Executive Summary

Key Numbers

Four numbers carry this filing. The first three are values Microsoft wrote down. The last slot is one where nothing has been written at all.

Sources: TechCrunch · Microsoft quarterly report (October 28, 2026)

On October 28 Microsoft reported results for the quarter ended September 30. Revenue of $92.3 billion made it a tenth consecutive record quarter, with Microsoft Cloud at $63.9 billion and Azure at $33.2 billion. The record is what became the headline, but inside the quarterly report submitted the same day sat a line that had not been visible on a quarterly basis until now. It says Microsoft recognized $10 billion in revenue this quarter from its commercial dealings with OpenAI. What it collects under the heading of revenue share is included there.

Against $92.3 billion, $10 billion works out to a share of about 11%. The party that produced that ratio, though, was not the company. On the earnings call Gabriela Borges of Goldman Sachs summed it up and asked back whether OpenAI accounted for 11% of revenue in the quarter, and Satya Nadella, the chief executive, declined to take the ratio up. He answered instead that customers mix several models and train their own as well. It is worth remembering whose mouth the widely quoted 11% came out of.

Next to that $10 billion sits one more number. It is the amount OpenAI agreed to pay that has not yet come in, which accountants call a receivable balance. That balance stood at $6 billion on June 30 and reached $7.2 billion on September 30. Over three months it grew by $1.2 billion, a rise of 20%.

Once those two numbers landed in the same paragraph, a reading spread quickly: Microsoft earned $10 billion and $7.2 billion of it never arrived. Read that way, the conclusion is that something like $2.8 billion actually reached the company. But the two numbers are not a pair you subtract.

In corporate accounting, revenue is booked on the day the promised thing is handed over, not the day the payment arrives. If Microsoft ran cloud computing for OpenAI over three months and earned its revenue share along the way, that amount becomes revenue for the quarter even when the payment has yet to touch a bank account. In any industry where goods go out first and payment follows, the same thing happens.

The part delivered but not yet collected does not simply vanish; it stays on the balance sheet under the name of a receivable. What sits there is not only what arose this quarter but also what carried over from earlier quarters. So the $7.2 billion is not the amount generated by business between July and September. It is everything still uncollected as of that one day, September 30, added together.

Appearing in the same report is all the two numbers share; what they count is different. The $10 billion is an amount accumulated over a stretch of three months, and the $7.2 billion is a photograph taken on the day that stretch ends. An amount over a period and a balance on a day carry different units, so neither addition nor subtraction applies.

So what does it mean that $1.2 billion more piled up. The most careful answer goes as far as saying the pace of handing over and the pace of being paid have moved apart. When OpenAI's usage climbs quickly, invoices whose due dates have not yet arrived climb along with it, so a growing receivable is natural for a business that is expanding. Payment terms stretching from 30 days to 60 produce the same shape. There is not enough here to translate the figure directly into an inability to pay.

Telling those three apart would take the amount actually past due and the payment terms on the contract. Neither appears in the material that has been released. What is certain for now is the single fact that the gap widened, and why it widened still has several answers to choose from.

Putting both numbers against the same four questions shows where they part ways. What is being counted, what moment the value captures, whether cash changed hands, and what news it makes when the value grows. All four questions come back with different answers. Which means that putting the two on one line and doing arithmetic with them does not hold up.

Laid out in a table, one more thing stands out. Two numbers this different in character ended up written into the same place in the same document, and it is worth asking why.

Few companies write down separately how much they earned from one customer and how much remains to be collected from that customer. Microsoft does so because OpenAI is not an ordinary customer. As of the end of September it holds roughly 24% of OpenAI, which under accounting rules classifies the business between them as dealings with a related party. Such dealings require the size and the outstanding balance to be stated on their own.

The stake keeps drifting down. Right after OpenAI reorganized its structure last year it stood at about 27%, then about 25% in June of this year and about 24% in September. Each new investor coming in dilutes what was already there. Even at this level, though, the related-party status holds, so the duty to disclose stays in place.

That means the duty is not something that arose this quarter. The stake dates to last year and the business between them has run since then. What changed this time is the unit of disclosure. Amounts that used to appear bundled into a full year now stand on their own as a three-month figure.

On the timing of the disclosure, one further reading has attached itself. Olga Usvyatsky, an accounting researcher at Nonlinear Analytics, suggests the first appearance of these numbers on a quarterly basis may not be unrelated to the timetable for OpenAI's public listing. That is a hypothesis and nothing more, and Microsoft has never said as much.

The name OpenAI turns up on more than one line of this report. There is $10 billion on the revenue line, $7.2 billion on the receivable line of the balance sheet, and another figure on the line for gains and losses from the equity stake. The stake-related result this quarter was net income of $272 million. In the same quarter a year earlier it was a net loss of $4.1 billion.

The money going the other way sits in the same document. Microsoft has committed a total of $13 billion to OpenAI, and $12 billion of that had been funded as of September 30. One company is a shareholder in the customer it sells cloud capacity to, and carries a balance that amounts to having lent that customer money.

The revenue sharing between the two companies also runs in one direction. Microsoft no longer carries an obligation to share revenue with OpenAI, while OpenAI's revenue-share payments to Microsoft are reported to continue through 2030. A condition is also said to be attached under which the payments stop earlier once the cumulative amount reaches a set ceiling, and how high that ceiling sits has not been disclosed. This asymmetry makes up a substantial part of the $10 billion.

The name given to the criticism of this structure is circular financing. If money Microsoft invested becomes OpenAI's cloud bill and returns as Microsoft's revenue, the argument runs, an outsider has a harder time telling demand a customer paid for out of its own pocket from demand that circulated within the industry. One analysis holds that a substantial part of the AI revenue Microsoft reports comes from OpenAI alone. Across the most recent full fiscal year, OpenAI-related revenue was $24.1 billion against total revenue of $331.8 billion, about 7.3%, and at this quarter's pace it would run near $40 billion a year.

Those making the criticism draw a line of their own. Every dollar that moved is real and disclosed, so this is a different matter from accounting fraud. The question left over is how to judge the quality of this revenue, and the detail needed for that judgment has not been released. A case where the same argument broke out over a chip seller buying stock in its customer was set down on this blog in the piece on Meta's equity raise. The story of AI companies counting the very word revenue in different ways sits in the comparison of how OpenAI and Anthropic tally theirs.

Microsoft's quarterly results are not something Pebblous needs to follow. The reason to look into this filing is where the numbers sit. Two of them are stuck together in one paragraph and both are denominated in dollars, yet one measures a period and the other measures a point in time. The table does not separate them; only the head of the person reading does.

To anyone who has handled data, the situation is familiar. One table holds a period-total column and a point-in-time balance column side by side, and the data type of both is a number. Unless which is which is written down somewhere, a query that subtracts the two runs without a single warning. The mistake a person makes reading "earned $10 billion, never collected $7.2 billion" is one a machine makes just as readily.

Totals that jump across layers are the same problem. In this arrangement Microsoft's $10 billion of revenue is, seen from OpenAI's side, a cost. Add up the revenue of chip companies and cloud companies and model companies in turn to size the AI market, and the revenue of the layer below gets counted once more as the cost of the layer above. Not one company's number is wrong, and yet the total alone swells.

If your organization publishes its own metrics outward, three things are worth settling.

• Is there a column recording whether each number measures a period or a point in time. Without one, subtraction happens inside the same table
• When a total is built, is the same money counted at two layers. Aggregations that sum revenue into a market size are especially prone to it
• Is the party that wrote the number the same as the party answerable if it turns out wrong. A value carried in a filing and a value computed on a call differ on exactly this point

That last item is the question this filing leaves behind. The number that traveled furthest this time was 11%, and the party that worked out the ratio and said it aloud was not the company but the analyst asking the question. As a number passes through several mouths, who wrote it down is the first thing to fall away.

Thank you for reading this far. The $10 billion in revenue, the rise in receivables from $6 billion to $7.2 billion, the $13 billion committed and $12 billion funded, and the stake of about 24% were confirmed in reporting that carried the contents of the quarterly report Microsoft filed on October 28, and the quarterly revenue of $92.3 billion along with the cloud and Azure results came from the earnings call held the same day. The 11% ratio, the answer from Nadella, the chief executive, and the remarks by Jackson Ader and Olga Usvyatsky are relayed from quotations in several outlets, since the original sources could not be confirmed. Setting revenue and receivables on separate axes in Section 2 and reading the whole thing as a data quality problem in Section 5 are this article's own interpretation. If your organization has put a period metric and a point-in-time metric into one table and lived with what followed, we would be glad to hear about it.

Pebblous Data Communication Team
October 30, 2026

Primary Filing

• 1.Microsoft Corporation. (2026). "Form 10-Q — Quarterly Report (FY2027 Q1, quarter ended 2026-09-30)." SEC EDGAR.

Reporting

• 2.TechCrunch. (2026). "Microsoft recorded $10 billion in revenue from OpenAI."
• 3.TechCrunch. (2026). "Microsoft hits 10 quarters of record-breaking revenue in a row."
• 4.24/7 Wall St. (2026). "OpenAI Now Makes Up 11% of Microsoft's Revenue."

Analysis

• 5.ExplainX. (2026). "Microsoft $10B Revenue From OpenAI in One Quarter — Filing, Receivables, Stake: What Is Confirmed."