Glossary · from SEC filings
What are remaining performance obligations (RPO)?
Remaining performance obligations, or RPO, is the revenue a company has under contract but has not yet earned. It is the closest thing the filings offer to a measure of contracted future demand, and in AI it has become one of the most quoted numbers in the stack.
What RPO means
Under the revenue accounting standard (ASC 606), a company must disclose the total transaction price of its contracts that relates to work it hasn’t yet performed. That is RPO: signed, non-cancellable contracts, valued at what the customer has agreed to pay, minus what has already been recognized as revenue. It includes amounts already billed but not yet earned (deferred revenue) and amounts not yet billed at all.
Companies also disclose roughly when they expect to recognize it, usually the share due within twelve months. Many call that part “current RPO” or cRPO. The split matters as much as the total: a backlog that converts slowly is a promise about the distant future.
RPO across the AI stack
Across the US-listed companies on our map, RPO totals $3.11T. The largest: Microsoft $683B (2.0× revenue); Oracle $664B (9.3× revenue); Alphabet $568B (1.2× revenue); GE Vernova $197B (4.5× revenue); Broadcom $179B (2.0× revenue); Micron $134B (1.0× revenue); Dell $132B (0.9× revenue); CoreWeave $104B (13.7× revenue); Salesforce $66.3B (1.5× revenue); Sandisk $59.8B (3.0× revenue); Cisco $46.7B (0.7× revenue); Quanta Services $41.8B (1.1× revenue).
The ratio to revenue is what stands out. A software company with RPO of one or two times its annual revenue is normal: customers sign multi-year subscriptions. Oracle’s RPO of $664B is 9.3 times its revenue, and CoreWeave’s about 14 times, because both have signed very large, very long contracts to deliver AI computing that hasn’t yet been built.
How quickly it turns into revenue
The timing disclosures show how different the same word can be. Oracle expects to recognize about 13% of its RPO in the next twelve months; Microsoft about 30%; ServiceNow about 46%. Sandisk, whose RPO comes entirely from long-term supply agreements, expects about 19%. GE Vernova’s equipment backlog runs over five years.
A slow-converting backlog isn’t bad in itself; it may simply reflect long contracts. But it is more exposed to change: the further out the revenue, the more depends on the customer still being able and willing to pay, and on the company building the capacity to deliver it.
Oracle: Share of backlog due within a year
“Remaining performance obligations were $664 billion as of August 31, 2026, of which we expect to recognize approximately 13% as revenues over the next twelve months, 37% over the subsequent month 13 to month 36, 34% over the subsequent month 37 to month 60 and the remainder thereafter.”
Oracle 10-Q, filed September 11, 2026
Microsoft: Share of backlog due within a year
“We expect to recognize approximately 30% of both our total company remaining performance obligation revenue and commercial remaining performance obligation revenue over the next 12 months and the remainder thereafter.”
Microsoft 10-Q, filed October 28, 2026
ServiceNow: Share of backlog due within a year
“As of December 31, 2025, the total non-cancellable RPO under our contracts with customers was $28.2 billion, and we expect to recognize revenues on approximately 46% of these RPO over the following 12 months.”
ServiceNow 10-K, filed January 29, 2026
Sandisk: Share of backlog due within a year
“Approximately 19% of the remaining performance obligations from these contracts with customers are expected to be recognized as revenue over the next twelve months.”
Sandisk 10-K, filed August 17, 2026
GE Vernova: Equipment backlog timing
“We expect to recognize revenue as we satisfy our remaining performance obligations as follows: (1) Equipment-related RPO of $64,245 million of which 37%, 69%, and 97% is expected to be recognized within 1, 2, and 5 years, respectively, and the remaining thereafter.”
GE Vernova 10-K, filed January 29, 2026
RPO, backlog and deferred revenue
Backlog is the older, looser term. Industrial companies still report “backlog” or “orders” in their own definitions, which may include contracts that can be cancelled. RPO is the standardized version: non-cancellable, measured the same way by every company that follows US accounting rules. Deferred revenue is narrower still: only what has been billed or paid but not yet earned, which sits on the balance sheet as a liability.
So when a company’s backlog headline is much larger than its RPO, the difference is usually contracts that can still be cancelled or adjusted. When RPO is much larger than deferred revenue, most of the contracted revenue hasn’t been billed yet, as with Sandisk, whose 10-K shows $58.7 billion of its $59.8 billion unbilled.
Sandisk: Long-term supply agreements
“As of July 3, 2026, the transaction price allocated to remaining performance obligations was $59.8 billion, of which $58.7 billion has not yet been billed and $1.1 billion has been recorded as contract liabilities.”
Sandisk 10-K, filed August 17, 2026
A worked example
Suppose a company signs a three-year contract on January 1 to provide cloud capacity for $300 million, billed $100 million at the start of each year. On signing, its RPO rises by $300 million. When it bills the first $100 million, that becomes deferred revenue, still inside RPO. As it delivers the service through the year, it recognizes the $100 million as revenue, and RPO falls to $200 million, of which the next year’s $100 million is current RPO.
The example shows why RPO and revenue move at different speeds. A company that signs many large multi-year contracts can see RPO jump in a single quarter while revenue rises only gradually, which is exactly what happened at Oracle and CoreWeave.
How Oracle’s backlog jumped
Oracle is the clearest example. Its RPO at the end of each of its last 8 quarters: Q4 2024 $97.3B, Q1 2025 $130B, Q2 2025 $138B, Q3 2025 $455B, Q4 2025 $523B, Q1 2026 $553B, Q2 2026 $638B, Q3 2026 $664B. Its quarterly revenue over the same period rose far more slowly, from $14.1B to $19.3B.
A backlog that grows several times over in a year reflects a few very large contracts rather than many small ones, and it will take years to turn into revenue: Oracle expects about 13% within twelve months. Its value to shareholders depends on Oracle building the capacity on time and the customers being able to pay, which is why Oracle’s free cash flow and borrowing matter as much as the backlog itself.
What RPO leaves out
RPO is not a complete measure of future revenue. Companies may leave out contracts of a year or less and usage-based fees, which is how much cloud and software revenue is earned; Sandisk’s 10-K, for example, notes that it excludes arrangements with an original expected duration of one year or less. Contracts that a customer can cancel without substantial penalty are also excluded.
So RPO understates future revenue for businesses that sell month-to-month or by usage, and it is most informative for businesses that sign large, long, fixed contracts, which is exactly the shape of the AI infrastructure deals that have made it famous.
Five questions to ask of any RPO figure
How fast is it growing compared with revenue? RPO growing much faster than revenue means contracts are being signed for capacity or products not yet delivered. How much is due within twelve months? The lower the share, the more depends on the distant future. Who are the customers? A backlog concentrated in a few customers is only as strong as they are.
Can the company deliver? A backlog of capacity that still has to be built depends on construction, power and financing. And what does it exclude? If the business earns much of its revenue from usage or short contracts, RPO will understate it, and comparisons with companies that sign long fixed contracts will mislead.
How to use RPO as an investor
Read three things together: the total, its growth over the year, and the share due within twelve months. Growth in RPO usually leads growth in revenue, which is why it gets so much attention. But the AI build-out adds a fourth question: who is the customer? When a large share of a company’s RPO comes from a few customers that are themselves funded by investors, the number’s quality depends on their funding as much as on demand.
Each company page shows the company’s RPO and, where it is disclosed, the timing. The is AI a bubble? page tracks backlog far ahead of revenue as one of its eight indicators.
Related terms
More from the glossary: Hyperscaler, Neocloud, High-bandwidth memory (HBM), AI data center, TPU vs GPU, CoWoS (advanced packaging), Vendor financing, and every other term you need to read AI companies’ filings.
Free · quarterly · from SEC filings
Get the AI Stack Report every quarter
After each Big Tech earnings season we update the map from the new filings and publish what changed: who spent, who got paid, who kept it and what the filings reveal about the loop. Free, in Rob May’s Investing in AI newsletter.
Form not showing? Subscribe on Substack. No spam; unsubscribe any time.
Questions about RPO
What is RPO?
Remaining performance obligations: the revenue a company has under signed, non-cancellable contracts but has not yet recognized. It includes both deferred revenue and amounts not yet billed.
What is the difference between RPO and backlog?
RPO is defined by the accounting standard and counts only non-cancellable contracts. “Backlog” is each company’s own measure and may include orders that can be cancelled, so it is often larger.
What is cRPO?
Current RPO: the part of remaining performance obligations a company expects to recognize as revenue within the next twelve months.
Why is Oracle’s RPO so large?
Because it has signed very large, long-term contracts to provide AI computing capacity. Its RPO is many times its annual revenue, and it expects to recognize only about 13% of it in the next twelve months.
Is a high RPO good?
It signals contracted demand, but its value depends on how quickly it converts to revenue, how concentrated it is among customers, and whether those customers can pay.
Does RPO include usage-based revenue?
Usually not. Companies may exclude variable, usage-based fees and contracts of a year or less, so RPO understates future revenue for businesses that bill by usage.
Why is CoreWeave’s RPO so high relative to revenue?
Because it has signed multi-year contracts with a few large customers, including Microsoft, Meta and OpenAI, for capacity it is still building. Its RPO is many times its annual revenue.
Is RPO the same as deferred revenue?
No. Deferred revenue is only what has been billed or paid but not yet earned. RPO adds contracted amounts not yet billed, so it is usually much larger.
Does every company report RPO?
Every company following US accounting rules must disclose it if it has long-term contracts, but the amounts are only meaningful where contracts are long and fixed. Hardware makers that sell order by order often report little or none.
Where do I find a company’s RPO?
In the revenue note to the financial statements in its 10-K and 10-Q, usually under “remaining performance obligations”, with a sentence on how much is expected within twelve months.
Research and education, not investment advice: nothing here is a recommendation to buy or sell any security. Figures come from companies’ SEC filings as reported; companies named were chosen because they show how the stack works, not as a view on their shares.