Crusoe confirms Google as end user of its Goodnight campus in Texas
The Armstrong County project connects directly to two wind farms totaling 531MW, and Aterio Research puts the planned build at more than a gigawatt.
Crusoe confirmed on September 28 that Google will be the end user of the data center campus the developer has been building since June 2025 near Amarillo, Texas, a project it calls the Goodnight Campus. The confirmation, reported by Data Center Dynamics, places the site at 10001 Lima Road in Claude, Armstrong County.
Chief executive and co-founder Chase Lochmiller described the project as an extension of an energy-first strategy that builds AI factories at the source of low-cost power, saying Crusoe broke ground in June 2025 and is building at the same pace it set at Abilene, the Texas campus used by OpenAI. "Building the future of AI requires a new kind of infrastructure, starting with energy," he said, and the framing points at the part of the project carrying the most information: how the campus gets powered, and what happens to the power it does not use.
Scale depends on which count you take. Data Center Dynamics, working from the project's filings, describes a campus ultimately planned to hold at least six single-story buildings—four at 805,380 square feet each and two at 484,955 square feet each—and notes that Crusoe filed in July for two buildings beyond the original plan. Aterio Research, cited by Data Center Dynamics, counts seven buildings and more than a gigawatt of capacity, split into a four-building, 530MW first phase and a three-building, 500MW second phase, against an expected total investment of $29 billion, and expects Goodnight to mirror the company's Abilene campus. One building's difference is a small gap and a useful reminder that the site plan is still being drawn.
Two wind farms and a 531MW first phase
The campus takes a direct connection to Serena's adjacent Goodnight 1 wind farm, a 265.5MW project already operating, and to Goodnight 2, a second 265.5MW farm still under construction, with any surplus beyond what the campus needs injected into the grid. Their nameplate sums to 531MW, one megawatt above the 530MW Aterio assigns to the first phase, which suggests the opening build was sized to the wind pair rather than to a queue position and leaves the second phase to track the second farm's completion.
Google's own Texas numbers are larger and older, with the company saying in November 2025 that it would invest about $40 billion through 2027 to expand cloud and AI infrastructure in the state, directed at existing campuses in Ellis County and three new campuses planned in Armstrong and Haskell counties. The Armstrong County campus named in that pledge is likely this one, which would make the September confirmation less a fresh commitment of capital than the naming of a tenant for capacity already budgeted.
The two companies have been paired before, and the earlier chapter is the one to keep in view: Crusoe developed Project Tembo, a 2.7GW campus near Cheyenne, Wyoming, and said in June that it had paused work there at the request of its customer. Reporting since has suggested that Google subsidiary Jupiter Star Holdings took over the site in Crusoe's place, and Crusoe has not commented on that reporting publicly, as the Data Center Dynamics account notes.
Crusoe's Texas book now runs to a 1.4GW campus in Childress, the Abilene site used by OpenAI where further capacity is set to be leased to Microsoft, and Goodnight; Abilene, the benchmark Lochmiller reached for when describing Goodnight's construction pace, is the reference point to hold the company to.
What Google's name does to the underwriting
Underwriting here begins with the name on the other side of the lease. Data center capital prices the anchor rather than the megawatt, and Goodnight now has Google as its anchor. The $29 billion Aterio figure is a build-horizon estimate rather than a signed number, and the disclosure carries no power price, no lease terms and no energization date, but the counterparty is named—which inverts the unpriced-announcement pattern in which a project milestone doubles as a financing milestone and the buyer stays unnamed.
The siting makes the case directly. Power rights are being optioned before offtake because the queue is the binding constraint; Crusoe's answer is to buy proximity to operating and near-complete generation instead of a place in line. Two wind farms next door to a gigawatt-scale campus is a bet that speed to power, not floorplate, decides which projects get leased, the same logic behind Google's willingness to stake exploration positions in new markets, as with the Lea County option from September.
Delivery is the part that cannot be settled until Goodnight 2 is energized, and as with Google's geothermal contracts in Utah, the doubt sits in the delivery chain rather than the resource. A data center's power story is only as firm as its construction schedule, so second-phase capacity hangs on a wind farm still being built and the campus's energy position is a function of how those two schedules converge.
The unglamorous test is whether Goodnight 2's turbines turn before the second phase's 500MW is ready to draw, which matters more than any headline about the anchor tenant. The next building filing will show whether Crusoe holds the July cadence; four buildings at 805,380 square feet apiece is a lot of concrete for a first phase matched to a pair of wind farms to within a megawatt.
Underwriting here begins with the name on the other side of the lease.
Save this analysis and keep the funds you follow together in My Desk.
Sign in to save articles or follow funds.