Report: Oracle to provide roughly $300B in commercial compute to OpenAI over five years
- Oracle’s reported five-year deal to supply OpenAI with about $300 billion of compute has become the clearest price tag yet on Stargate. - The key detail is the implied run rate — roughly $60 billion a year — on top of an earlier 4.5-gigawatt buildout. - That matters because AI’s bottleneck is no longer model ideas. It’s power, chips, and financing at utility scale.
Cloud compute is the story here — not software in the abstract, but the physical, power-hungry machinery needed to train and run frontier AI. The new wrinkle is a reported Oracle deal to provide OpenAI roughly $300 billion of compute over five years, starting in 2027, which would make it one of the biggest cloud contracts ever discussed. That number lands because it turns a vague AI arms race into something concrete: tens of billions a year, tied to actual data-center capacity. And it plugs directly into Stargate, the infrastructure push OpenAI has been building with Oracle and SoftBank. (datacenterdynamics.com) ### What is the deal, exactly? The reported structure is simple enough: OpenAI would buy about $300 billion of cloud compute from Oracle over roughly five years. The reporting around it says the contract begins in 2027. Oracle and OpenAI had already been linked through Stargate, but this is the first number that makes the scale feel real rather than promotional. (datacenterdynamics.com) ### Why is Oracle in the middle of this? Because Oracle has become one of the few companies willing to build AI infrastructure at extreme scale for a single customer. In July 2025, OpenAI said it had partnered with Oracle to develop 4.5 gigawatts of additional Starga(datacenterdynamics.com)an 5 gigawatts under development and run more than 2 million chips. That is not normal cloud capacity. That is national-infrastructure scale. (openai.com) ### Why does 4.5 gigawatts matter so much? Because gigawatts are the hard constraint now. A normal tech story is about users or revenue. This one is about electricity, land, transformers, cooling, and how many advanced chips you can actually house and keep fed. OpenAI’s own wording around Stargate (openai.com)ters are starting to look more like power projects than classic software deployments. (openai.com) ### How big is $300 billion in yearly terms? It works out to about $60 billion a year. That is the number to keep in your head. Earlier reporting tied OpenAI to an Oracle arrangement worth about $30 billion annually for data-center services tied to the 4.5-gigawatt expansion. The later $300 billion(openai.com). Even by hyperscaler standards, that is enormous. (techcrunch.com) ### Is this fully confirmed? Not in the clean, SEC-style way investors usually want. The $300 billion figure has been widely repeated from report-based coverage, and OpenAI publicly confirmed the 4.5-gigawatt Oracle partnership, but not every commercial term has been d(techcrunch.com)t-driven. (openai.com) ### Why does this change how people think about OpenAI? Because it reframes OpenAI from a model lab into an infrastructure buyer on the scale of a utility customer. If you are committing this much spend, the business challenge is no longer just “build a smarter model.” It is “finance, fill, and mon(openai.com) reacting not just to model launches, but to reports about OpenAI growth, Oracle exposure, and whether demand can absorb all this capacity. (thenextweb.com) ### So what’s the real takeaway? The headline is Oracle and OpenAI. But the deeper story is that frontier AI now runs on industrial inputs — power, chips, construction, and balance sheets. The reported $300 billion figure matters because it shows where the bottleneck moved. It is no longer mainly ideas. It is who can actually build the machine behind them.
