OpenAI admits demand is lower than expected, halts Oracle data center expansion

Oracle (ORCL) and OpenAI abruptly halted a massive AI data center expansion in Texas.

And just like that, billions of dollars in planned AI infrastructure spending suddenly evaporated.

The project was part of the Stargate buildout in Abilene, Texas. The site was supposed to scale from about 1.2 gigawatts to 2.0 gigawatts of computing power.

Instead, the expansion fell apart after financing talks stalled and OpenAI admitted it overestimated demand forecasts.

That comes at a striking moment for the company, especially as OpenAI’s new funding round pushed its valuation near $850 billion. This makes it the second most-valuable private company behind SpaceX.

The market reacted immediately. Oracle, Nvidia (NVDA), and CoreWeave (CRWV) stocks all slipped.

For investors betting on the AI infrastructure boom, the message was hard to ignore. The biggest most ambitious AI date centers depend on fragile economics.

Projects this large usually cannot move forward without long term customers locking in most of the capacity. Banks typically require tenants to commit to roughly 60 to 70 percent of a facility’s power load before they release financing.

And at the multi gigawatt scale, that’s a huge commitment.

A project like this can easily cost $10 billion to $15 billion to build. If OpenAI trimmed its expected demand even slightly, the project may have suddenly fallen below the threshold banks require to fund it. Mind that Nvidia recently stopped funding OpenAI's expansion. These two events could be related.

Point is, that’s how a single forecast revision can freeze billions in infrastructure spending almost overnight.

Meta Platforms is already in talks to take over the capacity that OpenAI backed away from. Nvidia is reportedly helping broker the deal and has placed a $150 million deposit with developer Crusoe Energy Systems to help move negotiations forward. And the broader 4.5 gigawatt agreement between Oracle and OpenAI is still intact.

But the cancellation revealed something the market has not fully priced in yet.

The AI infrastructure boom depends on constant demand growth. If that demand wobbles even slightly, billions of new data center spending can unravel much faster than investors expect.

The disruption behind the news: If AI demand slows, AI stocks valuations must come down too.

A two gigawatt facility isn’t small. That’s enough electricity to power roughly 350,000 homes. Canceling an expansion at that scale suggests the industry might be getting ahead of near term compute demand.

Many investors assumed AI demand would only move higher. OpenAI just showed that growth can flatten.

Building AI infrastructure requires enormous upfront capital. Oracle has already warned its cloud investments could push cash flow negative until 2030. The company is reportedly considering raising up to $50 billion to keep building data centers.

But here’s the challenge. Data centers only work if customers commit to filling them for years.

Training advanced AI models costs billions. Running them at scale requires steady demand. If AI companies hesitate on capacity commitments, the entire infrastructure chain becomes riskier.

That includes power providers, chip companies, and cloud operators.

The ripple effects help explain the stock reaction. Nvidia sells the GPUs. Oracle builds the infrastructure. CoreWeave rents compute capacity. A slowdown anywhere in that chain pressures all of them.

Meta stepping in also shows the next phase of the AI race. Large tech companies are starting to compete directly for the same computers and data centers needed to run AI.

What to watch next

Watch the data center leasing market closely.

Watch whether AI demand forecasts begin shifting across the industry.

Watch which companies can actually fill gigawatts of compute capacity.

Right now, the industry pipeline is massive. Analysts estimate more than 20 gigawatts of AI data center capacity planned across the United States alone. For context, the Abilene expansion would’ve represented about 10% of that.

If more projects stall, investors will start questioning how fast this infrastructure buildout can actually happen.

But if companies like Meta step in and absorb that capacity, something else starts to change. The AI race becomes more concentrated among the biggest tech giants.

That’s because only a handful of companies can consistently use compute at the gigawatt scale. OpenAI. Meta. Google. Microsoft. Maybe Amazon.

Everyone else ends up as a tenant.

Over the next 6 to 24 months, the winners probably won’t be the companies promising the biggest AI breakthroughs. The winners will be the companies that can guarantee demand for massive compute clusters.

And in that kind of fight, the biggest players usually have the advantage.

Oracle (ORCL) has a Disruption Score of 2.

Click here to learn how we calculate the Disruption Score.