OpenAI and Oracle to Build 4.5GW AI Data Centre Capacity
OpenAI has announced a major partnership with Oracle to expand its AI-focused Stargate data centre programme, adding 4.5 gigawatts (GW) of new compute capacity.
The deal is one of the largest AI infrastructure buildouts to date, supporting the scale of training and deploying large language models (LLMs) while addressing the industry's demand for high-performance data centre infrastructure.
First launched in early 2024, Stargate is OpenAI’s response to the shortage of compute power facing the AI sector. Building and operating frontier AI models requires sustained use of thousands of high-density chips, with data centre facilities built specifically for this purpose.
OpenAI’s first Stargate facility is already under construction in Abilene, Texas. Combined with the Oracle-backed expansion, the company is on track to surpass 5GW in total capacity, supporting more than two million AI-specialised chips.
This latest project moves OpenAI closer to its public goal of investing US$500bn in 10GW of US-based AI infrastructure over four years.
Oracle deal brings scale and jobs
The announcement was made jointly with Oracle at the White House as part of a broader initiative involving leading technology companies.
For OpenAI, the deal shows the strategic importance of controlling physical infrastructure to match the growing compute demands of advanced models.
“It’s easy to throw around numbers, but this is a gigantic infrastructure project,” says Sam Altman, CEO of OpenAI.
According to estimates shared by the company, the 4.5GW expansion will generate over 100,000 jobs across the US. These roles will span construction, operations, logistics and long-term maintenance of the data centre facilities.
That includes electricians, technicians, equipment operators and other skilled labour roles required for hyperscale data centre builds. At the Abilene site alone, workers have arrived from more than 20 states, showing the national scale of recruitment.
Construction work has already begun at the Abilene facility, with Oracle confirming it has started delivering Nvidia GB200 racks to the site. These chips are designed for high-density AI training and inference and are a critical component in large-scale LLM deployments.
Infrastructure partnerships extending beyond Oracle
The Oracle deal is one part of a much larger network of infrastructure collaborations forming around the Stargate initiative. OpenAI continues to work with Microsoft, which remains a key cloud partner, while also expanding its reach through partnerships with SoftBank and CoreWeave.
SoftBank is supporting site selection and contributing to facility design tailored for AI workloads, while CoreWeave specialises in supplying GPU resources for training AI models.
[It’s] easy to throw around numbers, but this is a gigantic infrastructure project.
This multivendor approach helps OpenAI diversify its infrastructure pipeline while maintaining control over core compute capabilities. It also ensures the company has flexibility in how it sources hardware, data centre capacity and energy, all of which are critical to meeting future AI needs.
Internationally, OpenAI is expanding Stargate through its “OpenAI for Countries” initiative, supporting foreign governments interested in building national-scale AI infrastructure. The company has indicated that AI deployment at scale requires global infrastructure, and Stargate will be part of that effort.
Political support and long-term goals
The White House has positioned Stargate and similar projects as key to the US’ long-term competitiveness in AI and computing. With government backing and private sector investment aligning, the momentum behind these infrastructure projects has increased.
“Together, we’re committed to delivering 10GW of new compute capacity through Stargate — one of the most ambitious infrastructure projects in US history,” Oracle and OpenAI said in a joint statement to CNBC.
“We’re moving with urgency on site assessments and reimagining how data centres are designed to power advanced AI and make its benefits widely accessible.”
The Oracle partnership also highlights the shift in data centre development toward AI-specific use cases. These facilities are designed from the outset to handle the power and thermal demands of AI hardware, including large-scale deployments of liquid-cooled GPU clusters.
With more partnerships lined up and facilities already delivering hardware, OpenAI’s infrastructure vision is moving from concept to execution.
The challenge now is scaling operations while keeping pace with the technical requirements of the most advanced AI models being built today.
