OpenAI, Oracle, and SoftBank expand Stargate with five new AI datacenter sites
OpenAI ● Covered by 2 sources
OpenAI, Oracle, and SoftBank are adding five more Stargate datacenter sites to their US buildout. The project is now racing toward $500B and 10 gigawatts of AI computing power.
Based on reporting by OpenAI — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Stargate just got bigger, again. OpenAI, Oracle, and SoftBank announced five additional datacenter sites this week, pushing their joint infrastructure venture closer to the eye-watering $500 billion figure they floated when the project launched back in January. The stated target: 10 gigawatts of AI computing capacity built on American soil, which is roughly enough electricity to power several million homes, except all of it is going toward training and running large language models instead.
The pitch from all three companies is consistent and, frankly, well-rehearsed by now. Tens of thousands of construction and operations jobs. Billions in local investment. A patriotic framing about keeping frontier AI infrastructure inside the United States rather than ceding it to state-subsidized buildouts elsewhere. It's the same script Stargate has used since its debut, just with more zeroes and more zip codes attached each time.
What's notable here isn't the announcement itself but the pace. Five new sites in one swoop suggests OpenAI and its partners aren't waiting around to see if demand for compute levels off. They're betting it won't, and they're betting big enough that walking it back later would be its own kind of news. Oracle brings the cloud and enterprise relationships, SoftBank brings capital and its own AI ambitions through Masayoshi Son's broader portfolio, and OpenAI brings the workloads that supposedly justify all of it.
There's also a quieter subtext: power grids. Ten gigawatts doesn't materialize out of thin air, and every one of these sites will need utility deals, cooling infrastructure, and probably a few tense conversations with local regulators about water usage and electricity pricing. Stargate has been vague on exactly how each site solves that problem, leaning instead on the scale of the commitment as if scale alone answers the engineering questions.
For now, the announcement is mostly a signal to markets and rivals: OpenAI isn't slowing its infrastructure bet just because valuations of AI companies have gotten wobblier in recent months. Whether the compute translates into models good enough to justify the concrete being poured is, as always, the part nobody can announce in a press release.
My take — AI-written commentary, not fact-checked reporting
I'll believe the jobs numbers when I see the payroll, and I'll believe the 10-gigawatt figure when the sites are actually drawing power instead of sitting in a press release. What strikes me most is how this pattern of gigantic, centralized compute bets keeps favoring three or four companies who can write checks this size, which is the opposite direction open-source AI needs the industry to move in. Someone should be asking who gets locked out of the frontier when the entry ticket is half a trillion dollars.
Read more about this at: OpenAI