Guide · Markets
Remaining Performance Obligations: Reading a Cloud Backlog
Remaining performance obligations (RPO) are revenue a company has under contract but has not yet recognised, the closest thing to an order book in software and cloud. Here is how RPO works, why Oracle's reached $664 billion in 2026, and what it does not tell you.
When a manufacturer such as Boeing reports a backlog, it means aircraft ordered but not yet delivered. Software and cloud companies have an equivalent, with a less memorable name: remaining performance obligations, or RPO. It is the total value of signed contracts for which the company has not yet delivered the service and therefore has not yet recognised the revenue. Under the US accounting standard for revenue, ASC 606, companies disclose it in their quarterly filings, usually with the share expected to turn into revenue over the next twelve months.
RPO includes two kinds of money. Some has already been billed or collected but not yet earned, which sits on the balance sheet as deferred revenue; most, in long cloud contracts, has not even been invoiced yet. A five-year agreement to buy $10 billion of computing capacity adds $10 billion to RPO on the day it is signed, and the amount runs down as the capacity is delivered and recognised as revenue. That is why RPO can jump by tens of billions in a single quarter when one large contract is signed, while revenue moves only gradually.
2026 has turned RPO into one of the most watched numbers in the market. Oracle reported RPO of $664 billion at the end of its fiscal first quarter on August 31, up $209 billion from a year earlier, after booking more than $30 billion of additional AI cloud contracts, and management said about half should turn into revenue over the following 36 months. Microsoft's commercial RPO reached $678 billion at the end of June, up 84%; the company had said earlier in the year that OpenAI accounted for 45% of its backlog, and that commercial RPO excluding OpenAI grew 25%. Alphabet's cloud backlog was $514 billion at the end of June.
Those figures explain why investors pay attention, and also what they leave out. RPO is a promise of revenue, not of profit or cash. To deliver it, the provider has to build the data centres first: Oracle's free cash flow was about minus $5 billion in that same quarter on roughly $28.5 billion of capital spending, and on October 28, 2026, at its investor day, the company capped new outside funding at $20 billion through fiscal 2030 while keeping a target of at least $225 billion of revenue in that year. The shares rose about 11%. On the same evening Alphabet, whose cloud revenue had more than doubled, fell about 9% after hours because its spending left free cash flow at minus $9 billion.
Four questions help read any RPO figure. How concentrated is it: a backlog where one customer accounts for a third or more depends on that customer's own ability to pay. How fast does it convert: compare the twelve-month portion with the total, and watch whether it rises. What can be cancelled: some contracts allow customers to reduce commitments, and filings describe the terms. And what does it cost to deliver: if capacity has to be financed with debt or new shares, the value to shareholders of each dollar of backlog is lower than the headline suggests.
In practice, read RPO next to three other lines in the same report: revenue growth for the segment it relates to, capital spending, and free cash flow. A rising RPO with steady free cash flow suggests a business converting demand into cash; a rising RPO with falling free cash flow means the company is financing its customers' future demand today. Both can end well, but the second carries more risk if the customers' plans, or the price of the chips they are renting, change before the contracts run out.