OpenAI’s Stargate Data Center Buildout: From Cloud Customer to Infrastructure Builder

Since January, OpenAI has been on a stealthy nationwide hunt for sites to host its ambitious Stargate data centers. Led by Keith Heyde, a former Meta executive and now OpenAI’s head of infrastructure, the company is transitioning from being a cloud customer to an infrastructure builder. This shift is crucial as OpenAI aims to scale up its compute resources to train increasingly powerful AI models.

The Hunt for Stargate Sites

Heyde, who joined OpenAI in October 2024, has been at the forefront of this search. His team has reviewed hundreds of proposals from potential sites across the U.S., prioritizing access to power, scalability, and local support. "My in-between Christmas and New Year’s last year was actually mostly spent looking at sites," Heyde told CNBC, highlighting the intensity of the process.

Key Criteria for Site Selection

• Access to Power: Reliable and abundant electricity is crucial for running the energy-intensive operations of AI supercomputing.
• Scalability: The ability to expand facilities over time without significant constraints.
• Local Support: Community buy-in and favorable regulatory environments are essential for smooth project execution.

While tax incentives play a role, they are "a relatively small part of the decision matrix," according to Heyde. The focus is on practical factors that ensure long-term sustainability and efficiency.

Advanced Stages of Diligence

As of October 2025, around 20 sites are in advanced stages of diligence. These locations span across the Southwest, Midwest, and Southeast, with massive tracts of land under review. One such site is in Abilene, Texas, where construction has already begun.

Transition from Cloud to On-Premise

OpenAI’s decision to build its own data centers marks a significant shift in strategy. Previously, the company relied heavily on cloud services to meet its compute needs. However, as AI models grow more complex and require more processing power, owning and operating physical infrastructure provides several advantages:

• Cost Efficiency: Long-term cost savings by reducing reliance on expensive cloud services.
• Control and Flexibility: Greater control over hardware and software configurations to optimize performance.
• Data Security: Enhanced security measures for sensitive data.

Technical Considerations

The Stargate data centers will be equipped with powerful GPUs (Graphics Processing Units) to handle the computational demands of training large language models. The architecture is designed to support high throughput and low latency, essential for efficient model training and inference.

• Power Management: Advanced cooling systems and energy-efficient designs to manage the high power consumption.
• Network Infrastructure: High-speed interconnects to ensure fast data transfer between nodes.
• Scalability: Modular design allows for easy expansion as compute needs grow.

Impact on the AI Ecosystem

OpenAI’s move into infrastructure building not only supports its own research and development efforts but also has broader implications for the AI ecosystem. By setting a new standard for AI supercomputing, OpenAI could influence other tech giants to follow suit, potentially leading to more innovation in data center design and operations.

Conclusion

Keith Heyde’s journey from Meta to OpenAI underscores the company’s commitment to building the physical infrastructure needed to advance AI research. As the Stargate project progresses, OpenAI is poised to become a leader in both AI development and data center innovation.