The AI boom has produced plenty of big numbers and bold headlines—but few rival “Stargate,” the massive data center project backed by Oracle and OpenAI.
Billed as a $500 billion bet on the future of AI computing, the plan envisioned building enormous facilities packed with NVIDIA chips to power the next wave of advanced AI models.
Last week, however, the first major setback emerged.
Plans to expand the flagship Stargate site in Texas were put on hold after negotiations between Oracle and OpenAI broke down.
Though the project itself continues and construction hasn’t stopped, the pause highlighted a rare moment of uncertainty in an industry otherwise defined by relentless optimism.
Even the companies building the future, it seems, are still figuring out how big that future should be.
Why the Texas expansion stalled
The story centers on Stargate’s 1,000-acre campus in Abilene, Texas—unveiled in 2025 as one of the most high-profile examples of the global AI infrastructure boom.
Oracle and OpenAI had discussed expanding the facility’s capacity from roughly 1.2 gigawatts to about 2 gigawatts, but those talks ultimately collapsed.
Financing terms proved difficult, and OpenAI’s projections for future compute demand kept shifting—a challenging combination for a project that demands decade-long commitments in a rapidly evolving industry.
The original agreement between the two firms remains intact.
Oracle still plans to deliver around 4.5 gigawatts of capacity for OpenAI across multiple sites, and parts of the Texas campus are already operational using NVIDIA hardware.
Still, the additional Abilene expansion is off the table—for now—leaving the door open for another company to step into the spotlight.
Nvidia steps in
The most interesting player in this episode may not be Oracle or OpenAI. It may be Nvidia.
The chipmaker reportedly placed a $150 million deposit with the project’s developer, Crusoe Energy Systems, and began helping attract a new tenant for the unused expansion space. One possible candidate is Meta Platforms, which is currently considering leasing the site.
The move says a great deal about Nvidia’s role in the AI economy.
Traditionally, semiconductor companies sold chips and let customers decide how to build infrastructure.
Nvidia now behaves more like an orchestrator of the entire ecosystem.
By helping match large data center projects with potential tenants, the company can influence where AI clusters are built and which hardware ends up inside them.
This shows that Nvidia’s influence now extends well beyond GPU design. It increasingly touches the infrastructure decisions that determine where hundreds of billions in AI spending will flow.
Oracle’s risky bet
For Oracle, Stargate represents the most ambitious strategic move in the company’s modern history. The firm built its reputation on enterprise software and databases. AI infrastructure is a very different business.
Large data centers require heavy capital spending and long construction timelines, and Oracle has taken on large operating lease commitments to support that expansion.
The company is also expected to raise up to $50 billion through debt and equity to finance its AI infrastructure buildout.
Analysts project that the scale of these investments could push Oracle’s free cash flow negative for several years before the spending begins to generate returns later in the decade, a fact that made investors worry about the stock.
Source: Bloomberg
The recent news explains why the market reacted quickly.
When reports emerged that the Texas expansion had been shelved, Oracle’s stock pulled back, and analysts lowered their price targets.
Another concern lies in customer concentration. OpenAI has become one of Oracle’s most important cloud customers.
If OpenAI spreads its infrastructure across multiple providers or adjusts its growth plans, Oracle’s projections for AI-driven revenue growth could change quickly.
Oracle is attempting something unusual for a software company. It is moving into an industry that behaves more like utilities or heavy infrastructure. Margins tend to be lower, and execution mistakes can become expensive.
Oracle’s stock is already down more than 20% year-to-date.
Meta’s appetite for compute
If Meta ultimately takes the additional space in Abilene, the decision would highlight another emerging trend in the AI race.
Some companies are spending far more aggressively than others.
Meta has projected capital expenditures of up to $135 billion in 2026 as it builds new data centers and expands its AI computing capacity.
The company operates massive social platforms and uses AI across advertising, recommendations and content moderation. That gives it several internal reasons to build larger clusters.
Unlike many AI startups, Meta also generates tens of billions in annual cash flow. Those resources allow it to finance projects that smaller companies could never attempt.
The potential move into the Stargate expansion site would therefore fit Meta’s strategy.
Build more computing capacity than rivals and run increasingly large models across its products.
The scale of spending now being discussed would have been difficult to imagine only a few years ago. AI infrastructure budgets increasingly resemble national infrastructure projects rather than traditional technology investments.
The real bottleneck of the AI boom
The Stargate episode reveals something deeper about the evolving AI economy: the key constraint is shifting from algorithms to physical infrastructure.
Training advanced AI models now depends less on code and more on the availability of electricity, cooling systems, and sheer physical space.
A single gigawatt-scale data center consumes roughly as much power as a nuclear reactor—enough to supply about 750,000 homes.
That staggering comparison, noted in reporting on the Abilene site, captures how immense these projects have become.
Land, energy access, and grid connections are emerging as strategic resources.
The companies racing to lead in AI are, in effect, competing for territory—seeking locations where power is abundant, and regulators are willing to approve massive new facilities.
The Texas expansion dispute offers an early glimpse of how this industry may develop. Demand for computing power will keep rising, but not in a steady or predictable line.
Projects will expand, pause, or even change hands as firms continually reassess their infrastructure needs.
Stargate was billed as one of the largest technology infrastructure efforts ever attempted, and that description still fits.
Yet this first slowdown underscores a sobering truth: even half-trillion-dollar ambitions eventually run up against practical limits—of energy, financing, and uncertainty about how much AI the world truly needs.
For investors watching the AI race, those limits may soon matter as much as the models themselves.
The post Who wins as Oracle, OpenAI’s $500B Stargate project stalls? appeared first on Invezz
