The squishy number behind the rise and fall of Oracle’s stock

A once-obscure footnote disclosure called “remaining performance obligations," or RPOs, has become one of the most closely watched numbers for investors in AI-themed stocks, especially Oracle. It is also a lot squishier than investors may realize, helping to explain the recent, sharp fall in Oracle’s stock.

RPOs, which have a standard definition across companies, represent contracted sales that have yet to be recognized as revenue. In other words, management believes the sales are probable, not definite.

Because of that, RPOs aren’t on the face of a company’s financial statements. Still, analysts use the number to help shape their future estimates of the company’s business. If they start to doubt the future sales are indeed probable, they are likely to take a dimmer view of a company’s outlook.

Oracle’s stock soared in early September, at one point jumping 36% in a single day, after a bombshell earnings report in which it said its RPOs had more than tripled since the previous quarter to $455 billion. Since then, Oracle’s RPOs have jumped to $523 billion as of Nov. 30, or about nine times Oracle’s revenue for the past four quarters.

Yet the stock has cratered, down 43% from its all-time high on Sept. 10. Investors have grown uneasy about the circular nature of many of the AI sector’s dealings and are looking askance at huge gains in RPOs. They are particularly wary about OpenAI’s ability to meet gargantuan future commitments.

It is widely known that about $300 billion of the increase in RPOs at Oracle was due to a five-year contract with OpenAI to supply computing capacity, although Oracle hasn’t expressly disclosed this. To meet the demand, Oracle is building new, colossal data-center complexes across the country.

The $300 billion portion from OpenAI has drawn skepticism from investors and even some Wall Street analysts. J.P. Morgan credit analyst Erica Spear titled her Dec. 15 note on Oracle: “If You Build It, Will They Pay?"

D.A. Davidson analyst Gil Luria in a Dec. 12 note said: “Since OpenAI is unlikely to deliver on its $300bn commitment, we believe the best course of action for Oracle would be to restructure that contract proactively in order to deploy capital more responsibly instead of pretending it has $523bn of RPO."

OpenAI is closely held and doesn’t disclose its financial reports. It has about $1.4 trillion of commitments over the next eight years, according to a Dec. 5 report by Moody’s. But it has comparatively little revenue, and it is still in the fundraising stage. OpenAI has said it is on pace to top $20 billion of annualized revenue by the end of this year. Equity investors include SoftBank, which has agreed to invest $30 billion.

Uncertainty about Nvidia’s plans with OpenAI, and the possible impact on Oracle, exemplifies the “circularity" that characterizes much of the AI sector. Oracle is a major customer of Nvidia, which someday might be investing in OpenAI, which recently signed a big contract with Oracle.

The pact between OpenAI and Nvidia has taken longer than initially expected. The companies on Sept. 22 announced a letter of intent for a deal in which Nvidia would invest as much as $100 billion in OpenAI over many years, and OpenAI would buy millions of Nvidia’s specialized chips for its AI infrastructure.

At the time, Nvidia and OpenAI said they expected to finalize the details “in the coming weeks." But that was almost three months ago. In its latest quarterly report, Nvidia said “there is no assurance that any investment will be completed on expected terms, if at all."

If Nvidia backs away from OpenAI, maybe another investor would step up to fill the gap, but maybe not. Or if it turns out OpenAI can’t pay Oracle the full $300 billion, maybe Oracle could land another gigantic customer to fill its place. Suffice it to say, if Nvidia’s investment in OpenAI fell through, it likely would be bad news for Oracle’s stock.

It might be far from certain that OpenAI will be able to pay Oracle the $300 billion. But the relevant question is how far. That is the difference between recognizing the contracted amounts as RPOs, or not.

Whether companies are allowed to record RPOs comes down, in part, to a simple question that might not have an easy answer: Is collectibility “probable" under the accounting rules?

This is a subjective determination. It is possible different people looking at the same facts could reasonably reach different conclusions. It is also possible for management’s conclusions to change as facts change. What looks probable today might not look that way a year from now.

The accounting rules don’t define “probable" in percentage terms, but it is a higher bar than “more likely than not." In practice, some accountants peg the threshold at 70% or higher.

The numbers behind the artificial-intelligence boom are mindbogglingly large. Whether they are all real is a judgment call.

Write to Jonathan Weil at jonathan.weil@wsj.com