OpenAI has been confirmed as the mystery client behind the $30 billion deal Oracle announced it had signed earlier this week.
The agreement, which is part of the previously announced Stargate project, represents the largest single cloud deal on record. According to the Financial Times, the deal will see OpenAI lease an additional 4.5 GW of data centre power in the US from Oracle – that is equivalent to about a quarter of the US’s current operational data centre capacity.
As part of the Stargate project, Oracle will be building brand-new facilities across the US to satisfy OpenAI’s need for more computing power. The first of these facilities is already under construction in Abilene, Texas, where capacity is set to almost double from 1.2 GW to roughly 2 GW, while additional campuses are being considered in Texas, Michigan, Wisconsin and Wyoming. There may also be further facilities constructed in New Mexico, Georgia, Ohio and Pennsylvania in future.
The deal with Oracle will be a huge blow to Microsoft, which just yesterday announced massive layoffs impacting 9,000 people across the business. Microsoft’s Azure cloud infrastructure has long been seen as the backbone of OpenAI’s infrastructure, but the AI company has since been diversifying – OpenAI is now using Google Cloud and specialist GPU-as-a-service providers like CoreWeave to reduce its reliance on Azure.
This comes amidst a reported rift between Microsoft and OpenAI, who are both competing to lead in this new AI era, despite Microsoft being the largest single shareholder in OpenAI. Given the increased appetite for computing power from OpenAI, however, Microsoft was forced to relax its exclusivity terms with the company in January 2025. That involved the terms being changed to ‘right of first refusal’, meaning if Microsoft couldn’t meet the demand required by OpenAI, the company would be free to explore elsewhere on the market – that is what led to deals with Google Cloud and now Oracle.
For its part, Oracle is capitalising on the race to deploy data centre capacity to meet the needs of AI. Earlier this year, Larry Ellison, the founder of Oracle, noted that the company would be “the number one builder and operator of cloud infrastructure data centres,” and that it would “build and operate more cloud infrastructure data centres than all of our cloud infrastructure competitors.”
It will achieve that only if it can procure enough GPUs, however – something the company is keen to keep in stock, with it reportedly spending $40 billion to buy around 400,000 of Nvidia’s GB200 chips for its Abilene, Texas data centre.