Decision Model and Forecasting
Bringing Oracle and OpenAI’s Stargate Campus Online in Abilene, Texas
When will the rest of the campus be ready to run OpenAI’s AI systems, and what could delay it?
Stargate Abilene · Crusoe at Lancium’s campus · Oracle infrastructure serving OpenAIOriginal campus public record
Parts of the campus are already operating. Crusoe is expanding the site, while Oracle and OpenAI bring additional AI capacity into service. The remaining buildings need working power and cooling, installed equipment and testing before they can run customer workloads. This model follows the decisions behind that expansion and the dependencies that could affect delivery.
An illustrative decision model grounded in Abilene’s public history. Motivations, unobserved choices and future actions are modeled to explore trade-offs; they are not reports of private deliberations.
Stargate Abilene decision timeline
Select any party to view its public record and request reasoning access.
24 / 24 decision points
● Past decision✓ decided× not taken○ Upcoming◆ Expected★ Best move
Land and power foundation
Land and power foundation
- Past decision point.
Lancium chose Abilene for a campus built around land and power access.
- Past decision point.
Lancium broke ground on a phased campus, initially attracting bitcoin and high-performance-computing interest.
AI build and capital
AI build and capital
- Past decision point.
Local partners amended the development agreements as Lancium invested in land and electrical infrastructure.
- Past decision point.
Crusoe commits to an AI campus
Crusoe and Lancium joined forces on an initial 200 MW AI build with a larger campus planned.
Build the initial AI phaseDefer the AI buildPartner on the AI buildKeep the site flexible for later tenants - Past decision point.
Partners fund the initial phase
Crusoe, Blue Owl and Primary Digital announced funding for the first phase.
Sponsor the initial phaseStage additional commitments - Past decision point.
OpenAI announced Stargate, placing the Texas build within a wider AI infrastructure program.
- Past decision point.
Crusoe starts the six-building expansion
Crusoe began the six-building expansion toward a planned eight-building, 1.2 GW campus.
Build the six-building second phaseSequence build against readiness - Past decision point.
Local partners approved Project Radiance, supporting expansion to eight buildings at Lancium’s campus.
- Past decision point.
Financing grows with the campus
The sponsors expanded the announced joint venture to $15 billion for the larger campus.
Sponsor the expanded campusCoordinate separate readiness tranchesCoordinate one common release gateBack the expanded funding envelopeCondition capital on ready-block milestonesPlan broad procurement for the expanded campusPropose smaller financed construction blocks Workloads and staged delivery
Workloads and staged delivery
- Past decision point.
OpenAI reported early training and inference on OCI; Crusoe reported the first phase operating.
- Past decision point.
Lancium announced a $600 million financing package to support Abilene and further campus development.
- Past decision point.
Crusoe announced a separate Microsoft campus next door and a year-end target for the original campus expansion.
- Past decision point.
Crusoe reported two buildings operating while six more were under construction.
- Past decision point.
Oracle reported 42% of total Abilene capacity delivered in its June update.
- Past decision point.
PUCT approved ERCOT’s Batch Zero framework for large electricity users.
- Past decision point.
Oracle’s September update reports 75% of total capacity delivered, with more to follow.
- Past decision point.
Oracle announced Texas wind investments supplying the grid that serves Abilene.
Remaining delivery decisions
Remaining delivery decisions
- Decision pointUpcoming.
Agree the next safe power ramp
Lancium, ERCOT and PUCT align the evidence and conditions for remaining load.
Select a move to inspect an illustrative alternative. ◆ Expected is the move the model expects. ★ Best move is the best fit to that party's modeled objectives among the alternatives assessed.
Expected for Lancium. The model expects Lancium to make this move, while a different alternative is Best move.Best move for Lancium. The model expects Lancium to make a different move; this alternative is best aligned with Lancium's modeled objectives among the alternatives assessed.Expected for ERCOT. The model expects ERCOT to make this move, while a different alternative is Best move.Best move for ERCOT. The model expects ERCOT to make a different move; this alternative is best aligned with ERCOT's modeled objectives among the alternatives assessed.Expected for PUCT. The model expects PUCT to make this move, while a different alternative is Best move.Best move for PUCT. The model expects PUCT to make a different move; this alternative is best aligned with PUCT's modeled objectives among the alternatives assessed. - Scheduled milestoneUpcoming.
ERCOT’s published request sets a response deadline for covered data-center developers.
- Decision pointUpcoming.
Choose how to finish the remaining buildings
Crusoe and its capital partners balance headline progress with ready-to-use handovers.
Expected for Crusoe. The model expects Crusoe to make this move, while a different alternative is Best move.Best move for Crusoe. The model expects Crusoe to make a different move; this alternative is best aligned with Crusoe's modeled objectives among the alternatives assessed.Expected and Best move for Blue Owl. The model expects this move, and it is best aligned with Blue Owl's modeled objectives among the alternatives assessed.Expected for Primary Digital. The model expects Primary Digital to make this move, while a different alternative is Best move.Best move for Primary Digital. The model expects Primary Digital to make a different move; this alternative is best aligned with Primary Digital's modeled objectives among the alternatives assessed. - Decision pointUpcoming.
Resolve operating conditions and local concerns
The developer and authorities address operating conditions and local-impact questions that could affect commissioning.
Expected for TCEQ. The model expects TCEQ to make this move, while a different alternative is Best move.Best move for TCEQ. The model expects TCEQ to make a different move; this alternative is best aligned with TCEQ's modeled objectives among the alternatives assessed.Expected for Abilene. The model expects Abilene to make this move, while a different alternative is Best move.Best move for Abilene. The model expects Abilene to make a different move; this alternative is best aligned with Abilene's modeled objectives among the alternatives assessed. - Decision pointUpcoming.
Decide which capacity is ready for customer work
Oracle and OpenAI match tested capacity to workloads while remaining delivery continues.
Expected for Oracle. The model expects Oracle to make this move, while a different alternative is Best move.Best move for Oracle. The model expects Oracle to make a different move; this alternative is best aligned with Oracle's modeled objectives among the alternatives assessed.Expected for OpenAI. The model expects OpenAI to make this move, while a different alternative is Best move.Best move for OpenAI. The model expects OpenAI to make a different move; this alternative is best aligned with OpenAI's modeled objectives among the alternatives assessed. - Scheduled milestoneUpcoming.
ERCOT plans to report on data-center state and community impacts with PUCT.
- Scheduled milestoneUpcoming.
Crusoe’s announced year-end target provides a checkpoint for the six-building expansion.
Use this decision model in your work.
Bring us into the decision, or request the structured decision data.
Illustrative decision model grounded in public sources. Hypothetical choices and effects are distinct from recorded events.
