a16z Texas Data Centers Analysis: Why ERCOT Put the Queue on Hold

a16z says Texas data centers face a new power test after their interconnection queue reached 474 gigawatts, more than five times record peak demand. About 90% of those requests involve data centers. Texas responded by requiring a comprehensive audit before additional projects could advance.

The pause does not mean that 474 gigawatts of facilities are under construction. It means ERCOT, the grid operator for most of Texas, received requests representing that combined demand. Many proposals remain preliminary, duplicated, or unsupported by completed technical studies.

That distinction is the center of the dispute. Texas wants AI investment, but it cannot plan transmission around projects that might never exist. Developers want rapid approvals, but local communities increasingly demand credible answers about electricity, water, noise, ownership, and public subsidies.

The conflict is therefore not simply data centers against regulators. It is speculative demand against buildable demand. Texas must decide which projects deserve scarce grid capacity before residents finance infrastructure for campuses that might disappear.

Texas Paused a Queue That Had Outgrown the Grid

Texas stopped treating every data center request as a credible forecast and started demanding evidence that projects can actually be built.

At the end of 2024, ERCOT had approximately 63 gigawatts of large customers seeking connections. By June 2026, the total had climbed to approximately 474 gigawatts.

ERCOT’s large-load presentation placed data centers at roughly 90% of that queue. More precise figures presented to state lawmakers identified 420.8 gigawatts of data center requests within a 474.7-gigawatt total.

An interconnection queue is a list of proposed projects seeking permission and infrastructure to draw electricity from the grid. It is not a construction schedule or an operating-load measurement.

That difference matters because one developer can investigate several possible sites before choosing a final location. Different developers can also pursue the same prospective customer without disclosing that overlap to planners.

ERCOT cannot simply add every request and build around the result. Transmission lines, substations, transformers, and generation take years to plan and construct. Customers ultimately pay many shared system costs.

The queue also contained projects at sharply different development stages. Some had completed studies and submitted technical models. Others had not provided a qualifying study at all.

In July, ERCOT reported that approximately 205 gigawatts were initially eligible for Batch Zero. Batch Zero is the first coordinated study under Texas’ revised process for large electricity users.

That figure was still substantial, but it was far below the headline queue. ERCOT identified 315 projects without qualifying studies and another 47 without required dynamic models.

A dynamic model estimates how equipment will behave during voltage changes, faults, and other grid disturbances. Without one, ERCOT cannot reliably simulate a proposed facility’s effect on the network.

Texas Governor Greg Abbott then ordered regulators to verify every data center advancing through the interconnection process. His August 3 audit directive required the review before additional projects could move forward.

The order covered more than electrical engineering. It requested details about project ownership, public assistance, power sources, water use, cooling systems, and protections for neighboring properties.

Projects that refuse the review face denial of grid access. ERCOT also delayed the Batch Zero process while regulators determined how to apply the directive.

Calling the action a permanent statewide ban would overstate it. The state paused advancement through a specific process while it checked whether applicants met legal and technical requirements.

Still, the interruption changes the negotiating position of every developer. A queue position alone no longer creates the same expectation of future capacity. Applicants must support their claims with documents, money, models, and credible operating plans.

The a16z Texas data centers argument starts here. The queue became too large and unreliable to function as a planning signal, so Texas converted it into a verification exercise.

Why the Texas Data Center Queue Became Unreliable

The queue grew because applying for electrical capacity created valuable optionality without proving that a project had a customer, financing, or a final site.

A proposed AI campus needs confidence that power will eventually be available. Developers therefore seek interconnection rights early, sometimes before completing commercial agreements or selecting a final location.

That behavior makes sense for each applicant. It creates a much harder problem for the grid operator.

A developer can submit several alternatives because losing every possible site would be costly. Yet ERCOT sees several separate demands, even when only one campus will survive.

The same prospective tenant can appear behind proposals from unrelated developers. A cloud provider evaluating four sites might generate four power requests while intending to build only one.

Texas officials cannot always identify that duplication because project details are commercially sensitive. Confidentiality protects negotiations, but it also obscures the relationship among applications.

Ryan McEntush’s a16z analysis describes this as a queue-quality problem. Many developers submit speculative projects across multiple sites before securing a committed customer.

The result resembles an option market. A position in the queue preserves a chance to obtain scarce power later, even if the holder remains uncertain about construction.

That option can also make land more valuable. A parcel associated with a plausible power allocation may attract buyers or tenants that would otherwise ignore it.

The grid absorbs the downside. Engineers must review applications, test contingencies, and identify upgrades for projects that can vanish before using any electricity.

Planning mistakes work in both directions. If ERCOT accepts inflated demand, Texas can approve unnecessary infrastructure and burden customers with stranded costs. If it discounts genuine projects, viable campuses wait while competitors build elsewhere.

The state’s answer is commercial-readiness verification. That means testing whether an applicant controls its site, has credible financing, possesses technical data, and maintains agreements supporting construction.

This concern extends beyond Texas. An Associated Press investigation found regulators across several states questioning whether utility forecasts contain speculative or duplicated requests.

The Data Center Coalition has also supported stronger methods for assessing commercial readiness. Serious operators do not benefit when weak applications consume engineering attention or damage public trust.

Yet verification is not easy. A legitimate early-stage project might lack a final tenant because customers want power certainty before signing. The power provider may want a signed tenant first.

This creates a sequencing trap. Developers need grid confidence to secure customers, while grid planners need customer commitments to treat demand as real.

Financial security helps, but it cannot answer every question. Well-funded firms can preserve several options, while smaller credible builders might struggle with early commitments.

Documents also prove preparation, not eventual operation. Land rights, equipment orders, and construction contracts can reduce uncertainty without eliminating market risk.

This is why the Texas data center queue should not be read as a conventional backlog. It is a mixture of credible demand, competing site options, exploratory proposals, and incomplete projects.

ERCOT’s task is to transform that mixture into a dependable planning range. The audit is designed to remove obvious weaknesses before engineers allocate transmission capacity.

That process will disappoint applicants who treated the queue as inexpensive optionality. It should benefit projects that already possess land, customers, equipment plans, and credible sources of electricity.

a16z Texas Data Centers Face a Cost Allocation Test

The decisive question is not whether Texas can build more infrastructure, but who pays when a new campus requires it.

A data center cannot reserve a direct electrical path simply by purchasing energy from a generator. Electricity flows across an interconnected network according to physical conditions, not corporate contracts.

A new campus can therefore require upgrades far beyond its property line. Planners may need larger substations, stronger transmission corridors, new transformers, or additional generation support.

Some upgrades serve one project directly. Others improve reliability for existing customers or create capacity for future development.

That overlap makes cost allocation contentious. Charging the first project for everything can deter investment. Socializing every expense can leave residents paying for infrastructure built around uncertain private demand.

Texas lawmakers addressed this tension through Senate Bill 6. The enacted SB 6 text requires qualifying large loads to contribute toward reasonable interconnection costs.

The law generally applies standards to sites exceeding a 75-megawatt threshold. It also requires applicants to disclose substantially similar requests elsewhere in Texas.

Applicants must demonstrate site control and provide financial commitments before receiving full consideration. They must also disclose significant on-site backup generation.

These provisions target the speculative behavior behind queue inflation. A developer should not reserve several large positions without telling planners that only one project will proceed.

However, assigning direct interconnection costs does not settle every dispute. Transmission upgrades often serve several users over many years, making their benefits difficult to divide precisely.

A facility might trigger an upgrade today that later supports housing, factories, batteries, or another data center. Charging the entire expense to the first project could overstate its responsibility.

The reverse problem is equally real. A campus might withdraw after a utility begins procurement, leaving customers exposed to equipment and planning costs.

Large transformers and other grid components have long production timelines. A late cancellation can strand both capital and scarce engineering capacity.

Texas has therefore combined study fees, capacity-linked security, milestone requirements, and possible capacity reallocation. The framework asks developers to place more capital at risk as their projects advance.

The policy creates pressure on two groups. Speculative developers must decide whether an uncertain project deserves additional commitments. Established operators must reveal enough information to distinguish themselves from speculative applicants.

Developers also face a choice between relying on ERCOT and bringing dedicated generation. On-site power can improve a project’s credibility, but it introduces fuel, permitting, emissions, and operational questions.

A gas-powered campus might connect faster while increasing local pollution concerns. A renewable-heavy design needs storage or another source when generation falls.

Nuclear projects promise continuous output but usually require longer development schedules. Batteries respond quickly, yet their duration limits how long they can support a campus.

No single technology removes the cost question. Even a campus with its own generation can depend on transmission for backup power or export electricity into the market.

Colocation creates another conflict. A data center built beside an existing plant can consume output that previously served the wider grid.

Texas permits some colocated arrangements only when facilities can curtail demand or switch to backup supply during emergencies. That requirement preserves ERCOT’s access to existing generation.

The a16z Texas data centers thesis favors infrastructure expansion and flexible consumption over a broad retreat from development. That direction has merit, but it requires enforceable commitments.

Texas cannot plan around a promise that a campus will reduce demand unless operators can verify the response. A contractual obligation needs compatible controls, telemetry, backup systems, and meaningful penalties.

Flexible Data Centers Offer a Path Through the Bottleneck

Projects that can reduce grid demand during stressed hours are easier to accommodate than campuses demanding uninterrupted power under every condition.

Data centers traditionally sell reliability to customers. Their systems must remain available despite electrical failures, hardware problems, or network interruptions.

Grid flexibility asks them to accept a different relationship with electricity. A flexible load can reduce consumption, use on-site generation, or shift computing when ERCOT needs relief.

That does not necessarily mean shutting down customer services. Operators can postpone some training jobs, move workloads to another region, or draw temporarily from batteries.

Not every workload can move. Real-time inference, financial services, healthcare systems, and active consumer applications can require continuous operation.

AI training is often more adjustable, but interruptions still have costs. A training run may involve thousands of accelerators working together, and poor coordination can waste computing time.

A credible flexibility plan must therefore identify which loads can move and how quickly. It must also explain what keeps essential services operating during curtailment.

Senate Bill 6 provides routes for large loads that supply their own power or accept binding curtailment obligations. Some expedited arrangements require commitments lasting at least ten years.

ERCOT’s revised approach studies qualifying projects together. The batch process allocates capacity by year, identifies required upgrades, and clarifies each developer’s financial obligations.

This coordinated view is important because projects interact. Approving one campus can consume transmission capacity needed by another, even when their sites are miles apart.

Batch studies also reduce the distortions created by evaluating applications one at a time. ERCOT can compare several credible loads against the same network constraints.

The mechanism does not create electricity by itself. It determines how much demand the existing and planned system can serve without violating reliability standards.

Grid planners test failures involving lines, transformers, generators, and other equipment. They must keep the network stable even when an important component unexpectedly becomes unavailable.

Data centers introduce another technical concern. Their electronic equipment can disconnect rapidly during a voltage disturbance to protect sensitive hardware.

A simultaneous drop in several large campuses can destabilize the grid. ERCOT has therefore developed voltage ride-through requirements, which require facilities to remain connected during defined disturbances.

Flexibility and ride-through solve opposite problems. Flexibility lets ERCOT reduce load when necessary, while ride-through prevents an uncontrolled reduction at the wrong moment.

The difference is control. Planned curtailment can protect reliability. An unexpected multi-gigawatt disconnection can create a new emergency.

Developers offering flexibility must show that their systems respond to ERCOT instructions within required times. The grid cannot rely on voluntary reductions that arrive after the constraint has passed.

On-site generation faces the same credibility test. A facility must demonstrate fuel availability, operating reliability, synchronization, and compliance with environmental requirements.

Batteries can respond almost instantly and support transitions between grid and on-site power. However, operators must preserve enough stored energy for the periods when ERCOT needs it.

Workload shifting can complement physical infrastructure. A company operating campuses in several regions can redirect suitable computing tasks toward available electricity.

That approach turns location into an operational variable. It can reduce pressure during short grid events without abandoning the Texas facility.

Yet shifting work is not free. Data movement consumes time and energy, while privacy rules or latency requirements can limit available destinations.

The promising model combines several tools. A campus can use batteries for immediate response, generation for longer events, and workload controls for discretionary computing.

Texas can reward such projects with earlier or larger capacity allocations. In return, operators must accept monitoring, testing, and enforceable performance requirements.

That trade offers a route beyond the current pause. It allows serious projects to advance without assuming the grid must serve every requested megawatt continuously.

Community Distrust Turned an Engineering Problem Into a Political One

The grid queue triggered the pause, but weak community engagement made aggressive state intervention easier to justify.

Residents near proposed campuses care about issues that an interconnection study does not fully address. Those include water consumption, generator emissions, construction traffic, light, and continuous equipment noise.

The concerns intensify when developers use code names or concealed ownership structures. Confidentiality can protect land negotiations, but it reduces accountability during tax and zoning decisions.

Hood County officials, for example, considered support for a project identified as “Project Patriot” without initially knowing the company behind it. That approach can feel one-sided to affected residents.

The state’s resource survey produced another warning. According to reporting cited by a16z, only 28 of 377 contacted companies had responded by June.

A low response rate does not prove misconduct by every nonresponding company. It does show that regulators lacked basic information needed for policy and public communication.

Abbott’s audit addresses that gap directly. It requests annual and peak electricity use, water sources, cooling technology, ownership, financial assistance, and community protections.

Those categories reveal how far the dispute has moved beyond grid engineering. Data center developers are now being evaluated as local industrial operators.

For communities, the central question is reciprocity. Residents want to know what a project contributes, what resources it consumes, and who bears the downside.

Construction spending and property taxes can benefit a region. Permanent employment may remain modest compared with the campus footprint and electricity demand.

Tax abatements complicate that calculation. A community can approve incentives before understanding the final operator, resource requirements, or infrastructure costs.

Water is especially sensitive in fast-growing or drought-prone regions. Cooling designs vary widely, so a generic estimate cannot replace project-specific disclosure.

Air-cooled systems can reduce direct water consumption but use more electricity under some conditions. Closed-loop systems recycle water, though they still require initial supply and maintenance.

On-site generators also change the local impact. Equipment intended for emergencies can become an important operating resource when grid power is limited.

Communities need to know how often those generators will run, which fuels they use, and what emissions controls apply. They also need credible noise measurements.

The industry’s strongest argument is that responsible projects should not be grouped with speculative applicants. That position supports the audit if officials conduct it efficiently.

The Data Center Coalition said the review should distinguish committed investors and responsible resource stewards from weaker proposals. It also urged Texas to move quickly.

The skeptical view concerns administrative overreach. A broad pause can delay credible projects alongside speculative ones, damaging investment without resolving transmission shortages.

Political incentives can also shape the process. Electricity bills and neighborhood development are visible issues, especially before elections.

Regulators must therefore publish consistent criteria and timelines. Without transparent standards, verification can become an unpredictable permitting layer.

Texas should not treat community engagement as a public-relations task completed after site selection. Local concerns affect permitting, schedules, financing, and ultimately queue credibility.

A project facing organized opposition is less certain than one with zoning approval and documented agreements. Community readiness belongs in commercial-readiness analysis.

For developers, early disclosure now has practical value. It can identify water, noise, traffic, and emergency-response problems before they trigger a wider political backlash.

The Texas data center queue became a symbol of projects arriving faster than public institutions could evaluate them. Restoring trust requires more than reducing the gigawatt total.

What Will Show Whether the Texas Pause Worked

The policy succeeds only if it produces a smaller, credible pipeline without turning verification into an indefinite barrier to construction.

The first signal is the revised Batch Zero population. ERCOT must identify which projects satisfy its study, model, security, and documentation requirements.

A substantial reduction would support the argument that speculative submissions distorted the queue. A limited reduction would suggest that many developers possess stronger commitments than critics assumed.

The key number is not simply total eligible gigawatts. The distribution by year matters because a phased campus imposes different demands than an immediate full-scale connection.

ERCOT’s allocations should show how much capacity each project can receive between 2028 and 2032. That schedule will reveal the practical size of the near-term bottleneck.

The second signal is new generation tied to credible demand. Data centers that bring dependable supply can reduce concerns that residents will finance capacity built only for private campuses.

Contracts alone are insufficient. Observers should track projects reaching construction, securing equipment, completing permits, and establishing firm operating dates.

Generation type also matters. A gas plant, battery system, renewable portfolio, and nuclear project offer different reliability and development profiles.

If serious data center projects stimulate dependable capacity without removing existing supply from other customers, the investment case becomes stronger. If they only reserve existing plants, opposition will persist.

The third signal is verified flexibility. ERCOT should test whether approved facilities can reduce grid consumption at the promised speed and duration.

Successful tests would strengthen the case for serving more load with existing infrastructure. Failed tests would show that contractual flexibility cannot substitute for physical capacity.

Performance should remain observable after approval. A facility that earns favorable treatment must maintain batteries, generators, controls, and communications throughout its commitment.

These signals will also influence markets outside Texas. Other grid operators face similar uncertainty around AI demand, duplicated applications, and the cost of new infrastructure.

Texas offers a notable experiment because it combines abundant development interest with an unusually isolated power system. ERCOT has fewer connections to neighboring grids than other major operators.

That isolation increases the value of local generation and controllable demand. It also raises the consequences of forecasting errors.

The a16z Texas data centers analysis is ultimately optimistic about technical solutions. It points toward batteries, flexible computing, improved transmission equipment, and better allocation rules.

That optimism should be tested against deployment rather than announcements. Useful technology must appear in approved projects, survive grid tests, and lower costs for other customers.

The queue’s headline size will remain tempting shorthand. Readers should resist interpreting every requested gigawatt as a facility that will be built.

A more useful measure is progression. How many projects secure studies, financial commitments, community approval, generation, and final energization?

The same discipline applies to claims that the pause ended Texas’ AI ambitions. Delays can change project economics, but verification does not automatically eliminate qualified development.

The likely result is selection rather than retreat. Projects with customers, land, power strategies, and local support should gain relative advantage.

Speculative developers will lose some of the optionality created by easy applications. Utilities will receive better information before ordering equipment or proposing transmission upgrades.

Communities should gain more leverage to request details before approving incentives or land-use changes. Regulators will still need to prevent local review from becoming inconsistent.

For AI companies, electricity strategy is becoming part of product capacity. Computing growth depends on power procurement, grid engineering, construction, and political legitimacy.

Executives evaluating Texas should now ask three practical questions. Can the project prove it is commercially real? Can it absorb or finance its grid impact? Can it reduce demand when ERCOT needs help?

Those questions are more useful than asking when the entire 474-gigawatt queue will receive power. Most of that queue was never likely to operate as submitted.

The next few months should reveal whether Texas can separate ambition from duplication without driving committed projects away. Watch Batch Zero allocations, generation construction, and flexibility tests.

If all three produce measurable progress, the pause will look like a queue repair. If timelines keep slipping, it will look more like an unresolved permitting freeze.

Either result will matter beyond Texas. AI infrastructure is forcing power systems to distinguish credible growth from speculative demand, and every major data center market faces that choice.