Menu
Prefer to listen instead? Here’s the podcast version of this article.
In January 2025, OpenAI, SoftBank, Oracle, and MGX announced the Stargate Project — a sweeping $500 billion commitment over four years to build U.S. AI infrastructure. The aim: scale compute, secure leadership in AI, and support advanced model training at unprecedented scale.
Now, the consortium has just unveiled plans for five new U.S. AI data center sites under that umbrella, expanding the footprint of Stargate. [Reuters] This announcement is a seminal moment in the “AI infrastructure race” — one that will reshape how the U.S. (and potentially the world) powers compute-intensive AI workloads.
Stargate LLC is the joint‑venture set up in early 2025 by OpenAI, SoftBank, Oracle, and MGX. SoftBank takes financial responsibility, OpenAI operational responsibility, while Oracle contributes cloud infrastructure and data center experience.
The announcement described Stargate as more than just data centers — it’s an infrastructure platform: compute, physical power, cooling, networking, energy sourcing, and site development all wrapped together. The initial equity funders include the founding partners, and the structure is intended to attract external infrastructure firms (land, power, A/E firms) through RFPs and RFQs.
Originally, OpenAI said it would start by deploying $100 billion immediately and ramp toward $500 billion over the full period. The buildout is intended to secure U.S. leadership in AI, foster jobs, and deliver strategic compute capacity.
Because this is not a single monolithic entity, but a modular infrastructure umbrella, the five new sites announced now are the next pieces in a much larger roadmap.
The five new data center sites include:
These are in addition to the flagship Abilene, Texas site, which is already partially operational under Oracle Cloud Infrastructure.
The new additions bring the planned capacity of Stargate to nearly 7 gigawatts (GW) of compute infrastructure. The long‑term target remains 10 GW.
The consortium expects these new sites plus expansions (e.g. 600 MW near Abilene) will generate over 25,000 onsite jobs, plus many thousands more indirectly. The capital committed for these is already crossing $400 billion in the near term, with room to hit the full $500 billion goal sooner than originally scheduled.
Oracle is bringing its cloud infrastructure muscle: the Abilene expansion already sees use of NVIDIA GB200 racks, and Oracle is building eight halls there. [WIRED] In some new sites, SoftBank’s SB Energy will provide power infrastructure (renewables + managed energy).
The idea is that compute, cooling, power, and networking are co‑optimized, giving a more efficient “AI factory” model.
7 GW of compute is massive. To power all this, Stargate will require large power generation, distribution, and cooling systems. The Abilene campus alone is projected to draw ~900 MW. Cooling in drought‑sensitive West Texas is a hurdle; the project uses closed‑loop water cooling and aims to lean on renewables, but will still rely on gas generation initially. [AP News]
Scaling infrastructure — fiber, redundancy, site preparation, geotechnical work — across multiple states will stress supply chains, permitting, and local grid capacities.
Though the headline number is $500 billion, actually raising and deploying that capital is nontrivial. Some critics argue that no funds had been fully committed or deployed initially. Managing debt, equity, cost overruns, and macroeconomic cycles will be a continual risk.
The partnership with NVIDIA is key: Nvidia has reportedly committed up to $100 billion in chip supply and infrastructure support for AI compute. [Financial Times] This helps reduce risk on the hardware side, but still leaves large infrastructure, site, and operations risks.
Large data centers often attract regulatory, environmental, and community pushback: water use, light/noise pollution, land disruption, and habitat loss. The Abilene site already faces concerns from local residents over lighting, energy, and resource stress. Compliance across multiple states with different permitting regimes, environmental rules, and utilities adds complexity.
Sustainability is a core narrative they will need to deliver. If the project is perceived as wasteful or environmentally damaging, it could suffer public backlash.
Ensuring that all sites begin operations in a timely manner, interconnecting the compute fabric, and amortizing costs over long periods are execution challenges at scale never attempted before. There’s always risk of delays, supply constraints, or cost overruns.
Historically, OpenAI has relied heavily on Microsoft Azure for compute partnerships. With Stargate, OpenAI is diversifying compute dependencies and giving Oracle and SoftBank prominent roles.
This gives OpenAI more flexibility, negotiating leverage, and redundancy (but also more risk).
As AI models scale (LLMs, multimodal models, foundation models), compute becomes a limiting factor. The entity that controls compute capacity — especially with efficiency and energy innovation — gains a competitive moat. Stargate signals that infrastructure is becoming as strategic as algorithms.
Announcing such investment at the White House, with Trump in January, underscores the geopolitical angle: AI leadership is now national infrastructure. Accelerating U.S. compute capacity helps the U.S. remain competitive against China and others.
Oracle becomes a more powerful AI cloud player. Other cloud providers (AWS, Google Cloud) will need to respond by improving their infrastructure, pricing, and specialization for AI. Edge and regional computing players could face consolidation or partnerships to stay viable.
Project Stargate is not just another infrastructure build — it’s a statement of ambition. By committing hundreds of billions and deploying gigawatt-scale compute, OpenAI, Oracle, and SoftBank are reshaping the playing field for how advanced AI will be built and run. Yet success is far from guaranteed: energy, financing, regulation, execution, and public buy‑in are real hurdles.
WEBINAR
