The headline number

The four largest cloud providers now report nearly $2.4 trillion of remaining performance obligations (RPO): revenue under signed contracts that has not yet been recognized. The totals are easy to add up. They are much harder to compare, because each company describes the timing of that backlog in a different way.

In this post we translate each disclosure into a common unit: the average quarterly revenue implied by the near-term backlog. That number says more about the pipelines than the headline totals do. Meta, the fifth company in our hyperscaler peer group, does not report a comparable backlog.

Four companies, three timing methods

ASC 606 requires companies to disclose when they expect to recognize RPO, but it lets them choose how. The four hyperscalers use three different approaches, and the windows don't match.

The definitions differ too. Amazon counts only contracts with original terms over one year. Microsoft splits out a commercial figure. Alphabet splits out Google Cloud. Oracle reports one company-wide number that includes its support contracts.

Translating to a common unit

With some simple algebra, each disclosure can be turned into implied average quarterly revenue from backlog. For the three companies that give a percentage p over a window of X months:

For Amazon, which gives only a weighted-average remaining life L in years, we take the reciprocal of the life as an implied annual recognition rate:

Applied to the June 2026 quarter:

• Microsoft: $684B × 30% ÷ 4 ≈ $51B
• Alphabet: $519.5B × ~50% ÷ 8 ≈ $32B
• Amazon: $496B ÷ (4 × 6.4) ≈ $19B
• Oracle: about $19B, using the RPO and 12-month percentage in its fiscal 2026 10-K (quarter ended May 31, 2026)

What the chart shows

Implied quarterly revenue from backlog has grown at all four companies, but the ranking by growth is very different from the ranking by headline RPO.

Alphabet grew fastest. Its implied figure rose roughly fourfold in a year, from about $7B in Q2 2025 to about $32B in Q2 2026. Most of the jump came after Q3 2025. Google Cloud's actual revenue grew 82% over the same period, from $13.6B to $24.8B, so contracted revenue is growing far faster than revenue itself. Part of Alphabet's level reflects its 24-month window, discussed below.

Microsoft is still the largest. At about $51B a quarter, it is well ahead of the others, up roughly 40% from a year earlier. Its RPO includes a large base of Microsoft 365 and other commercial subscriptions, not only cloud infrastructure, so it starts from a higher floor.

Amazon's backlog is getting longer, not just bigger. Amazon's implied figure rose about 60% in a year, far less than its RPO. The reason is the weighted-average remaining life, which stretched from about four years to 6.4 years. The filing points to two large, long contracts: an expansion with OpenAI of $100 billion over 8 years in Q1 2026, and an expansion with Anthropic of more than $100 billion over 10 years in Q2 2026. More backlog is landing further out.

Oracle is accelerating. Its implied figure rose from about $11B to about $19B over the year. Oracle's August 2026 10-Q, filed after the period in the chart, pushes it higher: $664B × 13% ÷ 4 ≈ $21.6B. Against actual revenue of $19.3B in that quarter, coverage rises to about 112%.

How much of a quarter is already contracted

Comparing implied backlog revenue with actual revenue in the same quarter shows how much each business already has under contract. The spread is wide: from under half at AWS to more than 100% at Google Cloud.

Google Cloud's coverage above 100% is the clearest sign that these are averages, not forecasts of the next quarter. Its 24-month window reaches deep into a ramp as new capacity comes online. Oracle's contracted book now matches its entire current revenue run rate. AWS sits at the other end: much of its revenue comes from usage outside long-term commitments, and Amazon's RPO counts only contracts with original terms over one year.

Caveats

The implied figures are approximations. Each rests on an assumption that the filings do not let us test directly.

• Straight-line within the window. Dividing by X /3 spreads revenue evenly across the window. These backlogs are likely back-loaded, so early quarters would be smaller than the average. The longer the window, the bigger the effect, which inflates Alphabet relative to the 12-month companies.
• Amazon's reciprocal. RPO ÷ weighted-average life assumes contracts recognize straight-line and have similar lengths. When lengths vary, short contracts contribute more to near-term revenue than the average suggests. For example, $100 on a 1-year contract and $100 on a 9-year contract have a 5-year average life, implying $40 in year one; actual year-one revenue is $111. Back-loading pulls the other way, so the net bias can't be signed.
• Rounding. Percentages are disclosed as “approximately” or “just over.” On Oracle's $664B, one percentage point is worth about $1.7B a quarter. Amazon's life is given to one decimal; 0.1 years moves its implied quarter by about $0.5B.
• What RPO excludes. RPO leaves out cancellable contracts and pay-as-you-go usage. Oracle also elects the practical expedient to omit certain variable consideration. RPO is a contracted floor, not a revenue forecast.
• Microsoft's duration is a different measure. The reciprocal of Microsoft's 2.3-year duration is about 43%, well above the roughly 30% it says will be recognized in 12 months. Its duration appears to measure the average timing of recognition rather than remaining contract life, so it can't be used to check the Amazon method.

Getting the data from Calcbench

Every number in this post comes from XBRL-tagged filing data that Calcbench stores. RPO is tagged as RevenueRemainingPerformanceObligation . Where companies tag the timing, the percentages and periods are captured under RevenueRemainingPerformanceObligationPercentage and RevenueRemainingPerformanceObligationExpectedTimingOfSatisfactionPeriod1 , broken out by expected start date.

That means the analysis here isn't limited to five hyperscalers. The same RPO and timing data is available for every filer that reports it, across quarters, in the Calcbench platform and through the API. Narrative disclosures that aren't fully tagged, such as contract descriptions, are searchable in Calcbench's footnote and disclosure text.