Robots Atlas>ROBOTS ATLAS
Infrastructure

Stargate

2025ActiveUpdated: 19 August 2026Published
Key innovation
An unprecedented-scale program (up to $500B) to build dedicated AI data-center infrastructure in the US, formed by a consortium of OpenAI, SoftBank, Oracle and partners.
Category
Infrastructure
Abstraction level
System
Operation level
Deployment
Use cases
Frontier-scale model trainingProduction-scale inference for OpenAI and partnersSecuring AI compute and power in the USExpanding national AI infrastructure

How it works

The consortium combines capital (SoftBank, MGX, Oracle), a model operator (OpenAI) and technology providers (incl. NVIDIA, Oracle, Arm, Microsoft) to finance and build AI data-center campuses together with power. The infrastructure is to be filled with accelerators (GPUs) and used for OpenAI workloads and more broadly. The project proceeds in phases — from initial campuses (e.g. in Texas) toward a target network of US facilities.

Problem solved

Training and deploying frontier AI models requires compute and power beyond existing resources. Stargate aims to address this shortage via coordinated, large-scale infrastructure buildout.

Key mechanisms

A capital-and-technology consortium (OpenAI, SoftBank, Oracle, MGX)
Up to $500B invested over several years
Building data-center campuses + power
Filling facilities with GPU accelerators
Phased execution (first campuses in Texas)

Strengths & limitations

Strengths
✓Unprecedented scale of compute
✓Coordinated capital and technology
✓Strengthening the US position in AI
✓An integrated compute-plus-power approach
Limitations
✗Enormous capital and execution risk
✗Dependence on GPU and power supply
✗Doubts about the full amount and timeline
✗Environmental and energy impact

Components

ConsortiumOrganizational structure

OpenAI (operator), SoftBank and MGX (capital), Oracle (cloud/data center).

Data-center campusesPhysical infrastructure

Large-scale compute facilities (first ones e.g. in Texas).

Official

Power and energyEnergy

Dedicated power for AI data centers.

Official

Implementation

Implementation pitfalls
Financial and timeline riskHigh

The stated amount and pace may not be fully realized.

Fix:Phasing, hard milestones, diversified financing.
Power constraintsHigh

Power demand may exceed local grid capacity.

Fix:Dedicated power, PPAs, siting near generation.
GPU supply dependenceMedium

The scale requires huge, timely accelerator supply.

Fix:Supplier contracts, hardware diversification.

Evolution

2025
Stargate announced
Inflection point

January 21, 2025 — OpenAI, SoftBank and Oracle announce the up-to-$500B program.

2025
First campuses and partners

Construction begins (e.g. Abilene, Texas) and the partner set expands.

Hyperparameters (configurable axes)

Investment scaleCritical
~100 mld USD początkowoInitial phase.
do 500 mld USDTarget over several years.
PartnersHigh
OpenAI, SoftBank, Oracle, MGXPrincipals.
NVIDIA, Microsoft, ArmTechnology partners.

Hardware requirements

Primary

Stargate campuses are filled with GPU accelerators for training/inference.