Oracle Stock Plummeted by 25% in the First Half of 2026, and This Dire AI Warning Might Be Why

Oracle (NYSE: ORCL) was founded as a software company in 1977, before transitioning first to cloud computing and later to artificial intelligence (AI) infrastructure. It operates some of the fastest and most cost-efficient AI data centers in the world, and its order backlog suggests demand is through the roof.

Nevertheless, Oracle stock plummeted by 25% during the first half of 2026, and it's down by more than 50% from last year's record high. Wall Street is concerned that some of Oracle's largest AI customers won't be able to fulfill their obligations, which would be catastrophic because the company has taken on an enormous amount of debt to build more data centers.

Those concerns reached a crescendo on June 10, when Oracle released its annual report for fiscal 2026 (ended May 31). It contained some unusual and very thorough warnings about the risks facing its AI infrastructure business, and here's why I think they give investors enough reason to avoid its stock.

Image source: The Motley Fool.

Oracle operates some of the best AI data centers

AI training and inference workloads require a lot of computing capacity, which is typically delivered by thousands of specialized data center chips called graphics processing units (GPUs). Oracle buys these chips from some of the world's top suppliers, like Nvidia and Advanced Micro Devices, installs them in its infrastructure, and rents the computing capacity to other businesses.

Oracle has a couple of advantages over other cloud companies. First, it uses a high degree of software-powered automation to run its data centers, which means it can bring new infrastructure online much faster than operators that rely on human-led processes. Second, its random direct memory access (RDMA) networking technology moves data between chips and devices faster than traditional Ethernet networks.

These features result in lower costs and faster processing speeds, the perfect combination for AI developers, who typically pay for computing capacity by the minute. That's why top companies like OpenAI, Elon Musk's xAI, and Meta Platforms are lining up to use Oracle's infrastructure.

In fact, the company ended fiscal 2026 with a staggering $638 billion in remaining performance obligations (RPO) from customers waiting for more data centers to come online, a figure that had soared by 363% year over year. RPO reflects the value of signed contracts for services that haven't been delivered yet, so it's similar to an order backlog. Oracle says the $638 billion figure is a good reflection of future revenue, but its latest annual report is littered with warnings to the contrary.

Almost half of Oracle's RPO is reportedly from one customer

According to a report by The Wall Street Journal from last September, around $300 billion of Oracle's RPO is attributable to OpenAI alone, and that poses a few problems. First, OpenAI has just $25 billion in annualized revenue. Second, it's losing money, so it relies on funding from outside investors to cover its operations -- in fact, it recently raised a whopping $122 billion in fresh capital.

Third, OpenAI has made commitments on a similar scale to other cloud providers like Microsoft, so the start-up is hundreds of billions of dollars short of being able to fulfill all its obligations. That might explain why Oracle provided the following warning in its fiscal 2026 annual report:

Some of our customers may be highly leveraged and subject to their own operating and regulatory risks and, even if our credit review and analysis mechanisms work properly, we may experience risks of non-payment and non-performance in our dealings with such parties. In certain OCI offerings, we are more concentrated among a number of large customers, which could increase these risks.

Oracle also warned that it could be overestimating customer demand and that it might be unable to release its AI data center capacity to other parties or repurpose it for other workloads. In other words, Oracle could wind up sitting on a ton of infrastructure it simply can't use.

Oracle expects to convert only around 12% of its total RPO into revenue over the next 12 months, followed by a further 34% in the 24 months after that. In simpler terms, less than half of its RPO will become revenue over the next three years in the best-case scenario. A lot can change in that time, especially in an industry moving as quickly as AI. There is a real risk that many of Oracle's customers won't need all the computing capacity they have signed up for.

This could have dire consequences for the company, as it's carrying over $122 billion in long-term debt and recently announced plans to raise another $40 billion through a mix of debt and equity to fund more data centers. Therefore, I'm not surprised investors are heading for the exits -- I certainly won't be buying this stock as things currently stand.

Anthony Di Pizio has no position in any of the stocks mentioned. The Motley Fool has positions in and recommends Advanced Micro Devices, Meta Platforms, Microsoft, Nvidia, and Oracle. The Motley Fool has a disclosure policy.