Feasibility File · OpenAI · Oracle
Abilene · TX
The strongest public Stargate proof point — partially operating, materially advanced, but not yet a fully de-risked ~1.2 GW AI campus.
Verdict
Bearing Labs feasibility score
Treat Stargate I as the strongest public Stargate proof point, but not yet a fully de-risked 1.2 GW AI campus. The appropriate posture is proceed-with-gates: credit the operating tranche and construction momentum, but hold the overall score in the low-5 range. Score should remain 5.2 until specific named evidence is public: ERCOT/AEP full-load approval and SB6 compliance; TCEQ final air authorization for the actual on-site generation fleet; a full-campus water balance and supply/wastewater commitments; building-by-building closeout for buildings 3–8; and OpenAI/Oracle offtake terms sufficient to prove durable use of the retained ~1.2 GW campus.
Where the story diverges
The existing ~1.2 GW Abilene flagship reaches full buildout through a hybrid of on-site natural-gas generation and ERCOT grid power, with ~875–1,000 acres, eight ~500,000 sq ft buildings, and roughly $15B invested at the existing site.
Oracle operates the campus for OpenAI workloads, with SoftBank and MGX as program-level Stargate partners. The retained ~1.2 GW campus continues to buildout while the separate 600 MW Abilene expansion was abandoned in March 2026.
Stargate I is not a paper campus. Roughly 0.3 GW is described as live as of April 2026, TCEQ records confirm on-site generating units under Standard Permit 177263, and high-voltage context around Abilene is real — including 345 kV infrastructure and AEP Texas North presence. The retained ~$15B / 1.2 GW scope sits in the Feasibility Index's 'plausible but unproven' band because structural support exists — land, sponsors, partial operation, TCEQ activity, high-voltage context — while the decisive full-load proof remains non-public.
ERCOT's April 2026 large-load update reports ~410 GW of large loads seeking interconnection (~87% data centers), with post-2026 inclusion conditional on executed interconnection agreements satisfying PUCT Project 58481. DOE and CISA/NIAC transformer-resilience materials cite 36-month average and up to 60-month maximum transformer lead times. Public evidence reviewed did not show full-scale turbine procurement, gas supply, transformer delivery, completed ERCOT/AEP studies, SB6 compliance, or building-by-building closeout for buildings 3–8. The abandoned 600 MW expansion is contained but adverse — it supports a haircut, not a failure call.
Risk 01
Single most decision-relevant exposure
The strongest cross-cutting issue is the mismatch between data-hall construction timelines and power-infrastructure timelines. Data-center buildings can be delivered in 18–24 months; gas generation, transformers, and transmission upgrades can run much longer. Stargate's partial operation proves the first tranche — it does not prove the remaining tranche. ERCOT queue inflation (~410 GW seeking interconnection) shifts the burden of proof onto executed readiness evidence, not queue volume or sponsor ambition.
Claim Scorecard
Each claim assessed independently against supporting and contradicting evidence in the full report.
Meaningful but qualified
Plausible and materially advanced, but not substantiated. Should remain in the 5–6 band until full-load power evidence is public.
Preliminary to meaningful
Physical progress is real, but the mid-2026 full-campus completion claim remains aggressive on the public record.
Preliminary
Closed-loop design is favorable, but the no-material-municipal-impact claim is not yet proven; water is the weakest non-power claim.
Preliminary
Regional power context is favorable for plausibility, but full coexistence under ERCOT West constraints and SB6 remains unproven.
Meaningful but qualified
Expansion abandonment is contained but adverse — supports a haircut, not a failure call on the retained campus.
Meaningful but qualified
Current scope is economically plausible, not bankable on public evidence — one-year slip moves illustrative NPV negative by $1B+.
Preliminary
Demand is credible, but the public record does not prove durable full-campus offtake — Oracle/OpenAI contract terms remain undisclosed.
Meaningful but qualified
Existing operations are likely durable; modifications, renewals, abatements, and expansion headroom remain politically exposed.
Cross-Cutting Risk Factors
On-site gas improves firming, resilience, and ERCOT deliverability — but introduces turbine procurement, gas supply, emissions, TCEQ, community-health, and Title V questions. Parker Hannifin's reported turbine involvement is relevant, but a supplier reference is not the same as public proof of installed, permitted, synchronized, and fueled 1+ GW generation.
ERCOT's April 2026 large-load update is central context. In a 410 GW large-load environment, feasibility credit should attach to executed readiness evidence — interconnection agreements, completed studies, upgrade-cost allocation, and SB6 compliance — rather than queue volume or sponsor narrative.
Closed-loop cooling lowers water use, but water scarcity remains legible to residents. On-site gas lowers grid-dependence, but raises air and health concerns. Together they create a coherent opposition narrative even if each issue is technically manageable.
If OpenAI/Oracle demand is durable, the project can absorb high power costs and infrastructure burdens. If demand softens, the fixed-cost burden of a $15B campus becomes unforgiving. The abandoned 600 MW expansion is therefore both a financial signal and a power-queue signal.
The ~57 permanent-jobs figure cited in the project specification, the federally hosted Oracle fact sheet's broader onsite-jobs language, and local tax-abatement materials appear to describe different scopes or time periods. Until reconciled, community-benefit claims should be treated carefully.
The Abilene / West Texas AI corridor includes Vantage Frontier in nearby Shackelford County and Lancium-hosted capacity. Even if Stargate I clears as a single campus, cumulative load can affect congestion, curtailment, network upgrades, and reliability conditions.
Project Profile
Bearing Labs · Feasibility Brief
Full claim-by-claim assessment with supporting and contradicting evidence, reference-class comparisons, cross-cutting risks, key data gaps, and recommendation.
Companion reports — Power & Grid Path, Permitting & Litigation, Community Impact, and Schedule Forensics — are produced on request.