The Buildout Behind AI Is Shaking Investor Confidence - The Information
The $500 billion Stargate project was supposed to be the most concrete expression of America's AI ambitions.
The $500 billion Stargate project was supposed to be the most concrete expression of America's AI ambitions. What it has become is a case study in how the AI infrastructure buildout is colliding with its own physics — and how the market is starting to price in the gap between announcement and execution.
The Information reported that Oracle's role in the Stargate joint venture may be narrower than investors assumed. The specific ownership structure: SoftBank and OpenAI each committed $19.5 billion for 40 percent ownership stakes, while Oracle and MGX each contributed $7 billion, with the remainder from outside investors. Oracle is not the lead infrastructure provider. It is a minority partner providing cloud services and technology.
Then Bloomberg and Reuters confirmed that Oracle and OpenAI had abandoned plans to expand the flagship Stargate campus in Abilene, Texas — a 600-megawatt addition to the existing 1.2-gigawatt facility. The reasons, reported by The Information, Bloomberg, and Tom's Hardware: financing negotiations collapsed, winter weather disrupted liquid-cooling infrastructure at the site, and — most consequentially — power will not be connected for roughly a year. By that point, Nvidia's next-generation Vera Rubin chips will be available. OpenAI does not want Blackwell-based capacity that will be a generation behind before the buildings are even live.
This is the structural problem nobody is talking about openly: AI infrastructure takes 12 to 24 months to build, but Nvidia is now shipping a new GPU generation every year. Nvidia's own specifications for Vera Rubin, compared to Blackwell: up to 10 times higher inference throughput per watt and one-tenth the cost per token. For a company racing to maintain frontier model leadership, committing to yesterday's hardware is not a viable option.
Oracle disputed the coverage, posting on X that "Crusoe and Oracle are operating in lockstep to deliver one of the world's largest AI Data centers in Abilene at record-breaking pace. Two buildings are completely operational and the rest of the campus is on track." The company also said it has completed leasing for the additional 4.5 gigawatts to deliver on its OpenAI commitments. Oracle shares fell on the initial Bloomberg report, recovering some losses after the denial.
The fallout has created an opening for Meta. According to Reuters, Meta is in talks to lease the Abilene expansion site from developer Crusoe, facilitated by Nvidia. OpenAI, for its part, has appointed new infrastructure leaders following a strategy shift: the company is now renting more AI servers from major cloud providers rather than building its own data centers. Former Intel executive Sachin Katti now oversees OpenAI's Stargate groups.
The market is watching closely. Nasdaq 100 futures have shown heightened sensitivity to capital expenditure reports in the AI sector, and Nvidia's own earnings — while beating expectations — produced a muted stock reaction. Investors are trying to figure out whether the AI infrastructure cycle is durable or overheating.
The deeper question Stargate is forcing: can the economics of AI infrastructure survive the collision between a 24-month build cycle and a 12-month GPU refresh cycle? Every data center that gets announced is a bet that the hardware it will run on is still competitive when the power comes on. The Vera Rubin numbers suggest that bet is getting harder to win.
Sources:
