OpenAI isn’t having the best time of late.
After Grok4 climbed to the top of the model charts, and Kimi K2‘s release allegedly caused OpenAI to delay the release of its its open-source model, and its International Mathematics Gold over overshadowed by Google which got its results ratified by the organizers, OpenAI’s ambitious compute plans also seem to have hit a roadblock. WSJ reports that OpenAI’s $500 billion Stargate project appears to have scaled back its plans, and will only build a small data center by the end of the year.
WSJ reported that SoftBank and OpenAI, which together lead the joint venture, have been at odds over crucial terms of the partnership, including where to build the sites. In a joint statement, the two companies now told Reuters they were moving “with urgency on site assessments” and were also advancing projects in multiple states.
In January this year, OpenAI had announced plans for a $500 billion investment in building Stargate over four years. Softbank, along with Oracle CEO Larry Ellison and other investors were backing the project, which had been unveiled in the presence of US President Donald Trump. At that time, Larry Ellison had said that the first of the project’s data centers was already under construction in Texas.
When the $500 billion project had been announced, it had caused some eyebrows to be raised in the AI space. Elon Musk, however, had been characteristically blunt. “They don’t actually have the money,” he’d said on X, implying that the investors wouldn’t be able to cough up $500 billion as planned. “Softbank has well under $10 billion secured. I have that on good authority,” he’d added. Sam Altman had then said that Musk was wrong, and had invited him to visit the site of the first data center that was already underway.
It now appears that the Stargate project has indeed appears to have been delayed. These datacenters are critical for top AI labs to train their models — larger datacenters, in general, can produce more powerful AI models. Companies have been scrambling to set up these datacenters as fast as possible — xAI set up a datacenter in record 19 days, and Meta has said that it’s now building datacenters the size of Manhattan and will have the most compute per researcher in its lab. Amid all these competitive pressures, OpenAI will need to make sure that it can also similarly build out enough compute resources if it wants to maintain its lead long-term in the AI model race.