OpenAI's ambitious $500 billion "Stargate" initiative, designed to secure computing capacity, is undergoing significant adjustments, with some components being abandoned entirely. However, the company's readiness to pursue any necessary deals to acquire computing power has given it a competitive edge in the race to build artificial intelligence infrastructure.
In recent weeks, the group has halted planned data centers in the UK and Norway, abandoned an expansion of its flagship site in Abilene, Texas, and seen several executives associated with the Stargate project depart to join rival firm Meta.
The Stargate plan was initially announced by Donald Trump in early 2025 as a $500 billion joint venture between OpenAI, Oracle, the Abu Dhabi fund MGX, and Japan's SoftBank. Its original purpose was to construct and operate dedicated data centers for OpenAI, pooling resources to tackle the high costs and complex construction of AI infrastructure.
This vision has rapidly given way to a more flexible model: OpenAI is increasingly relying on third-party providers, preferring to lease computing capacity rather than build and own the facilities itself.
A person involved in the initial data center planning stated, "Now I don't even know what 'Stargate' means. I think the term is completely outdated."
This evolving strategy reflects OpenAI's overall operational style. From core technology to partnerships, the company pursues multiple options simultaneously before abandoning all but the optimal choice. An individual close to the company attributed this approach largely to CEO Sam Altman's "venture capital mindset."
According to people familiar with the Stargate changes, OpenAI has effectively abandoned the joint venture model in favor of pursuing a series of large bilateral deals. However, OpenAI executives maintain that the core guiding principle remains "building more computing capacity."
Oracle has played the most central role, agreeing last year to provide OpenAI with 4.5 gigawatts of computing power under a five-year agreement valued at $300 billion.
The company has also secured partnerships with firms including AMD, Broadcom, Nvidia, CoreWeave, and Cerebras, with total commitments once exceeding $1 trillion, though some have since been scaled back.
External doubts persist regarding its ability to shoulder the massive infrastructure costs. A recent report that OpenAI failed to meet internal revenue and user growth targets caused share prices of associated companies like SoftBank, Oracle, and CoreWeave to fall. OpenAI responded that its business is "operating at full capacity."
By shelving or modifying the Stargate plan, OpenAI has reduced costs. However, this pattern of renegotiating or abandoning projects has unsettled partners and raised questions about OpenAI's reliability as a counterparty.
When the plan was launched, OpenAI faced two major challenges: a complex supply chain and lender caution towards an unrated, loss-making startup. The intention was to build data centers under the "Stargate" banner.
The joint venture—pooling funds from blue-chip partners and receiving the US President's endorsement—aimed to solve both problems simultaneously.
Subsequently, a surge in demand for AI tools prompted action from chip, cloud computing, and infrastructure firms, easing investor concerns.
A person involved in the Stargate project said, "We created enough demand in the market, and other companies followed, so we marginalized our own data center construction plans."
OpenAI stated, "'Stargate' is an umbrella term for our computing capacity strategy, and we have fully delivered on our commitment: securing the computing capacity OpenAI needs at an unprecedented scale. We have established a broad partner network covering cloud, chips, and infrastructure. Capacity is coming online, ahead of schedule. Construction at this scale is intended to make powerful AI accessible to more individuals, businesses, and developers."
This month, OpenAI shelved a data center project in northeast England and adjusted another project in Narvik, Norway. Both were previously touted as part of "Stargate" and were being developed in collaboration with AI cloud computing startup Nscale.
OpenAI blamed the halt of the UK project on strict regulation and high energy costs, a move that annoyed the UK government. UK AI Minister Kanishka Narayan stated, "The only thing that has changed since the original commitment is OpenAI's financing environment."
UK-based Nscale had previously invested heavily expecting to lease computing capacity to OpenAI. Microsoft has now taken over Nscale's site in Narvik. OpenAI stated it still plans to access Norwegian computing capacity through its overall partnership with Microsoft, rather than leasing directly from Nscale.
Last month, the AI lab abandoned plans to expand its Abilene, Texas site, opting not to exercise a lease option with developer Crusoe.
A person involved in the Stargate project revealed that OpenAI found cheaper alternatives, including in Michigan, because "there is a finite amount of capital in the world, and regardless of what Sam [Altman] says, the expansion costs were astronomically high."
Microsoft has taken on the additional computing capacity in Abilene. A person close to Crusoe said Microsoft is a more ideal tenant, stating, "The outcome is better, [they have] stronger creditworthiness."
A person familiar with Microsoft's thinking said the decision "helped partners who were disappointed and misled by OpenAI's abandonment of the project." The person added, "Stargate has changed three or four times. I don't know what it is now. I couldn't say. Perhaps it never really existed in the first place."
When the Stargate plan was launched, the company aimed to secure 10 gigawatts of computing capacity by the end of the 2020s—equivalent to about 10 nuclear power plants—at a cost of roughly $500 billion.
The company's targets are now higher. It claims to have secured over 8 gigawatts already and expects total expenditure to exceed $600 billion by the end of 2030.
This means the loss-making company's spending will far outpace rivals like Anthropic and Elon Musk's xAI, approaching the scale of tech giants like Amazon, Google, Microsoft, and Meta, which generate annual profits of tens of billions of dollars.
SoftBank, a key financial backer in the original joint venture plan, is building its own data centers, though they are not yet operational. A knowledgeable source indicated that computing capacity from a project in Ohio might eventually be supplied to OpenAI through a tender process.
Another person familiar with the Japanese investor's perspective said, "Stargate hasn't disappeared; it has evolved. Now, people can largely define what 'Stargate' is for themselves. To some extent, any computing capacity project involving SoftBank or Oracle could be called 'Stargate'." SoftBank declined to comment.
A person involved in the Stargate project said the term lost its meaning when OpenAI's policy and PR teams began using it as a blanket term for infrastructure efforts. They stated that including the tenancy of a European data center under the Stargate umbrella "was always false advertising."
Despite this, OpenAI's aggressive infrastructure investments may still yield an advantage. Anthropic CEO Dario Amodei has criticized competitors' aggressive plans. However, facing power constraints that are beginning to hamper its ability to meet rapidly growing demand, Amodei this month also approved hundreds of billions in long-term computing capacity spending.
If OpenAI can fulfill its massive spending commitments, its moves may prove prescient.
Earlier this month, OpenAI Chief Revenue Officer Dennis Draes wrote in an internal memo to sales staff, "We recognized the exponential growth in computing capacity demand earlier and acted faster, giving us a tangible, structural advantage today."
Comments