, Abilene is the clearest picture of what Stargate means in electrical terms: a campus the size of a large power plant, built in phases, and powered by whatever combination of grid, gas and renewables can be brought online fastest.