Key takeaways
- For the quarter ended 30 June 2026, AWS reported $16,621 million of operating income on $42,232 million of net sales, a 39.4% operating margin. Google Cloud made 35.6%.
- Microsoft publishes no Azure profit line at all. Its Intelligent Cloud segment earned $56,972 million on $137,791 million of revenue, but that segment also holds SQL Server, Windows Server and GitHub.
- Oracle's cloud and software margin of $34,468 million on $58,530 million reads as 58.9%. A footnote says it excludes research and development, general and administrative costs, and stock compensation.
- Alibaba's Cloud Intelligence Group reported adjusted EBITA of RMB14,265 million on RMB158,132 million of revenue, a 9.0% margin struck before share-based compensation and impairments.
- Amazon alone discloses segment capital spending: AWS added $48,604 million of property and equipment in one quarter, against $16,043 million a year earlier.
Only two of the five biggest cloud businesses publish a margin you can compare
You want to know which cloud business earns the most on what it sells. The filings won't answer that, and not because anything is being hidden. Each provider reports a different number under the same word.
Two of them do publish the same measure. For the three months to 30 June 2026, Amazon reported AWS net sales of $42,232 million and AWS operating income of $16,621 million. That is an AWS operating margin of 39.4%. Alphabet reported Google Cloud revenue of $24,768 million and Google Cloud operating income of $8,814 million, or 35.6%. Same measure, same period, same currency. Those two you can set beside each other.
Both have moved a long way in a year. In the same quarter of 2025, AWS made $10,160 million on $30,873 million, a 32.9% margin. Google Cloud made $2,826 million on $13,624 million, or 20.7%. The direction is the same and the gap has narrowed. That comparison holds only because the two companies happen to define segment profit the same way.
After that, cloud segment reporting stops being comparable. The chart above plots the figure each of five providers reports for its cloud unit. Read as a ranking, it says Oracle runs the most profitable cloud business in the world and Alibaba the least. Read properly, it says five companies use five definitions.
Microsoft publishes two cloud numbers, and neither one is Azure
Microsoft's Form 10-K for the year to 30 June 2026 gives two figures a reader could mistake for the cloud business.
The first is Microsoft Cloud revenue: $214.4 billion. That is not a segment. The filing defines it as revenue "which includes Microsoft 365 Commercial cloud, Azure and other cloud services, the commercial portion of LinkedIn, and Dynamics 365", and the filing places those amounts in two different reportable segments. No operating income is attached to it. There is a gross margin, and it fell: Microsoft Cloud gross margin percentage "decreased to 66% driven by continued investments in AI infrastructure and growing AI product usage".
The second figure is Intelligent Cloud, which is a segment. It earned $56,972 million of operating income on $137,791 million of revenue, a 41.3% operating margin. Set beside AWS at 39.4%, that looks like a win. It is not a like-for-like one, because Intelligent Cloud is not Azure. The segment description lists SQL Server, Windows Server, Visual Studio, System Center, GitHub cloud services, Enterprise Support Services and Industry Solutions alongside Azure.
Azure itself gets one sentence in the entire annual report: "Azure and other cloud services revenue increased 41%." A growth rate, with no revenue, no cost and no margin beneath it. Anyone quoting an Azure operating margin is quoting an estimate, not a filing.
Oracle's 58.9% is not an operating margin, and Oracle says so in a footnote
Oracle's fiscal 2026 ended on 31 May 2026. Its cloud and software business reported revenues of $58,530 million and a line the company labels "Margin" of $34,468 million. That is 58.9%, which would make it comfortably the most profitable of the five.
The footnote directly beneath the table explains why it isn't the same animal: "The margins reported reflect only the direct controllable costs of each line of business and do not include allocations of research and development, general and administrative and certain other allocable expenses, net." The same footnote also excludes amortisation of intangible assets, restructuring, stock-based compensation and interest expense.
Those are real costs. Oracle spent $10,272 million on research and development in fiscal 2026 and $1,618 million on general and administrative expenses. Charge all of both to the cloud and software line, which certainly overstates its share, and the margin falls to $22,578 million, or 38.6%. That's an outer bound rather than an estimate of the true figure. It lands roughly where AWS sits, not far above it.
Two things sit underneath that margin and outside it. Oracle's capital expenditures were $55,663 million in fiscal 2026, up 162% on the prior year. Its free cash flow, which Oracle calculates as operating cash flow less capital expenditures and labels non-GAAP, was negative $23,686 million. The segment measure excludes very nearly the whole of the spending that pushed free cash flow below zero.
Alibaba reports adjusted EBITA, and IBM has no cloud segment at all
Alibaba's Cloud Intelligence Group reported revenue of RMB158,132 million for the year to 31 March 2026, and adjusted EBITA of RMB14,265 million. That's a 9.0% margin. The 20-F defines the measure as net income before interest, tax and "certain non-cash expenses, consisting of share-based compensation expense, amortization and impairment of intangible assets, impairment of goodwill, and others".
So Alibaba's 9.0% is struck before several costs that AWS's 39.4% is struck after. It's also in renminbi, on a March year end, at a company whose chief operating decision maker reviews adjusted EBITA rather than operating income.
IBM goes further and has no cloud segment at all. Its four reportable segments are Software, Consulting, Infrastructure and Financing. For the quarter to 30 June 2026 it reported $17,110 million of segment revenue and $4,092 million of segment profit, and states the margin as 23.9%. Cloud appears in the narrative and in product names, never as a reporting line. A reader who wants IBM's cloud margin has nothing to build one from.
What cloud segment reporting leaves out is the capital
The larger gap sits below the profit line, in the assets that produce it.
Amazon publishes segment capital spending, and the numbers are startling. In the quarter to 30 June 2026, net additions to property and equipment in the AWS segment were $48,604 million, against $16,043 million in the same quarter of 2025. AWS depreciation and amortisation for the quarter was $8,076 million, up from $4,844 million, a rise of 66.7%. AWS property and equipment stood at $263,750 million at the end of June. Depreciation ran at 16.6% of that quarter's own additions, which is what a business early in a capital cycle looks like.
Nobody else in this group gives you that. Microsoft's 10-K is explicit: "Assets are not allocated to segments for internal reporting presentations," and "It is impracticable for us to separately identify the amount of amortization and depreciation by segment that is included in the measure of segment profit or loss." Alphabet states that "Our operating segments are not evaluated using asset information." Oracle states that "We do not track our assets for each business." Alibaba says it does not allocate assets to segments either.
The consolidated capital spending is enormous and unallocated. Microsoft added $115,948 million of property and equipment in fiscal 2026, against $64,551 million the year before, a rise of 79.6%. Alphabet spent $80,598 million in the six months to 30 June 2026, against $39,643 million in the same half of 2025. Neither company tells you how much of that went to the cloud segment whose margin it reports.
That matters because depreciation on those assets lands in segment operating income later, and the same AI capex depreciation lag flatters every reported cloud margin while spending is still climbing. A margin that rises while its own asset base goes unmeasured has a known direction of error and an unknown size. AWS is the one place in this group where you can size it.
The objection: a common definition would describe a business nobody runs
The strongest objection to everything above comes from the standard itself, and it deserves a hearing. Segment reporting under ASC 280 is a management-approach rule. A company reports what its chief operating decision maker actually reviews, not what an outside reader would prefer to see.
The filings name those people. Amazon's decision maker is its President and Chief Executive Officer. Alphabet's is Sundar Pichai. Oracle's are its Chief Executive Officers and Chief Technology Officer. If Oracle's management runs the cloud and software business on direct controllable costs, then a margin carrying allocated research and development would describe a business nobody inside Oracle is actually managing. Forcing a common definition would make every segment number less true to how these companies are run, which is the opposite of what disclosure is for.
Disclosure has also improved. Oracle's segment note now presents, in its own words, "significant segment expenses categories and amounts on a segment basis and included within each reported measure of a segment's profit or loss, that is regularly provided to our CODMs". Alphabet now splits Google Cloud's costs into employee compensation of $6,933 million and other costs and expenses of $9,021 million for the quarter. That is more than a reader had two years ago.
Alphabet keeps its frontier model spending outside Google Cloud
One line in Alphabet's segment note deserves more attention than it gets. Certain costs are not allocated to any segment, and the first item on that list is "certain AI-focused shared research and development activities, including employee compensation expenses and technical infrastructure usage costs associated with the development of our general AI models".
Those Alphabet-level activities cost $5,789 million in the quarter to 30 June 2026, up from $3,372 million a year earlier. Google Cloud sells access to the output. The same 10-Q lists Vertex AI and Gemini Enterprise among Google Cloud Platform's services. The models are developed at Alphabet level, and the revenue is booked in Google Cloud.
The arithmetic is worth doing in both directions. Charge every dollar of Alphabet-level activities to Google Cloud and the segment's margin drops from 35.6% to 12.2%. Allocate it by revenue instead, where Google Cloud is 20.7% of the $119,690 million booked across the three segments, and the charge is roughly $1,198 million and the margin becomes 30.7%. Neither figure is correct. The distance between them is the size of the question the disclosure leaves open, and it is wider than the entire gap between AWS and Google Cloud.
The same structural point runs through AI gross margin comparisons generally: the cost of building a model and the cost of serving it land in different places, and often at different companies.
What these filings cannot tell you
Start with what none of them contain. No company in this group publishes an audited profit figure for Azure, for Oracle Cloud Infrastructure, or for Google Cloud Platform separately from Google Workspace. Every widely quoted margin for those individual products is somebody's estimate, built on assumptions the filings do not confirm.
The periods don't line up either. Amazon and Alphabet report calendar quarters. Microsoft's year ends on 30 June, Oracle's on 31 May, Alibaba's on 31 March. Comparing a quarter against a full year flatters whichever business is growing faster, and all of them are growing.
Allocation methods belong to management, and they move. Amazon says technology infrastructure assets are allocated among segments "based on usage", and that the allocation of net additions "can fluctuate on a quarter-to-quarter basis". Alphabet allocates technical infrastructure and office facilities to segments "as a service cost generally based on usage, headcount, or revenue". A margin built on those allocations is an accounting output, not a measurement of a standalone business.
None of this is evidence that any of these companies is better or worse run than another. It is evidence that the published figures do not answer that question. A segment margin, however defined, also says nothing about the profitability of any individual contract inside it, and the disclosed Google Cloud operating margin is an average across a very mixed book.
What would change the picture
The cleanest change would be Microsoft breaking Azure out as a reportable segment. Under the management-approach rule that follows from how Microsoft organises itself rather than from what investors ask for, so it would signal a genuine reorganisation rather than a disclosure decision.
A second change is already scheduled. Alphabet's 10-Q describes a new expense-disaggregation standard that, upon adoption, requires it to "disclose in the notes to the financial statements a disaggregation of certain expense categories included within the relevant expense captions on the consolidated statements of income". The filing states that the standard "is effective for our 2027 annual period, and our interim periods beginning in 2028". That disaggregation is consolidated rather than segmental, so it leaves the capital question open.
The figure that would settle most of this is smaller and duller than any of the ones above. It is segment capital expenditure. Amazon already publishes it, which is why AWS is the only cloud business here whose reported margin can be read against the assets generating it. Until the other four do the same, cloud segment reporting supports exactly one honest comparison on this page, the one Amazon and Alphabet permit between themselves: 39.4% against 35.6%, for the same three months, on the same definition. Everything else on the chart answers a different question while wearing the same label.