oracle phone
Oracle expands OpenAI partnership, putting Codex and ChatGPT Work in its marketplace
Oracle Cloud customers can now buy OpenAI models with existing Universal Credits, layering a software deal on top of a $300 billion infrastructure pact
Oracle and OpenAI are deepening a relationship that was already one of the largest in tech. Oracle Cloud Infrastructure customers can now buy access to OpenAI’s models, including Codex and ChatGPT Work, directly through the Oracle Marketplace.
The expansion was announced on June 10, 2026, with an update following on August 7, 2026. Oracle was already building servers for OpenAI. Now it sells OpenAI’s products too.
What the expanded deal actually does
The core change is procurement. Enterprises on Oracle’s cloud can pay for OpenAI models using their existing Oracle Universal Credits.
Universal Credits are Oracle’s prepaid consumption system. Companies commit spending up front, then draw it down across Oracle services as needed.
Adding OpenAI to that menu means a company does not have to sign a separate contract with OpenAI. Nor does it need to put a new vendor through months of security reviews and legal paperwork.
The research describes the setup as offering AI capabilities within a trusted governance framework. Oracle is positioned as an early mover among cloud providers in packaging OpenAI models this way.
Two products are named in the arrangement. Codex is OpenAI’s coding-focused model, and ChatGPT Work is aimed at business users.
The $300 billion foundation underneath
This marketplace move sits on top of a much bigger commitment. In July 2025, OpenAI agreed to a contract with Oracle worth roughly $300 billion over five years.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
That deal centers on expanding data center capacity in the US. It works out to approximately $30 billion annually, with payments starting in 2027.
Under the agreement, Oracle is set to supply an additional 4.5 gigawatts of computing capacity. That is equivalent to the output of more than two Hoover Dams.
The capacity forms part of the Stargate project. Stargate targets total investment of up to $500 billion to build 10 GW of US AI data center capacity.
The buildout has not been perfectly smooth. By early 2026, some site-specific plans had faced delays, including scrapped expansions in Abilene tied to shifts in financing.
Even so, the overall financial commitment is described as remaining solid. Oracle is expected to start seeing significant revenue from the arrangement around 2027.
Why Oracle and OpenAI keep getting closer
Senior executives, including co-CEO Clay Magouyrk, have expressed optimism that Oracle can meet OpenAI’s growing appetite for infrastructure. Scaling Oracle Cloud Infrastructure to handle heavy AI workloads is central to that ambition.
For OpenAI, the relationship reflects a broader shift. The company has been widening its computing partnerships beyond Microsoft, its longtime primary backer.
What this means for enterprises, Oracle and the cloud race
The marketplace integration changes Oracle’s role in a meaningful way. It moves Oracle from being only a supplier of raw computing power to also being a distribution channel for OpenAI’s software.
A data center contract is lumpy and long-dated, with payments that only begin in 2027. Marketplace sales tied to Universal Credits, by contrast, could keep enterprise customers spending inside Oracle’s ecosystem.
The biggest financial payoff depends on a five-year contract whose revenue has not yet started flowing. The Abilene setback shows that building gigawatt-scale data centers involves financing and execution challenges.
For investors, the key milestones are fairly clear. The first is whether OpenAI models gain real traction among Oracle’s enterprise customers through the marketplace. The second is whether the Stargate buildout stays on track after the early 2026 adjustments. The third is whether revenue from the $300 billion agreement begins arriving around 2027 as expected.