Microsoft's OpenAI Equity Method Loss: $4.1B in One 10-Q, Then a $10B Gain

In the quarter to 30 September 2025, Microsoft booked a $4.1 billion pre-tax loss from its OpenAI stake. One quarter later the same line showed a $10.0 billion gain. Neither number appears in capex, and neither touches operating income. The Microsoft OpenAI equity method loss lives in one footnote of the 10-Q, and it is the cleanest public window into what the partnership costs Microsoft as an owner rather than as a supplier.

Where the OpenAI equity method loss appears in the 10-Q

Equity-method accounting applies when a company owns enough of another business to influence it, without controlling it. The investor does not consolidate the other company's revenue and costs. Instead it books its share of the investee's profit or loss as a single number.

Microsoft holds an approximate 25% interest in OpenAI on an as-converted basis, according to its fiscal 2026 10-K, and accounts for it under the equity method. The result does not sit in cost of revenue, research and development, or any operating line. It sits in other income (expense), net, which is calculated after operating income.

Note 3 carries the pre-tax number

Every Microsoft 10-Q has a note titled Other Income (Expense), Net. It is Note 3 in the fiscal 2026 quarterly filings. The table breaks the line into interest income, interest expense, gains on investments, derivatives, currency remeasurement, and a residual called "Other, net".

Below the table is one sentence that matters more than the table. In the September 2025 10-Q it reads that other income (expense), net included $4.1 billion and $688 million, respectively, of net losses from investments in OpenAI. Those losses are recorded mainly in "Other, net", which that quarter was a $4.923 billion expense.

So the computation the headline promises is short. Read the OpenAI sentence in Note 3. That is the pre-tax equity-method result for the quarter. Everything else in this post is context for that one number.

The press release carries the after-tax version

Microsoft's earnings release reports a second, different figure. Since fiscal 2026 it publishes non-GAAP results that exclude the OpenAI impact, and it states that impact after tax. For the September 2025 quarter the release says OpenAI losses reduced net income by $3.1 billion and diluted earnings per share by $0.41.

That is why coverage of the same quarter quoted both $4.1 billion and $3.1 billion. Neither was wrong. One is pre-tax from the 10-Q, the other is after tax from the release. I would always cite the pair together and label which is which, because quoting one against the other's prior-year figure produces a false growth rate.

Eight quarters of Microsoft's OpenAI line, computed

Microsoft gives quarterly figures in the 10-Qs and annual figures in the 10-K. The fourth quarter of each fiscal year is never stated directly. It has to be derived by subtracting the nine-month figure from the full year, which is what the starred rows below do.

Pre-tax figures from Note 3 of each 10-Q and the fiscal 2026 10-K. After-tax and EPS figures from the earnings release reconciliations. *Derived: full-year Note 3 figure less nine-month figure, so subject to rounding of up to $0.1 billion.

Fiscal 2025: a loss that kept growing

The pre-recapitalization pattern is the useful one for anyone modelling the stake. Losses ran from $688 million in the September 2024 quarter to about $2.1 billion in the June 2025 quarter, then $4.1 billion in September 2025. That is roughly a sixfold increase across four quarters.

Across fiscal 2024, fiscal 2025 and the first quarter of fiscal 2026, Microsoft recorded about $10.4 billion of pre-tax OpenAI losses. That is close to the $11.9 billion Microsoft had funded by June 2026. On the income statement, most of the cash Microsoft put in had already been expensed through other income before the recapitalization reset the account.

Those losses grew in step with OpenAI's spending, which is the expected pattern when an investee is burning cash to buy compute. How OpenAI's own valuation compares with that burn is covered in the analysis of OpenAI's valuation against its run rate.

How the recapitalization turned a loss into a $10 billion gain

In the December 2025 quarter the OpenAI line flipped to a $10.0 billion net gain. Note 3 says the gain primarily relates to the dilution gain from the OpenAI Recapitalization, completed in October 2025.

A dilution gain arises when an investee issues new equity at a price above the investor's book value per share. The investor's percentage falls, but the book value of its slice rises, and the difference is booked as a gain. Microsoft's 10-K says its proportionate ownership of OpenAI decreased during fiscal 2026 because of the recapitalization and other funding activity.

The effect on reported results was large. The December 2025 quarter earnings release shows GAAP net income of $38.458 billion. Of that, $7.583 billion came from the OpenAI gain after tax. That is 19.7% of the quarter's net income, from a line that sits below operating income.

HLBV, in one paragraph

Microsoft says it calculates its OpenAI result using the hypothetical liquidation at book value method, usually shortened to HLBV. The method asks what Microsoft would receive if OpenAI were liquidated at book value on the balance sheet date, under the actual liquidation preferences. The change in that amount between two dates is the quarter's income or loss. It is used when an investor's rights to proceeds differ from its headline ownership percentage, which was true of Microsoft's capped-return arrangement with OpenAI.

That is the reason the line behaved so differently after the recapitalization. A change in OpenAI's capital structure changes the waterfall that HLBV runs through. So the post-recapitalization figures of a $19 million loss and a roughly $0.6 billion gain should not be read as OpenAI suddenly breaking even. They show a new structure, not a new business.

How much of Microsoft's AI bet sits outside capex and operating income

Most coverage of Microsoft's AI spending tracks capital expenditure. Capex captures servers, chips and buildings, and it reaches the income statement through depreciation inside operating costs. That is the visible half of the bet, and it is enormous: $115.9 billion of additions to property and equipment in fiscal 2026, up from $64.6 billion a year earlier.

The OpenAI stake follows a different path through all three statements. The cash goes out as an investment, not as property. The returns come back through other income, not through operating income. And since fiscal 2026 Microsoft strips the result out of its non-GAAP numbers entirely.

Sources: Microsoft fiscal 2026 10-K and fourth quarter earnings release. The $24.1 billion revenue figure and the funding figures are as quoted from the 10-K by Techzine, August 2026. Revenue includes revenue-sharing payments and is not broken down further.

Operating income never sees it

The ratio worth computing is the OpenAI line against operating income, quarter by quarter. In the September 2025 quarter, the $4.1 billion loss equalled 10.8% of Microsoft's $37.961 billion operating income. It also equalled 21% of that quarter's $19.394 billion of additions to property and equipment.

In the December 2025 quarter the gain equalled 26% of operating income. Over fiscal 2026 as a whole, the $6.5 billion net gain was 4.2% of operating income of $155.2 billion. Those are not rounding errors, and none of them show up in the operating margin that analysts quote.

Cash funded, not capitalised

The $11.9 billion Microsoft has put into OpenAI is the part of its AI bet that no capex tracker counts. It is smaller than one quarter of current capex, so it does not change the headline. It does change the shape of the risk. Capex buys assets Microsoft controls. The equity stake buys a claim on a company that Microsoft does not control, accounted for through a method that can swing $14 billion between two quarters.

Microsoft's aggregate equity-method investments, which include OpenAI and every other such holding, rose from $1.8 billion at 30 September 2025 to $11.5 billion at 31 December 2025, and stood at $12.0 billion at 30 June 2026. The jump tracks the dilution gain closely. The filings give the aggregate, not an OpenAI-only carrying value, so treat that as a strong inference rather than a disclosed figure.

This is a different question from the circular-deal debate, though the two meet. A dollar Microsoft invests can return as Azure revenue, and the auditor tests for that loop are set out in the piece on how auditors test circular AI revenue. The equity-method line is the owner's side of the same relationship, and Nvidia's equivalent disclosures are traced in the Nvidia filings trace on its OpenAI investment.

Where this analysis is weakest

The argument above rests on footnotes that Microsoft writes for compliance, not for clarity. Three gaps limit what any outside reader can conclude, and one popular calculation is wrong.

The dilution gain is not disclosed

Microsoft reports one net figure per period. The $10.0 billion December 2025 gain is a dilution gain minus Microsoft's continuing share of OpenAI's losses, and the filing does not separate the two. If the dilution gain was much larger than $10.0 billion, then the underlying loss that quarter was also large, and the "near zero" quarters that followed may hide more than they show. I cannot settle that from public documents, and neither can anyone else outside the company.

The same gap affects the fiscal 2026 total. A $6.5 billion net gain sounds like a profitable year for the stake. It is more accurately one large accounting event and three quarters of ordinary results netted together.

The gross-up trap

Several outlets divided Microsoft's September 2025 loss by its ownership percentage to infer OpenAI's own quarterly loss. Under HLBV, that arithmetic does not hold. Microsoft's share of a loss depends on the liquidation waterfall, not on a fixed percentage. I would not quote any OpenAI loss figure derived that way, including ones that happen to sound plausible.

There is also a timing issue that limits all of these figures. OpenAI is a private company, and equity-method results are based on the investee's financial information, which can lag. A quarter's OpenAI line may not reflect that calendar quarter's spending precisely.

Finally, the case against this whole exercise deserves a hearing. Microsoft's management excludes OpenAI from non-GAAP results precisely because it does not see the line as operating performance. On that view, focusing on it overstates its importance. I think the exclusion is reasonable for judging Azure, and wrong for judging the total return on Microsoft's AI strategy, because the stake and the commercial contract are two halves of one deal.

What to read in the October 28 10-Q

Microsoft has scheduled its fiscal 2027 first-quarter results for 28 October 2026, after the market closes. That quarter runs to 30 September 2026, so it is the first full year-on-year comparison against the $4.1 billion loss.

The comparison flatters by construction. Unless the post-recapitalization pattern breaks, the OpenAI line will show a large improvement against a quarter that was the low point of the series. Do not read that as a turnaround in OpenAI's economics. Read it as the base effect it is.

The second check is the one most people skip. Microsoft's commercial revenue from OpenAI sits inside Azure and the segment numbers, while the ownership result sits in Note 3. A quarter in which Azure accelerates and the OpenAI line deteriorates would be the clearest sign that the two halves of the relationship are moving apart. For how the broader capex picture feeds into cash, the fiscal 2026 numbers are set out in the analysis of hyperscaler capex against operating cash flow.

The depreciation side of Microsoft's spending has its own accounting judgement, and the useful-life assumptions behind it are examined in the piece on server useful lives and hyperscaler profit. Between that and Note 3, you have both of the places where Microsoft's reported earnings depend on an estimate rather than on cash.

Frequently asked questions

Where to start

Open the September 2025 10-Q on EDGAR and find Note 3. Copy the OpenAI sentence and the "Other, net" row into a spreadsheet, then add one row per quarter from the table above. It takes about 20 minutes, and on 28 October you will have the base the new number has to be read against.

If you hold Microsoft or sell into its ecosystem, add one column next to it: Azure growth for the same quarter. The relationship between those two columns is the real story of the partnership, and no single headline will give it to you.

Last verified: 26 October 2026, against Microsoft filings for the quarters ended 30 September 2025 through 30 June 2026. Re-check after the 28 October 2026 filing.

Related reading