OpenAI has partnered with Oracle Corp. to significantly expand its artificial intelligence infrastructure, announcing a deal to develop an additional 4.5 gigawatts of data center capacity in the U.S.
This collaboration is a key component of “Stargate,” a broader initiative aimed at constructing a network of AI data centers for OpenAI. The Stargate project, envisioned as a colossal undertaking, initially aimed for a total investment of up to $100 billion to build 10 gigawatts of AI infrastructure. OpenAI stated in a blog post that it anticipates exceeding its initial commitment due to “strong momentum with partners.”
However, reports suggest a scaling back of its near-term plans for the project’s scope. The Wall Street Journal reported that the initial phase of Stargate has been “sharply” reduced. OpenAI reportedly aims to complete a smaller data center by the end of the year, likely in Ohio. Additionally, the company is developing two other data center campuses in Texas.
OpenAI’s ambitious construction efforts are reportedly complicated by ongoing disagreements with SoftBank Group Corp., which emerged as OpenAI’s largest backer this year, committing to lead a funding round potentially worth up to $40 billion. A significant point of contention is SoftBank’s ownership of the “Stargate” trademark. Key terms of the partnership, including the specific role of SB Energy, a SoftBank subsidiary, in the construction initiative, have not yet been finalized. OpenAI stated in its blog post that “with SoftBank, we’re moving quickly on site assessments and reimagining how data centers are designed to power advanced AI.”
Regarding the SoftBank-led funding round, OpenAI has reportedly raised a quarter of the potential $40 billion. The company is expected to launch an effort to secure the remaining $30 billion on July 28, with SoftBank poised to contribute up to three-quarters of that sum.
Oracle stands to gain substantially from its AI infrastructure deal with OpenAI. The tech giant anticipates generating over $30 billion annually from this partnership within three years. Beyond the recently announced 4.5 gigawatts of capacity, Oracle is also responsible for constructing one of OpenAI’s Texas data center campuses. This particular facility is designed to house a massive array of up to 400,000 of Nvidia Corp.’s GB200 Grace Blackwell accelerators. Each of these powerful chips integrates two Blackwell B200 graphics cards and a 72-core Grace central processing unit.
OpenAI confirmed that Oracle has already begun the process of moving GB200-equipped server racks into this Texas data center. Furthermore, OpenAI’s researchers are actively utilizing this advanced hardware to conduct AI training and inference workloads. The Texas data center is projected to achieve full operational status by mid-2026.
