Stargate: OpenAI’s Ambitious Project Fizzles Out

This was the project that was supposed to reshape the face of global AI. In January 2025, Donald Trump summoned the cameras to unveil Stargate: a $500 billion joint venture uniting OpenAI, Oracle, the UAE sovereign fund MGX, and Japan’s SoftBank, aimed at financing and building data centers dedicated to OpenAI.

Fifteen months later, according to a Financial Times investigation, the project looks more like a millefeuille of bilateral deals than a structured alliance.

“I no longer know what ‘Stargate’ means today,” a person involved from the outset in deploying the infrastructure told the FT. “It’s a concept that’s completely outdated.”

Abandoned Projects, Frayed Partners

Warnings have mounted in recent weeks. OpenAI suspended a data center project in northeast England and reconfigured another in Narvik, Norway; both were built with the neo-cloud Nscale. The inventor of ChatGPT also declined to expand its flagship site in Abilene, Texas, refusing to exercise a lease option with developer Crusoe.

Read also: OpenAI Trial: Elon Musk and Sam Altman clash over the origins of a betrayal

In both cases, Microsoft stepped in to take over the sidelined capacity. A Crusoe source cited by the FT described Redmond as “a preferable tenant,” implying greater solvency. A barely veiled dig at OpenAI, whose deficits remain astronomical.

The rupture with Nscale drew particular ire from the British government. The start-up had invested heavily in anticipation of promised contracts. UK AI Minister Kanishka Narayanan publicly contradicted OpenAI’s justification (regulatory constraints and excessively high energy costs) by saying that “the only thing that has changed since those commitments is OpenAI’s funding environment.”

From Joint Venture to Bilateral Deals

Behind the scenes, the transformation is radical. According to FT sources close to the matter, OpenAI has effectively abandoned the joint-venture structure in favor of a string of major bilateral agreements.

Oracle is playing the lead role with its pledge to supply 4.5 gigawatts of compute capacity over five years for $300 billion. AMD, Broadcom, Nvidia, CoreWeave, and Cerebras have also signed partnerships, with the total value at one point surpassing $1 trillion before being partially pared back.

This tactical flexibility mirrors what those close to Sam Altman describe as his “venture mindset”: place many bets, keep only what works. “We’ve created enough demand in the market, and other players have arrived. So we pulled back our own data centers,” one Stargate insider told the FT.

$600 Billion in Sight

OpenAI insists the overall strategy remains intact, claiming it has already secured more than 8 gigawatts of capacity (toward an initial target of 10 gigawatts by 2030) and is now aiming for more than $600 billion in total spending, up from the $500 billion initially announced.

These ambitions place it in the same league as Amazon, Google, Microsoft, or Meta—groups that post tens of billions in annual profits. OpenAI, by contrast, continues to lose money, a fact that has unsettled markets: in early April, reports of missed revenue and growth targets triggered stock declines for SoftBank, Oracle, and CoreWeave.

The company dismissed these anxieties: “Our business is firing on all cylinders.”

Anthropic Caught by Reality

Meanwhile, its main rival in large language models, Anthropic, had previously shown more restraint. CEO Dario Amodei had criticized the ambitious plans of its competitors. But energy-supply constraints beginning to bite into its ability to meet rising demand led him to approve, this month, commitments worth hundreds of billions of dollars over the long term, according to the FT.

Read also: Microsoft relinquishes exclusivity on OpenAI models

As for Stargate, its definition remains deliberately fuzzy. “Stargate hasn’t disappeared, but it has evolved, and everyone can now give it their own definition,” a SoftBank-associated source told the FT. In OpenAI’s own characterization, the term has become a convenient catch-all. As bluntly summarized by its head of revenue, Denise Dresser, in an internal memo: “We saw the exponential curve of compute earlier, we acted faster, and today we have a real structural advantage.”

The question now is whether the coffers will follow.

Dawn Liphardt

Dawn Liphardt

I'm Dawn Liphardt, the founder and lead writer of this publication. With a background in philosophy and a deep interest in the social impact of technology, I started this platform to explore how innovation shapes — and sometimes disrupts — the world we live in. My work focuses on critical, human-centered storytelling at the frontier of artificial intelligence and emerging tech.