Crusoe’s controversial Goodnight campus, under construction in the Texas Panhandle, has found itself at the center of the state’s debate over how to manage co-located data centers.
The gigawatt-scale campus, developed in partnership with Google, will leverage 933 megawatts of off-grid fossil gas turbines; the latter will emit more emissions than the average coal plant. The headline news this week was that Google, which has long had ambitious climate goals, is embracing fossil fuel generation. But one phase of that same data center campus will also co-locate behind the meter with an onshore wind farm, and Crusoe’s bid for regulatory approval of that deal has made Goodnight the test case for ensuring such arrangements don’t hurt ratepayers in Texas.
In late October, Crusoe and Goodnight Wind — Serena Energy’s 265.5-MW farm, which came online in 2024 — submitted a proposal to the Public Utilities Commission of Texas for a net metering arrangement under Senate Bill 6, the state’s legislative overhaul of large load interconnection.
Under that proposal, the first phase of the Goodnight campus and the wind farm would share a single point of interconnection, and net their power usage against the wind farm’s production. As a result, that segment would come online later this year.
Co-locating data centers with behind the meter generation is the subject of intense regulatory scrutiny around the country, including at the Federal Energy Regulatory Commission, where the Department of Energy’s October directive to study large load interconnection envisioned a fast lane for co-located facilities, and in PJM, which is in the midst of reforming its own rules for behind-the-meter net billing. But while PJM is working on more flexible (read: FERC-directed) reforms, Texas is doubling down on a strict approach that requires state approval and relies heavily on curtailments and mandatory dispatch retention.
In the past, loads and generation that wanted to share an interconnection point simply had to register both resources, but didn’t require ERCOT study or approval. But in 2022, when it became clear that rapid load growth, including from crypto miners and early hyperscalers, could cause a surge of unstudied connections, ERCOT implemented an interim process that added a transmission study requirement to the co-location process.
SB 6, passed last summer, went further, formalizing reliability study requirements and adding rules about both PUC approval and emergency dispatch — but only for new large loads pairing with existing stand-alone generators. It was an effort to prevent merchants from self-supplying hyperscalers to avoid regulations that apply to new generation.
Last week, the PUCT adopted final rules for net metering arrangements for co-located data centers 75 MW or larger. Now, the portion of the Goodnight campus set to be wind powered is Texas’ first test case for the new rules.
Notably, according to PUCT filings, Crusoe sold the would-be co-located portion of the Goodnight project to a company called Ensign Infrastructure in mid-March. Ensign, incorporated in Delaware last summer, is now the official “large load” in the co-location application.
According to Latitude Media’ s review of filings, Ensign appears to be a special purpose vehicle for the co-located project — a common project finance practice that would isolate that portion of the wider Goodnight project from Crusoe’s broader liabilities. It could also serve to create a legal firewall between the carbon-free portion of the campus and its extensive gas-fired expansion, allowing Google’s contract to be technically separate from the 4.5 million tons of greenhouse gasses the off-grid gas turbines are expected to emit each year.
Crusoe didn’t immediately respond to Latitude’s request for comment on Friday. Google, for its part, confirmed that it does have an agreement for the wind energy secured in the co-location deal, but told Wired it did not have a “contract in place” for the gas turbines.
Co-location trade-offs
Against the objections of data center developers in the rulemaking docket, the PUCT approved a set of strict, standardized conditions, including mandatory emergency curtailment and a prohibition on participating in any paid demand response programs.
Those would require the Goodnight data center load to fully curtail its consumption within 30 minutes when notified of a grid emergency by ERCOT. Whether by ramping down compute or switching to backup power, the data center must make 100% of the wind farm’s capacity available to the grid during that time.
In the case of co-located data centers, the commission added, it must be the wind farm’s own capacity that is made available: “Bring your own capacity” approaches, in which data centers leverage off-site virtual power plants in response to ramp-down orders, are not allowed.
Crusoe, in its filings to the PUCT, argues an AI data center — and the Goodnight project specifically — simply can’t respond within 30 minutes. At least an hour is necessary to avoid “cascading hardware failures across server clusters” and the “loss of active computational work that cannot be recovered.”
Additionally, Crusoe argues, thatSB6 only required that “dispatchable capacity” be made available to the grid. Wind, as a non-dispatchable resource, should be exempt.
The PUCT maintains that ERCOT’s operational needs “override any individualized concerns” regarding the safety of data center hardware.
Just as in other parts of the SB6 implementation process, the financial implications for data centers that opt to co-locate are steep. Under the commission’s rules, data centers won’t be compensated for the curtailment, and nor will they be allowed to participate in other paid demand response programs.
Those rules, Ensign wrote in its latest filing, effectively create “a massive, free ancillary service that allows [ERCOT] to curtail all co-located loads.” They argue that this is particularly relevant for non-dispatchable generators, like renewables, where load might be forced to shut down even if it isn’t producing energy. Barring participation in commercial demand response, Ensign added, is “an affirmative market distortion that privileges incumbent supply-side resources over equally capable alternatives.”
Data centers are “among the highest-quality demand response resources available in ERCOT,” they added. Blocking data centers from demand response and ancillary services programs means the region would be losing access to large, controllable loads “precisely at the moment when demand response depth is most needed.”
Ultimately, Crusoe and Ensign argue that the rules unduly disadvantage developers pursuing shared grid points, and will disincentivize co-location. Texas regulators, however, say the steep rules, designed to both maintain system reliability amid massive load growth and protect ratepayers from paying for stranded or underutilized transmission assets, present a “known risk” that data centers can undertake voluntarily if they want speed to power.
Wider co-location rules
Crusoe, of course, won’t be the only data center developer to grapple with Texas’ new net metering rules. Across the state in Freestone County, for instance, developer CyrusOne is also seeking approval of its co-location agreement with Calpine, which would power a 760 MW data center campus behind-the-meter with fossil gas.
In finalizing its strict guidelines, the state has doubled down on what amounts to a “trust but verify” approach, formalizing a pathway that could help data centers get online more quickly — but at a high price. It’s a balancing act that other parts of the country, most notably PJM, are also navigating as load growth surges and ratepayers worry about footing the bill.
In the wider co-location docket, the PUCT grappled with a few additional challenges that weren’t front and center in the case of Goodnight’s wind application.
Perhaps the most controversial element of the final rules is the commission’s approach to “underutilized assets.” If a co-location arrangement results in a power line sitting even partially idle, the parties face financial liability.
Industry groups sought to raise the threshold, pointing out that the rule creates a long tail of financial risk for developers, because potential utilization costs won’t be ironed out before a net metering arrangement is approved. But the final rules define underutilization as an asset that is expected to transmit at least 25% less power as a result of the co-location agreement.
“Customers should be held harmless for a transmission asset that is no longer consistently used for the primary function it was built to serve,” the commission wrote.
