The $500 billion “Stargate” AI data center initiative—unveiled by President Donald Trump on January 21, 2025—is reportedly running into major turbulence behind the scenes.
More than a year later, The Information reports the original OpenAI–Oracle–SoftBank joint venture hasn’t staffed up and isn’t actively developing data centers, with sources describing the effort as effectively shelved amid disputes over control and structure.
The partnership that stalled
According to The Information, the three partners began clashing over responsibilities and governance well before the project was publicly positioned as a marquee AI infrastructure bet. Those disagreements reportedly dragged on long enough that little—if any—concrete development work took off under the original three-way structure.
The report says OpenAI explored building data centers on its own, but struggled to secure financing for capital-heavy projects. Lenders were reportedly wary of underwriting billion-dollar builds for a company still burning significant cash, pushing OpenAI back toward partner-led infrastructure.
Oracle and OpenAI move forward—without SoftBank at the center
Rather than reviving the original consortium, OpenAI and Oracle appear to have shifted to a more direct arrangement. On July 22, 2025, the companies announced an expansion covering 4.5 gigawatts across multiple U.S. sites—positioning Oracle as a major infrastructure backbone for OpenAI’s compute needs.
The Information also claims the two companies are sharing part of the economic exposure, with both sides taking on some risk tied to delays and overruns while also benefiting if costs come in under expectations. Those specific terms haven’t been publicly detailed, but they underline how expensive—and fragile—these mega-project timelines can be.
OpenAI leans on partners to stay “light”
For now, OpenAI is increasingly relying on cloud and infrastructure partners rather than trying to own the full stack. OpenAI CFO Sarah Friar has framed that approach as a deliberate decision to keep the company’s balance sheet “light,” prioritizing flexibility while demand for compute keeps rising.
At the same time, The Information reports that OpenAI’s previously discussed 10-gigawatt Nvidia deal (announced in September 2025) won’t proceed as originally communicated—another sign that capacity plans across the AI industry are being reshaped by financing, supply constraints, and project execution realities.
Delays raise costs—and push OpenAI to alternative suppliers
The Stargate slowdown has had downstream effects. The Information reports OpenAI missed an internal target to lock in commitments for 10GW of capacity over the next three years through SoftBank and Oracle by the end of 2025—a claim that stands in tension with some earlier public positioning that Stargate was progressing toward that goal.
To keep up with compute demand, OpenAI has been expanding its supplier mix. Over the past year, it has inked and/or expanded relationships involving Amazon Web Services, Google Cloud, AMD, and Cerebras—a pragmatic “get capacity wherever it exists” approach as the AI arms race intensifies.
Texas is still central to the plan
Despite the early friction, parts of the Stargate concept are still materializing. OpenAI has publicly highlighted Milam County, Texas as a major Stargate site, tied to SB Energy (linked to SoftBank) and described as a 1.2GW project in OpenAI’s own updates. The exact on-the-ground construction milestones vary by report, but the direction is clear: OpenAI needs enormous new power and compute footprints, and Texas remains a key piece of the map.
If this reporting holds, Stargate is less a single clean mega-project than a shifting set of deals—restructured as OpenAI and its partners negotiate who pays, who builds, who controls the design, and who absorbs the risk when timelines slip.