A water-use survey is now an ERCOT gating item

On September 1, 2026, the Texas Senate Committee on Water, Agriculture and Rural Affairs held an interim hearing on data center water use. Representatives of the Texas Water Development Board and the Public Utility Commission told that committee something that has not travelled far enough.

Responding to their water-use surveys, they said, has become a de facto prerequisite for future state interconnection approvals, particularly Batch Zero, on the authority of the Governor's August 2026 audit directive.

A water agency survey is now functioning as a grid gate. Not a bill. Not a rule. A survey.

That is reported through Holland & Knight's account of the hearing rather than published as an ERCOT requirement, and it should be treated at that weight. But it points at the thing most of the market is not tracking, because most of the market spent September looking somewhere else.

The story everyone covered instead

On September 3, ERCOT issued conditional classifications for Batch Zero, the first system-wide study of large load interconnection requests in the United States. It did not publish the results. Everything known came five days later, when three listed sponsors chose to disclose.

Read those disclosures together and a pattern appears. Among the projects that disclosed, the ones that landed in the top tier were not the ones with the most megawatts, the best sites, the strongest tenants or the deepest capital. The tier turned on which agreement a project held on a particular date.

The two tiers are not degrees of the same thing

Under ERCOT Planning Guide Section 9, a Large Load sorted into Batch Zero lands in one of two places.

Base Load is modeled into the study as given. It is not evaluated further and it is not subject to capacity allocation. The megawatts are treated as known demand.

Studied Load is included in the batch, but the allocation is determined by system-wide evaluation. It carries additional grid studies, potential transmission upgrade cost allocation, possible curtailment obligations and schedule risk that Base Load does not.

Everything else sits outside the batch.

That is not a ranking. It is the difference between capacity you keep and capacity you apply for.

The discriminator is a contract type

Under Section 9.2.1.1, a project requesting energization on or after January 1, 2028 qualifies as Base Load if it holds valid interconnection studies and an executed Interconnection Agreement meeting Section 9.7.2, filed by the eligibility deadline.

Under Section 9.2.1.2, a project that misses that test still enters the batch as Studied Load if it holds studies sufficient under Section 9.2.1.4 and an executed Intermediate Agreement meeting Section 9.7.1.

One agreement buys the top tier. A different one buys the lower tier. Project scale, tenant credit and capital structure do not appear in the test.

That is not a complaint about bureaucracy. An executed Interconnection Agreement is a real thing: it means a transmission service provider has studied the load and agreed to serve it. The document is a proxy for genuine project maturity, and ERCOT chose a defensible proxy. The point is narrower and more useful. Two sponsors of comparable quality can land in different tiers because of which agreement they had signed, not because of how good the project is.

Hut 8 and AEP Texas executed an interconnection agreement for 1,000 MW of utility capacity at Beacon Point in Nueces County, disclosed in May 2026. Beacon Point came back Base Load. Galaxy submitted Caspian, 700 MW, as Base Load. It came back Studied Load.

Galaxy did not say why, and ERCOT does not publish reasons. Reading the downgrade against the eligibility criteria is our inference, not a disclosed fact. But Galaxy is Nasdaq-listed, runs a pipeline it puts above 5.7 GW, and has a flagship campus already generating revenue. Whatever moved Caspian, project quality was not the missing ingredient.

The disclosed roster

Galaxy disclosed roughly 4.2 GW across five projects, 1,630 MW in the top tier and 2,600 MW in the lower one. IREN disclosed 2,000 MW, all Base Load. Of the 6,230 MW that Galaxy and IREN disclosed on a comparable basis, 42 percent came back in the lower tier.

Every megawatt of it is conditional. ERCOT's language in Market Notice M-A080326-04 is that conditional inclusion is not final interconnection approval and does not establish an energization date.

The ERCOT counterparty is not the tenant

Helios is worth pausing on, because it demonstrates something the market keeps collapsing.

CoreWeave is Galaxy's anchor tenant at the Dickens County campus, contracted for 526 MW of critical IT load across Phases I through III under 15-year leases with two five-year extension options, which Galaxy expects to generate more than $1 billion in average annual revenue. Phase I was delivered in July 2026, putting roughly 200 MW of gross power and 133 MW of critical IT into service, with rent commencing in the second quarter of 2026. Phase II covers 260 MW of critical IT with hall deliveries expected in the first half of 2027.

That is a real tenant on a real lease paying real rent.

But Galaxy states plainly that as developer, owner and operator it holds all retail electricity service agreements itself, with power costs folded into the broader hosting or lease arrangement.

So the entity ERCOT is auditing is Galaxy, not CoreWeave. The attestations under Section 9.2.1 were sworn by Galaxy. The verification RFI goes to Galaxy through its service provider. A tenant of any credit quality does not answer for the ILLE's filings, and a lease does not cure an attestation problem.

Hut 8 is the same structure with the name withheld. It describes its Beacon Point tenant only as high-investment-grade, rated AA− or higher, across two 15-year triple-net leases totalling 704 MW of IT capacity and $19.6 billion of base-term value. The campus is designed to NVIDIA's DSX reference architecture. The Financial Times has reported the tenant is NVIDIA. Hut 8 has not confirmed it.

Two problems follow for anyone underwriting these positions. You would be relying on a press report rather than a disclosure for the tenant's identity, and you would be assessing the credit of a party that does not answer to ERCOT for anything.

What the Governor's audit is collecting

On August 3, 2026, Governor Abbott directed the PUCT and ERCOT to conduct a comprehensive verification and audit of all data centers advancing through ERCOT's interconnection process, to be completed before any project moves forward. His office was direct about the consequence: a project that fails to satisfy the requirements set by the PUCT and ERCOT will be denied connection to the grid.

ERCOT stopped classification the same day, filed for good-cause exceptions on August 10, received them on August 20, and issued conditional classifications on September 3 after missing two of its own deadlines.

The audit runs as two legally distinct workstreams, and the distinction matters.

The Batch Zero eligibility verification tests the attestations. ERCOT has said it will remove a project where an attestation was materially false, or where the entity fails to respond within the specified time. That is the workstream with teeth today.

The Community Impact Audit reaches every computational load of 25 MW or more that has not yet energized, whether or not it is in Batch Zero, which ERCOT has put at hundreds of facilities and roughly 8,800 MW. It is a data collection exercise, and ERCOT intends to hand the result to the Commission, the Governor and the Legislature.

What it collects is the part that has been underread. Tax incentives and public financial assistance. Ownership and controlling interests. Projected peak and annual electricity consumption. Onsite generation. Water sources. Cooling technology, specifically whether the facility will be air-cooled, closed-loop or another water-efficient system. And measures addressing noise, light, traffic, setbacks and emergency response.

None of that appears anywhere in the Base Load test. All of it is now on a form, per project, headed for a Legislature that convenes in January 2027.

Whether the second workstream becomes an eligibility screen is unresolved. The TWDB and PUCT testimony above is the first public signal that the line between the two is already blurring in practice.

Four things the classification says nothing about

Water and cooling. Of the responses the Texas Water Development Board did receive to its survey, almost all Texas data centers purchase water from public systems: roughly 90 percent surface water, 8 percent groundwater, 2 percent unknown. The draft 2027 State Water Plan projects existing supplies declining about 10 percent, from 15.5 million acre-feet to nearly 14 million, between 2030 and 2080, primarily from aquifer depletion.

Holland & Knight's read of the September 1 hearing is that the committee signalled intent to prohibit evaporative cooling, require monthly water-use reporting, water-source disclosures and drought contingency plans, expand groundwater district and county authority, and repeal the data center sales tax exemption. The Texas Alliance of Groundwater Districts has told legislators its member districts are not receiving accurate water data for long-term planning and are wary of litigation if they challenge development plans.

So a project drawing groundwater has now put its water source and cooling design on a state form, and it has done so months before a legislative session that has been asked to give groundwater districts more power to say no.

No onsite generation. The PUCT's stated direction, well before the audit, was to reallocate transmission cost responsibility and to give data centers a choice between providing their own generation and paying for the transmission upgrades their load requires. PGRR145 built optional pathways for projects willing to be flexible, through Provisional Controllable Load Resource and Withdrawal-Limited Private Use Network designations. A project that is pure, inflexible, grid-dependent load has elected neither, and now appears on the audit form under onsite generation with nothing to report.

That is not a disqualification. It is a position at the wrong end of every cost-allocation debate currently open at the Commission and in the interim hearings.

Public money. Tax incentives and public financial assistance are collected per project. The Texas qualified data center sales and use tax exemption is under explicit legislative scrutiny, with reform and repeal both on the table. Any project whose economics depend on that exemption, on a municipal abatement, or on a city or county funding infrastructure build-out, now has that dependency itemised ahead of a session where legislators have been openly critical of local incentive behaviour. Representative John Smithee questioned how well local officials are regulating data center water use, describing cities and counties as "hungry for economic development."

Unincorporated siting. Texas counties hold very limited land use authority, which has made unincorporated county land the path of least resistance for data center siting. Expanded county authority to require community hearings, buffer zones and sound and lighting standards is on the record as legislative intent, alongside expanded groundwater district authority.

So the jurisdictional advantage of unincorporated siting is real today and is a named legislative target for January. A project that chose county land specifically to avoid a municipal process has no vested protection if a standard arrives, and fewer local relationships to draw on when it does.

The delivery dates in the market came before the study, and the study has no completion date

Beacon Point targets initial energization in Q1 2027, a first data hall in Q3 2027, and a Phase 2 first hall in Q2 2028 that is not yet financed. IREN targets 300 MW gross at Sweetwater 1 in Q4 2027, with its high-voltage substation already energized. Galaxy says Helios II remains on schedule to energize in 2028.

Against that, ERCOT's SVP for regulatory policy told the PUCT on August 20 that the Batch Zero study will not be finished by April 9, 2027. That date has not been waived, replaced or extended, and it does not appear in the good-cause relief granted the same day. Final transmission planning is now expected in fall 2027. ERCOT is working toward a December 10, 2026 filing, with verification RFI rounds running through October and November.

Every delivery date in circulation was set before a classification framework existed, and none of them sits behind a completed study. You can build a schedule around a deadline, even a bad one. You cannot build a schedule around a process that has no end date at all.

We only know the outcomes of the projects that chose to tell us

The visible sample is three listed sponsors with a reason to announce good news. ERCOT is keeping classifications private. Nobody has issued a release saying they were excluded and are heading to alternative dispute resolution under Protocol Section 20.4. December 10 is the first look at the real distribution.

What is visible on the way in is harsher. Of roughly 474 GW of large load requests in ERCOT, 315 projects carrying about 274 GW were placed outside Batch Zero for lacking required studies, before the audit began. Forty-one requests sought exemptions from the eligibility rules, and 21 of those asked to skip the interconnection study requirement outright. Of 290 dynamic models ERCOT reviewed, roughly 18 percent were acceptable on first submission.

Those three figures come from trade coverage of ERCOT's August 20 presentation rather than from a published document, and we flag them as reported rather than as ERCOT-published. If they are close to right, they carry the argument. Each of those was a paperwork test. The screen that removed 274 GW checked whether required studies existed. The dynamic model round checked whether submissions were complete. The verification audit now running is the third such test in a year, and it is being applied to the projects that already passed the first two.

What belongs in diligence now

Which tier, and can the seller show you the notice? Base Load and Studied Load are different assets with different cost structures. A megawatt number without a tier is a request.

Which agreement supports it, 9.7.2 or 9.7.1? That is the document that produced the outcome. Read it rather than accepting a characterisation of it.

Who is the ILLE? Not who is the tenant. The entity that swore the Section 9.2.1 attestations is the entity ERCOT can remove from the batch, and a lease does not stand behind a filing.

What is the cooling architecture and water source, in writing, and did the project respond to the TWDB and PUCT surveys? On the testimony above, that response is already functioning as an interconnection prerequisite.

Does the project carry onsite generation or any flexibility election? If not, it sits at the wrong end of every cost-allocation proceeding now open.

What public money is in the capital stack, and what happens if the sales tax exemption is repealed? On a capital base at this scale that is a nine-figure sensitivity, and it is a live bill topic.

Is the site in an unincorporated county, and what protects it if county authority expands? The reason unincorporated land is easy to permit today is exactly the gap the Legislature has been asked to close.

The sponsors most exposed are not the ones with indentures and investor relations teams. They are the ones who attested in good faith to a project that has since changed. A site swapped. An ownership chain restructured. A megawatt number revised after the qualification date. A cooling design not yet settled. None of that is fraud. All of it is an inconsistency to be explained inside a window measured in business days, with no cure period written into the attestation review, only a reasonable opportunity to explain.

---