Locator: 49301STARGATE.
One can worry that Oracle is getting over-extended.
Stargate. This is where I track Stargate.
See Barron's update on Stargate: OpenAI, Oracle, and SoftBank.
See the graphic at this post. Knowing most folks won't click on the link, here's the drawing:
Now the explanation:
OpenAI's compute infrastructure
follows a multi-pronged approach, drawing on multiple partners and strategies to power its vast training and inference workloads. Instead of relying solely on a single cloud provider, the company strategically builds and accesses large-scale GPU clusters to handle the immense computational demands of its frontier AI models.
The flow of this compute infrastructure can be broken down into three main categories:
- massive supercomputing projects: Large, multi-partner initiatives focused on building dedicated, high-power data centers.
- strategic cloud partnerships: Accessing and scaling existing cloud resources to meet both research and user-facing needs.
- specialized software and orchestration: The internal tools and systems that manage and optimize OpenAI's computational workloads.
1. Large-scale supercomputing projects: The Stargate initiative:
- the most significant known component of OpenAI's compute flow is the "Stargate" project, a long-term, multi-gigawatt data center initiative aimed at building the infrastructure for future AI. This is the "fuel" that drives the company's biggest breakthroughs.
- partnerships: Stargate is a joint venture that includes OpenAI, Oracle, SoftBank, and NVIDIA.
- physical locations: Projects include a massive data center in Abilene, Texas, that is already operational, with five new U.S. sites announced in late 2025.
- energy and power: The sites are being developed to support immense power requirements, reaching gigawatt-scale capacity and necessitating new energy solutions.
- purpose: This infrastructure is designed to handle the multi-month, continuous computation needed for training the most advanced AI models.
2. Strategic cloud and hardware partnerships:
While Stargate focuses on long-term expansion, OpenAI also relies on strategic partnerships to secure the hardware and cloud resources needed for its day-to-day operations.
-
Microsoft Azure: As a key investor and partner, Microsoft provides access to its Azure AI supercomputers and cloud infrastructure for training and scaling models like GPT-4.5. OpenAI also leverages Azure for enterprise-grade services.
-
NVIDIA: In a recent landmark deal, OpenAI and NVIDIA entered a strategic partnership to deploy at least 10 gigawatts of NVIDIA-powered systems. NVIDIA will also invest up to $100 billion to help fund this infrastructure expansion. The first phase of this new GPU hardware will be based on NVIDIA's Vera Rubin platform and is scheduled for 2026.
-
Oracle: In addition to its role in Stargate, Oracle serves as a cloud infrastructure partner. Oracle is specifically leading the development of several new data center sites. Other providers: OpenAI also contracts with smaller cloud providers to secure additional compute capacity.
3. Internal software and infrastructure orchestration:
Underpinning all this hardware is a sophisticated software stack developed by OpenAI to manage and optimize its compute resources.
- model training workflow: OpenAI engineers work on designing and scaling architectures, including techniques for training large models. This involves managing multi-datacenter training and ensuring fault tolerance across clusters of tens of thousands of GPUs.
-
GPU infrastructure management: Internally, OpenAI's GPU infrastructure team designs and operates the systems that manage the immense GPU fleet. Their work includes:
- building user-friendly scheduling systems to maximize GPU utilization.
- automating Kubernetes (see below) cluster provisioning and upgrades.
- optimizing model startup times by ensuring fast snapshot delivery across storage and hardware caching.
- model serving infrastructure: For inference, or serving its models to users, OpenAI leverages a microservices and containerization architecture. This allows for efficient, low-latency, and highly available deployment of models like ChatGPT for hundreds of millions of users. For enterprise use, many customers also access OpenAI models through Azure's enterprise-ready services.
**************************
More
Kubernetes
Kubernetes, often abbreviated as K8s, is an open-source platform for automating the deployment, scaling, and management of containerized applications. It acts as a container orchestrator, enabling organizations to run applications consistently across different environments by abstracting the underlying infrastructure. Key features include declarative configuration, automated rollouts and rollbacks, self-healing capabilities, and service discovery, making it a powerful tool for modern DevOps and cloud-native development.
In plain English:
Kubernetes (often called K8s) is a free, open-source system that helps you run and manage apps that are packaged into containers (small, self-contained units of software).
It takes care of automatically starting, stopping, and adjusting how many copies of your app are running, so everything stays reliable and balanced. It also hides the messy details of the servers underneath, so your apps can run the same way no matter where they’re hosted — on your laptop, in your company’s data center, or in the cloud.
Some of the useful things it can do include:- setting up and updating apps automatically
- undoing updates if something goes wrong
- restarting apps that crash
- letting apps easily find and talk to each other
In short, Kubernetes is a tool that helps developers and IT teams easily run large, complex applications without constantly managing servers by hand.
**********************************
Disclaimer
Brief
Reminder
Briefly:
- I am
inappropriately exuberant about the Bakken and I am often well out front
of my headlights. I am often appropriately accused of hyperbole when it
comes to the Bakken.
- I am inappropriately exuberant about the US economy and the US market.
- I am also inappropriately exuberant about all things Apple.
- See disclaimer. This is not an investment site.
- Disclaimer:
this is not an investment site. Do not make any investment, financial,
job, career, travel, or relationship decisions based on what you read
here or think you may have read here. All my posts are done quickly:
there will be content and typographical errors. If something appears wrong, it probably is. Feel free to fact check everything.
- If anything on any of my
posts is important to you, go to the source. If/when I find
typographical / content errors, I will correct them.
- Reminder: I am inappropriately exuberant about the Bakken, US economy, and the US market.
- I am also inappropriately exuberant about all things Apple.
- And
now, Nvidia, also. I am also inappropriately exuberant about all things
Nvidia. Nvidia is a metonym for AI and/or the sixth industrial
revolution.
- I've now added Broadcom to the disclaimer. I am also inappropriately exuberant about all things Broadcom.
- Longer version here.
