Supercomputers/Systems/Stargate Abilene
USA · operational · ai training
Stargate Abilene
Deep dive: Stargate Abilene: 1.2 GW, four of eight live
Also known as OpenAI/Oracle Stargate flagship
Operated by Oracle at Lancium Clean Campus (Abilene) .
Phase 1 seed dataset, compiled by hand. These rows were built from public operator, laboratory and vendor sources. A mechanical second-reader pass has since fetched every cited source: 1092 of 1153 systems have a readable citation that names them, and 40 are genuinely weakly sourced. Every claim carries its source and a confidence tier. Treat anything below verified as a lead, not a citation.
Measured performance
- Rmax
- -
- Rpeak
- -
- Rmax ÷ Rpeak
- -
- Cores
- -
- Power
- -
- Per watt
- - GF/W
Figures are as last publicly reported for the configuration described below, not a live measurement. Where a system was upgraded in place, the post-upgrade configuration is shown and the earlier one appears in the timeline.
What this machine is made of
One row per supplier relationship. “Supplier at build” is the company that shipped the part at the time; where that company has since been acquired, the parent it rolls up to today is shown beside it. That distinction is what makes ticker-level aggregation possible across a thirty-year dataset.
| Role | Part | Supplier at build | Quantity | Confidence | Source |
|---|---|---|---|---|---|
| Integrator | - Designed and built by Crusoe at the Lancium Clean Campus | Crusoe | - | Reported | crusoe.ai |
| Accelerator | NVIDIA GB200 NVL72 Accelerators Arm CPU + CUDA GPU · TSMC 4NP Oracle began delivering the first NVIDIA GB200 racks in June 2025 | NVIDIA | - | Verified | crusoe.aigroup.softbank |
| Cooling | Direct-to-chip cold plate Cooling Cold plate on the package, air for the remainder · Removes roughly 50-80 percent of node heat to liquid direct-to-chip liquid cooling via a zero-water evaporation cooling system | Crusoe | - | Reported | crusoe.ai |
AI datacenter metrics
Figures beyond an accelerator count, each with who stated it. What can and cannot be compared across AI datacenters.
| Metric | Value | As of | Basis | Note | Source |
|---|---|---|---|---|---|
| Accelerators | 201,600 accelerators | 2026-05-23 | Third-party estimate | Epoch AI chip estimate, latest per chip type (B200+B300); some rows are company disclosures, others modelled. | epoch.ai |
| Stated investment | 15 USD bn | 2025-05 | Operator stated | Crusoe, Blue Owl and Primary Digital joint venture to fund the 1.2 GW eight-building campus; a financing figure, not an audited build cost | crusoe.ai |
| Facility power | 590 MW | 2026-07-28 | Third-party estimate | Epoch AI modelled gross facility power (IT plus cooling and overhead) at this timeline date. | epoch.ai |
| Facility power | 1,200 MW | 2026-03 | Design target | Crusoe release describes the existing Abilene site as 1.2 GW across eight buildings; design capacity, not measured draw; excludes the new 900 MW Microsoft campus | crusoe.ai |
| IT power | 421 MW | 2026-07-28 | Third-party estimate | Epoch AI modelled IT power from satellite imagery and permits at this timeline date; not operator-stated. | epoch.ai |
| PUE | 1.3 | 2024-12 | Design target | Crusoe Impact Report 2024: data centers designed for average PUE of 1.2 to 1.3, Abilene included; upper bound shown, not measured | cdn.prod.website-files.com |
How it is powered
| Source | Provider | MW | Status | As of | Basis | Note | Source link |
|---|---|---|---|---|---|---|---|
| Grid | - | 1,200 | planned | 2025-03 | Design target | Crusoe states a 1.2 GW grid interconnection (ERCOT) for the campus. Epoch: 200 MW substation built, a 1 GW substation near completion by Jul 2026. | crusoe.aiepoch.ai |
| Grid | American Electric Power | 200 | operating | 2026-07 | Third-party estimate | Epoch: 200 MW substation (built 2022, expanded 2024) serves the first buildings. Epoch lists American Electric Power as the energy company for the site. | epoch.aiepoch.ai |
| On-site gas | Crusoe | 360 | operating | 2026-07 | Regulatory filing | Texas Tribune: 360 MW on-site gas plant of 10 turbines under TCEQ minor permits; a proposed expansion adds 41 turbines. Epoch counts 10 turbines by imagery. | texastribune.orgepoch.ai |
| Battery storage | Crusoe | - | planned | 2025-03 | Operator stated | Crusoe says the Lancium power plan includes large-scale behind-the-meter battery storage; no MW or MWh stated. | crusoe.ai |
| Wind and solar | - | - | planned | 2025-03 | Operator stated | Crusoe cites behind-the-meter solar and use of nearby wind; Epoch lists local wind. No capacity or contract disclosed. | crusoe.aiepoch.ai |
How AI campuses are powered: capacities are what the operator, utility or filing states, not what has been delivered.
The read
Crusoe built the campus and Oracle operates it as OCI capacity for OpenAI; Crusoe and the joint OpenAI, Oracle and SoftBank release both say Oracle began delivering NVIDIA GB200 racks in June 2025 and that the site is running. Campus power capacity is stated as 1.2 GW across eight buildings, which is a design capacity and not a measured draw, so power is blank. No installed accelerator count is stated by any operator (Crusoe says up to 50,000 GB200 NVL72 per building, which is ambiguous between racks and GPUs), so no quantity is recorded, and an analyst estimate that names B200 and B300 parts is not used because it contradicts the operators. Later buildings were still being commissioned in 2026.
Timeline
- 2025-06 First operational source
Site & facility
Lancium Clean Campus (Abilene)
- Location
- Abilene, TX, USA
- Cooling
- Direct-to-chip liquid cooling
- Commissioned
- 2025
Where sources disagree
We record conflicts instead of picking a winner quietly. All open questions.
- Installed GPU count: Up to 50,000 GB200 NVL72 per building (Crusoe) against About 201,600 B200 or B300 (analyst estimate). Crusoe's wording is ambiguous between racks and GPUs and the estimate names different parts, so no count is recorded.
Change history
- 2026-10-04 AI datacenter metrics: Stargate Abilene
- 2026-09-19 Stargate Abilene added, verified on Crusoe and the joint OpenAI, Oracle and SoftBank release
Source check
We fetched this system's own citations and recorded whether each page actually mentions it. This is a corroboration signal, not a fact check, and it is published so you can see how well the sourcing holds up rather than take it on trust.
| Cited source | Result | Found on the page |
|---|---|---|
| group.softbank | names system and part | NVIDIA GB200 NVL72 |
| crusoe.ai | names system and part | NVIDIA GB200 NVL72 |
Checked 2026-10-09 by pnpm verify. Re-run it and the table changes with the web.
Further reading
Inside a node
Why accelerators exist, what the memory hierarchy costs, and how the CPU and accelerator merged onto one package.
Operating a supercomputer
Acceptance testing, why "installed" and "in production" are six months apart, and the machine lifecycle.
Power and cooling
Why air ran out at 20 kW a rack, how warm-water cooling removes the chillers, and what PUE conceals.