Emerging AI infrastructure and ERCOT base-load capacity: ERCOT base-load scarcity could lift AI-data-center lease pricing and favor power-secured miners

Overview

The report assesses ERCOT's initial Batch Zero classifications for large Texas power-load projects and argues that conditional base-load status materially improves visibility on power availability and project timing for AI-infrastructure developers. Bernstein believes the scarcity of approved sites should support lease pricing and favor covered operators with secured power capacity.

Core views

ERCOT's Batch Zero screening has created a clearer division between projects with conditional base-load status and those still subject to system-level study. Bernstein states that emerging AI-infrastructure developers have disclosed 8 GW of conditional base-load capacity and another 7 GW of conditional studied-load capacity. A base-load designation gives a project greater near-term visibility on power availability and execution timing, allows the site to enter lease negotiations, and is expected by Bernstein to become increasingly valuable as new data-center development faces political scrutiny, moratoriums and state-level directives. Of the 8 GW conditionally classified as base load, more than 5 GW is available for new AI deployment contracts. Bernstein highlights IREN's 2 GW Sweetwater campus and CIFR's 1 GW Colchis campus as two large available sites. It also notes that the screening process has reduced expected base-load allocations for some proposals: CIFR submitted 3.6 GW of potential Batch Zero sites but received only about 1 GW of incremental approved capacity. This reinforces the report's view that already approved capacity is scarce rather than interchangeable with prospective pipeline capacity. The report distinguishes the three Batch Zero outcomes. Base-load projects include operating facilities and advanced developments supported by qualifying prior studies; they are not subject to further evaluation or MW reallocation. Studied-load projects require additional system-level analysis to establish the amount and timing of capacity, with no assurance that the full requested load will be granted. Excluded projects must wait for later application rounds, with Batch 1 submissions expected to start in summer 2027. ERCOT completed its initial screening on September 3, but classifications remain conditional on requirements including Governor Abbott's audit; ERCOT expects a formal Batch Zero verification report in December 2026. Studied-load allocation is not expected to be finalized until at least April 2027. Across individual operators, Bernstein cites conditional base-load capacity at IREN's 2 GW Sweetwater hub; CIFR's 1.1 GW across Colchis and Stingray, alongside 2.1 GW of studied load; CLSK's 285 MW Sealy and 300 MW Brazoria sites; CORZ's 431 MW Hunt and 300 MW Pecos sites; MARA's roughly 280 MW Granbury and roughly 200 MW Garden City sites; and Hut 8's 1 GW Beacon Point campus. CORZ also has a further 300 MW Pecos expansion classified as studied load, while 371 MW at Denton is not subject to Batch Zero. MARA's up-to-2 GW Matagorda site is classified as studied load. Bernstein's broader thesis is that grid regulation can decongest the queue by removing speculative positions and elevate the value of genuine projects with development history. It therefore favors covered AI-infrastructure players with secured power and strong uncontracted pipelines, arguing that conditional base-load capacity enhances their competitive position. The report also notes that crypto miners have roughly 32 GW of planned power capacity across key data-center markets and have contracted roughly 9 GW with hyperscalers, frontier AI labs, neoclouds and AI-chip manufacturers over more than 20 deals worth over $180 billion during the past two years. Bernstein maintains Outperform ratings on WULF, CIFR, IREN, CORZ, RIOT and CLSK, while MARA is Market-Perform.

Analysis framework

Bernstein reviews ERCOT's project classifications, separates capacity with protected base-load status from capacity still awaiting study, and compares the disclosed power portfolios of covered miners. It then links scarcity of approved, immediately leaseable capacity to project execution visibility, competitive positioning and potential data-center lease pricing.

Methodology notes

• Industry AnalysisSupply-demand framework

  Power-capacity scarcity and AI-data-center demand

  The report treats conditional base-load capacity as scarce supply relative to demand for AI infrastructure, arguing that this scarcity can increase the value of approved sites and support lease pricing.

• Industry AnalysisUpstream-Midstream-Downstream Transmission

  Transmission from ERCOT classification to data-center leasing economics

  ERCOT classifications affect power availability and development timelines, which the report links to miners' ability to offer powered data-center capacity to AI customers.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• IREN (IREN)

  Covered AI-infrastructure operator with a conditionally approved 2 GW Sweetwater base-load campus.

  Strengths

  Large approved Sweetwater capacity and an Outperform rating with a $100 target price.

  Comparison

  Sweetwater is one of two highlighted gigawatt-scale campuses available for AI deployment contracts.

  Risks

  Conditional classification remains subject to required verification and audit conditions.

• Cipher Mining (CIFR)

  Covered operator with conditional base-load capacity at Colchis and Stingray, plus studied-load projects.

  Strengths

  Approximately 1.1 GW conditionally classified as base load; Outperform rating and $32 target price.

  Weaknesses

  Only about 1 GW of incremental approval was received despite 3.6 GW of potential Batch Zero submissions.

  Comparison

  Colchis is highlighted as a 1 GW campus available for AI deployments.

  Risks

  Its 2.1 GW of studied-load capacity is not assured of full allocation.

• Core Scientific (CORZ)

  Covered operator with conditional base-load capacity at Hunt and Pecos.

  Strengths

  431 MW at Hunt and 300 MW at Pecos classified as base load; Outperform rating and $32 target price.

  Weaknesses

  An additional 300 MW Pecos expansion is studied load.

  Comparison

  Denton's 371 MW is approved but not subject to Batch Zero.

  Risks

  Studied-load expansion capacity remains subject to further system-level analysis.

• TeraWulf (WULF)

  Covered AI-infrastructure operator favored by Bernstein for secured power and pipeline positioning.

  Strengths

  Outperform rating and $36 target price.

  Comparison

  Included among Bernstein's preferred covered operators with secured power and uncontracted pipelines.

• Riot Platforms (RIOT)

  Covered AI-infrastructure operator favored by Bernstein in the evolving ERCOT framework.

  Strengths

  Outperform rating and $35 target price.

  Comparison

  Included among covered operators whose secured power position is viewed favorably.

• CleanSpark (CLSK)

  Covered operator with 285 MW at Sealy and 300 MW at Brazoria classified as base load.

  Strengths

  Conditional base-load capacity across the Texas portfolio and an Outperform rating with a $24 target price.

  Weaknesses

  The 300 MW Brazoria Phase 2 project is classified as studied load.

  Comparison

  Its Texas portfolio is under an exclusivity arrangement with the Sandersville tenant.

  Risks

  Studied-load capacity requires further review and may not receive its full requested allocation.

• MARA Holdings (MARA)

  Covered operator with conditional base-load sites at Granbury and Garden City.

  Strengths

  Approximately 280 MW at Granbury and approximately 200 MW at Garden City classified as base load.

  Weaknesses

  MARA is rated Market-Perform with a $17 target price.

  Comparison

  Its up-to-2 GW Matagorda site is studied load rather than base load.

  Risks

  Matagorda's requested capacity remains subject to system-level allocation.

Key data

• Conditional ERCOT base-load capacity disclosed8 GWTotal disclosed by emerging AI-infrastructure developers.
• Conditional ERCOT studied-load capacity disclosed7 GWCapacity requiring further system-level analysis.
• Base-load capacity available for new AI contractsMore than 5 GWIncludes IREN's 2 GW Sweetwater campus and CIFR's 1 GW Colchis campus.
• CIFR potential Batch Zero submissions versus incremental approval3.6 GW submitted; about 1 GW approvedIllustrates lower base-load allocation than initially expected.
• Miners' planned power portfolioApproximately 32 GWAcross key data-center markets.
• Power contracted to AI-related customers over the past two yearsApproximately 9 GW across 20+ deals worth over $180 billionContracts were with hyperscalers, frontier AI labs, neoclouds and AI-chip manufacturers.

Impact & implications

Bernstein believes conditional base-load capacity is becoming a differentiating asset because it offers greater certainty on power availability and project timing than studied-load capacity. The institution expects this advantage to improve the relative position of operators with secured, uncontracted power pipelines and to support higher lease pricing for qualifying sites.

Risks

• Conditional Batch Zero classifications remain subject to Governor Abbott's audit and other requirements.
• Studied-load projects have no assurance that ERCOT will grant the full requested load or approve it on a given timetable.
• Projects reclassified from base load to studied load may receive less approved capacity than developers initially expected.

What to watch

• ERCOT's formal Batch Zero verification report, currently expected in December 2026.
• Final studied-load allocation, which is not expected until at least April 2027.
• Whether scarcity of approved base-load sites translates into higher lease pricing and additional AI-data-center contracts.
• Future Batch 1 submissions, expected to begin in summer 2027.