OpenAI, SoftBank and Oracle Clash Over Stargate as the Data-Center Megaproject Stalls

The Stargate megaproject, which was supposed to be OpenAI's compute foundation, has stalled before it has even fully launched. On the surface it looks like an ordinary construction delay, but behind it lies a growing conflict of interests between OpenAI, SoftBank and Oracle.

Why Stargate is stalling

Stargate was announced on January 21, 2025 as an AI infrastructure project worth up to $500 billion over four years. The idea was simple: build a network of data centers that would give OpenAI and its partners enough capacity to train the next generations of models. But a year after the splashy launch, it has become clear that a gap has opened between the presentation and the actual execution.

According to The Information, the joint venture never got a fully dedicated team, and the participants argued even over basic questions: how the project should be structured, who is responsible for what, and where exactly to build new sites. It is also telling that the data center in Abilene, which OpenAI calls Stargate I, is not being built by the joint venture itself but by Crusoe for Oracle, and work there began before the official announcement.

• The project lacks the pace promised at the start
• Participants disagree on structure and governance
• The key construction site in Texas sits outside the framework of a classic joint venture
• OpenAI is simultaneously seeking capacity from other providers

As a result, Stargate looks less and less like a single mechanism and more like a sign under which each participant pursues its own agenda. For the market this is a bad signal: when an infrastructure project worth hundreds of billions starts to fragment into separate deals, investors inevitably ask questions about its real status.

Money and control

OpenAI, SoftBank and Oracle had different interests from the outset, and now those differences are too visible to ignore. OpenAI needs compute, but without a gigantic burden on its own balance sheet. Oracle needs cloud growth and big contracts, but debt financing for such a race is getting harder. SoftBank, for its part, sees the project as a chance to gain a foothold in the scarcest layer of the AI market: infrastructure.

"Stargate is not yet formed," is how

Oracle's Safra Catz described the project's status to investors.

This is where the main shift in power shows. If Stargate was once presented as an alliance for the next leap in AI, it now looks more like a venue for bargaining over control of land, energy, contracts and future data centers. In that logic, SoftBank does not necessarily have to believe in the quick arrival of AGI: it is enough for it to understand that rights to electricity, sites and capacity will remain valuable even in a cooler market.

The situation is also nerve-racking for Oracle. The company has already taken on a heavy infrastructure load, and any doubt around Stargate hurts not only the project's reputation but also investors' expectations for the cloud segment. If the pace of construction and contract signing fails to match the promises, pressure on Oracle and its cost of capital will only grow.

What OpenAI is doing

Against this backdrop, OpenAI is hedging more and more through parallel channels. The company has deepened its direct cooperation with Oracle, kept buying capacity from CoreWeave and struck agreements with AWS and Google Cloud. Separately, it is working with Nvidia alternatives, including AMD and Cerebras. In other words, OpenAI is no longer betting on a single megaproject, even if it continues to publicly back the Stargate idea.

The monetization of ChatGPT itself is also a symptom of this pressure. On January 16, 2026, OpenAI announced it would begin testing ads in the free and Go tiers in the US, and on February 9 it launched a pilot. The ads appear separately from the model's answer and do not affect paid subscriptions, but the fact itself matters: the company is looking for new money not only in capital markets but also inside a product with hundreds of millions of users.

This does not mean OpenAI is abandoning its ambitions. If anything, the opposite: the company is trying to preserve access to compute at any cost without tying itself to a single organizational scheme. But that is exactly why Stargate is ceasing to be the center of the whole story. If capacity can be topped up through individual partners, the role of the joint venture inevitably blurs.

What this means

The Stargate saga shows that the AI race has entered a phase of hard economics. Big announcements are no longer enough: the market is now looking at debt load, ownership structure, actual construction and sources of revenue. For OpenAI this means a move from a glossy showcase to a more pragmatic growth model, and for the whole industry, a test of how much of the infrastructure boom is real demand and how much is expectation.