On October 15, the Public Utility Commission of Texas took up three applications that ask the same question in three different forms: when a very large electricity customer wires itself straight into a single power plant, and the plant’s output is netted against what the customer draws, what does the grid get in return? The three arrangements paired a West Texas solar project with a data-center company, the Comanche Peak nuclear plant with a large-load customer, and a Haskell County wind farm with a data-center owner. As of publication, no orders or vote outcomes from the October 15 meeting had surfaced in publicly verifiable sources, so this article reports what was before the commission — not how the votes went.
The stakes are larger than three dockets. Texas spent the summer discovering that its large-load queue — roughly 474 gigawatts of tracked requests, about 90 percent of it data centers — was mostly reservations rather than real projects, and it responded with an audit and a pause on new energization approvals. Co-location is the path that lets a big customer route around the queue. The commission’s job under Senate Bill 6 is to make sure that shortcut does not quietly subtract generation the rest of the grid was counting on.
What netting means, and how the clock works
Start with the mechanics, because “co-location” covers several electrical arrangements and only one triggers this process. In a net-metering arrangement under SB 6, a large load sits behind the same point of interconnection as an existing, standalone generation resource, and the generator’s output is netted against the customer’s consumption for settlement purposes. The load and the plant share a meter. From the grid’s perspective, the pairing can look like a smaller net load — or, when the plant is generating strongly, like almost no load at all. The concern is what happens when the grid needs that plant’s output and the customer is absorbing it behind the meter.
SB 6, enacted in 2025, added Section 39.169 to the Public Utility Regulatory Act to govern this situation. It applies to new net-metering arrangements between an existing standalone generation resource — one registered in ERCOT’s network operations model as of September 1, 2025 — and a new large-load customer of 75 megawatts or more. Before implementing such an arrangement, the generation owner must notify ERCOT. ERCOT then has 120 days from receiving all the information it needs to complete a reliability study and send its findings to the commission. The commission has 60 days after that to approve, deny, or approve with conditions. If it lets the 60 days run out without acting, the arrangement is deemed approved.
Two features of that design deserve attention. The deemed-approval clock means inaction is approval. And the table is narrow: the only parties entitled to participate are the commission, ERCOT, the transmission provider, and the parties to the arrangement. A transmission provider may object for reasonable cause before a final decision, but community and environmental groups have no standing.
The three cases
The fullest public picture exists for the solar case. Docket 59115 concerns Silver Basin Digital Infrastructure, a proposed 180-megawatt data-center load co-located behind the 204-megawatt-AC Taygete II solar plant in Pecos County, according to the industry newsletter Barrio Energy, which tracked the docket ahead of the commission’s October 1 meeting. Initial briefs were filed September 23, and ERCOT proposed a 30-minute curtailment condition — the same emergency-disconnect obligation at the heart of the July precedent. The pairing is the most intuitive of the three. The question is what happens at sunset, when the plant’s output falls to zero and the load does not.
The nuclear case, Docket 59399, is the one the industry is watching most closely. The application comes from the operator of the Comanche Peak nuclear plant — Luminant Generation, a subsidiary of Vistra Corp. — whose two-unit, roughly 2.4-gigawatt station sits in Somervell County southwest of Fort Worth. What the publicly reachable filings did not establish is the size of the paired load or the identity of the customer. Press reports have separately described data-center development interest around the plant, including a county development agreement involving Amazon Data Services, but those reports should not be read as identifying the counterparty in this docket; the filing itself did not make that link in materials this article could verify. The nuclear case matters because it is the hardest version of the question: a dispatchable, always-on plant whose output the grid counts on around the clock, paired with a load that would absorb it behind the meter.
The wind case, Docket 59943, is the least documented of the three. It involves a Haskell County wind farm paired with a data-center owner, per the Texas Dispatch’s pre-meeting reporting. How much power the arrangement would move, and who stands behind the data-center entity, could not be established from filings that reporting could reach. That opacity is itself part of the story: the process is designed to surface exactly these facts through ERCOT’s study and the commission’s review, but much of the underlying filing material is not easily visible to the public before the commission acts. The three had also been listed for the commission’s October 1 meeting.
The July precedent
The commission has decided one net-metering case under this provision, and it set a template the three new cases will be measured against. In Docket 59220, decided July 24, the commission approved an arrangement netting the 265.5-megawatt Goodnight Wind facility in Armstrong County against Crusoe Load Two, a proposed 260-megawatt AI data center developed in partnership with Google, with the load assets later transferred to a special-purpose entity, Ensign Infrastructure. The commission used the case to establish that emergency curtailment is not capped by the size of the paired generator.
The core of the order was a curtailment obligation. Crusoe Load Two must be capable of fully disconnecting — its entire load, not just the portion offset by wind output — within 30 minutes of an ERCOT instruction during a transmission emergency, with physical breaker disconnection if necessary, and without compensation. The commission required ERCOT to give at least 60 minutes’ advance notice “when practicable.” The logic was straightforward: if one load behind the shared interconnection curtails but a second does not, the second load absorbs the wind farm’s generation behind the meter, and none of that capacity reaches the grid.
The arithmetic is what made the case notable. A prior commission order, issued in April, had already imposed a 265.5-megawatt, 30-minute curtailment obligation on the adjacent Crusoe Load One facility. ERCOT recommended, and the commission adopted, a cumulative 525.5-megawatt curtailment obligation across the two data centers behind the single wind farm — roughly twice the wind farm’s capacity. The developers argued that was double-counting. The commission rejected that argument, holding that the conditions exist to ensure the existing generator’s capacity is fully available to the grid during emergencies, and that ERCOT’s operational requirements supersede the operator’s hardware concerns about 30-minute shutdowns. The shutdown risk, in the commission’s view, is a commercial trade-off the developer accepted in exchange for faster access to power.
The July order also imposed the standardized conditions from the commission’s implementing rulemaking, Project 58479. The arrangement is barred from participating in paid ERCOT demand response, ancillary services, Emergency Response Service, or transmission-provider load management programs — mandatory emergency curtailment is a reliability obligation, not a service to be sold back to the market. The conditions attach to the physical assets and bind future owners. And they carry an expiration date of sorts: between 36 and 60 months from the order, the parties must jointly apply for a determination of whether the conditions should be extended, modified, or rescinded.
Why nuclear tests the template differently
Apply that template to Comanche Peak and the fit is awkward in instructive ways. The statute requires conditions ensuring a dispatchable resource makes available to ERCOT in an emergency at least the same amount of power it was providing before the arrangement. For a wind farm, that means the load gets out of the way so the turbines’ output can flow to the grid. For a nuclear plant, the plant itself cannot be curtailed on a 30-minute signal — nuclear units run at steady output, and their value is precisely that they do not move. The curtailment lever falls entirely on the load: the data center must disconnect so the plant’s output keeps flowing to the grid rather than being absorbed behind the meter.
That raises a question the July order did not have to answer. When the paired generator is variable, the emergency condition mostly matters in hours when the generator is actually producing. When the generator is a nuclear plant running around the clock, the condition matters in every hour — and the load’s promise to disconnect becomes the entire reliability protection. The practical question is how much backup capability the load must maintain to ride through a disconnection lasting hours rather than minutes. One analyst, commenting on the July order, argued the architecture now essentially requires backup capacity capable of carrying the load off-grid for a meaningful period — an interpretation never tested against a nuclear pairing.
Why variable generation tests it differently
The solar and wind cases pose the mirror-image problem. A solar plant’s output falls to zero every night; a wind farm’s output varies with weather ERCOT cannot schedule. Netting a steady, always-on data-center load against a variable generator means that for much of the time, the “net” load is simply the data center’s full draw on the grid — the arrangement saves the customer transmission and settlement exposure in the hours the plant produces, while the grid still serves the load in the hours it does not. That is not a defect; it is the economics of pairing a variable resource with a flat load. But it is why ERCOT’s study matters: the reliability question is not the average hour, it is the tight hour, when the paired plant is producing little and the system is short.
The gaming concern ERCOT raised in July applies here too. If a developer could pair a very large load with a nominal generator — a small plant whose output is a rounding error next to the load — and thereby shield most of the load from curtailment, the process would become a loophole rather than a safeguard. The commission’s answer in July was full-load curtailment regardless of the generator’s size. The October cases test whether that answer holds when the generator is a 204-megawatt solar plant paired with a 180-megawatt load — much closer to one-to-one — and when it is a wind farm whose size is not public.
The conditions toolbox, and the awkward timing
Beyond curtailment, the commission has other levers. The statute allows conditions requiring load reduction, dispatchable generation obligations, and assurance that other system users are held harmless for stranded transmission costs. The standardized framework adds a financial backstop: if an arrangement causes a shared transmission asset to transmit at least 25 percent less power than expected, the parties face direct financial liability, protecting retail ratepayers from paying for wires a behind-the-meter pairing left underused. Non-time-limited conditions must be reviewed at least every five years.
All of this is happening while the rest of Texas’s data-center apparatus is frozen. Governor Abbott’s August 3 directive ordered the audit of every data center in the interconnection process and halted approvals to energize new large-load data centers and virtual currency mining facilities; his September 21 directive halted TCEQ permits for data centers until the audit finishes, with TCEQ’s compliance report due to the governor’s office on October 19 — the day this article publishes. ERCOT expects to publish its audit report by December 10, and the Batch Zero grouped study is stalled behind the audit. The net-metering dockets, by contrast, are adjudications of arrangements already filed, proceeding under a statute with its own clock.
What to watch
When the commission publishes its decisions, the telling detail will not be approve-or-deny but the conditions: whether the 30-minute full-curtailment obligation appears verbatim in all three, whether the nuclear case draws an additional backup or availability condition, and whether the solar and wind cases get the same demand-response bar and the same 36-to-60-month review window. Divergence by technology would signal that the July template is a starting point, not a rule; uniformity would signal that Texas has settled on a single price for co-location, whatever the fuel. Also watch the legislative session convening in January 2027, where a framework whose standing rules are narrow and whose inaction counts as approval is a natural target for lawmakers.
For developers, the lesson of the autumn is already clear, whatever the three orders say. Texas will let a data center bring its own power — but the state intends to keep the right to take that power back, in full, on half an hour’s notice, and to revisit the deal in three to five years. That is the price of skipping the queue. The October cases will show whether the price is the same for everyone.