When investors want to know whether an AI or cloud company's growth is durable, one of the cleaner forward indicators in the filings is remaining performance obligations, usually abbreviated RPO. The figure answers a specific question: how much revenue has the company already contracted for that it has not yet been able to record? It is, in effect, a disclosed backlog — committed, signed business that will convert to recognized revenue as the company delivers on its obligations.

RPO is not a marketing metric a company invents; it is mandated by the revenue-recognition standard, ASC 606, which governs how and when companies record revenue. Under that standard, a 'performance obligation' is a promise in a contract to transfer a distinct good or service to a customer. Revenue is recognized as those obligations are satisfied. The remaining performance obligations are the ones still owed — the contracted value the company is obligated to deliver but has not yet recognized as revenue. ASC 606 requires companies to disclose the aggregate amount of the transaction price allocated to remaining performance obligations, and typically an indication of when they expect to recognize it. That disclosure lives in the revenue note within the audited financial statements.

As of January 26, 2025, revenue related to remaining performance obligations from contracts greater than one year in length was $1.7 billion, which includes $1.6 billion from deferred revenue and $151 million which has not yet been billed nor recognized as revenue.— NVIDIA Corporation, Form 10-K (fiscal year ended January 26, 2025), source

That disclosure shows the mechanics. NVIDIA's RPO of $1.7 billion for contracts longer than a year is split between $1.6 billion already collected and sitting in deferred revenue (cash received, revenue not yet earned) and $151 million that has not yet even been billed. The same note adds that approximately 39 percent of revenue from these longer-than-one-year contracts will be recognized over the next twelve months — the timing element ASC 606 calls for, which tells a reader how quickly the backlog is expected to convert.

RPO, deferred revenue, and backlog are not the same thing

The three terms are related but distinct, and precision matters. Deferred revenue is cash the company has already received for goods or services not yet delivered — it sits on the balance sheet as a liability. RPO is broader: it includes deferred revenue plus contracted amounts that have not yet been billed or collected, as NVIDIA's split between $1.6 billion deferred and $151 million unbilled illustrates. 'Backlog' is an informal term companies sometimes use in commentary; RPO is its disciplined, standard-defined cousin, with a fixed definition and a required disclosure location. When a company cites a 'backlog' figure that is not the ASC 606 RPO number, the two should not be conflated.

For AI and cloud businesses specifically, RPO carries extra weight because so much of the business is sold on multi-year contracts. A large and growing RPO suggests customers are committing to future capacity and services; the proportion expected to convert in the next twelve months indicates how front-loaded that commitment is. Because the figure is contract-based, it can be a steadier signal than any single quarter's recognized revenue, which is why it features in many cloud and AI-infrastructure earnings discussions.

How to read the RPO disclosure

Three points keep the interpretation grounded. First, RPO reflects contracted obligations, not guaranteed cash — contracts can have cancellation or modification terms, and companies sometimes disclose RPO only for contracts above a certain duration (NVIDIA's disclosure is for contracts 'greater than one year in length'), so the figure may exclude shorter contracts. Second, the timing disclosure — what share converts within twelve months — is as informative as the headline number, because it distinguishes near-term backlog from long-dated commitments. Third, RPO is a revenue-side metric and says nothing about profitability; it is the contracted top-line, not the margin on it.

It helps to see how RPO sits within the wider set of revenue-recognition disclosures the standard requires. ASC 606 governs the entire lifecycle of a contract: identifying the contract, identifying the distinct performance obligations within it, determining the transaction price, allocating that price across the obligations, and recognizing revenue as each obligation is satisfied. RPO is the tail end of that machinery — the portion of the allocated transaction price tied to obligations not yet satisfied. That lineage is why the figure is comparable, in concept, across companies that apply the standard: it is not a bespoke metric each firm defines for itself but a disclosure anchored to a common accounting framework. For AI infrastructure providers and cloud platforms, where revenue increasingly arrives through committed multi-year capacity deals, that common anchor is what lets an analyst read one company's contracted backlog against another's without first untangling two different private definitions.

The disciplined method is to read the RPO figure straight from the revenue note in the financial statements, note the split between deferred and unbilled amounts, and read the recognition-timing percentage alongside it. Each of those is an ASC 606-mandated disclosure with a fixed home in the filing — which makes RPO one of the more rigorous backlog signals available, grounded in the revenue-recognition standard rather than in management's choice of words on a call. For an AI or cloud business whose value rests on durable, contracted demand, that disclosed backlog is one of the clearest forward signals the filings provide — provided it is read as what it is: contracted revenue not yet earned, not cash in hand.