The OpenAI Stake

The OpenAI Stake

Microsoft owns about 27% of OpenAI, funds it, and sells it Azure capacity — one relationship that now sits on both sides of the ledger, flattering reported earnings while backing much of the contracted demand behind the AI build. In the December 2025 quarter investments in OpenAI added $10.0 billion to other income and $1.02 to diluted EPS, lifting reported EPS to $5.16 (+60%) against adjusted EPS of $4.14 (+24%), while in the March 2026 quarter that same OpenAI relationship carried Microsoft's commercial RPO to +99% growth ($627 billion) versus only +26% excluding OpenAI and turned commercial bookings to -4% including OpenAI versus +7% without it — and Microsoft's equity-method loss from OpenAI was about $4.1 billion in the September 2025 quarter alone, nearly the entire $4.7 billion 'Other, net' drag recorded across all of fiscal 2025, even as Microsoft surrendered its right of first refusal to be OpenAI's compute provider as OpenAI's Azure commitment rose by an incremental $250 billion. One counterparty thus supplies about 20% of the December-quarter diluted EPS ($1.02 of $5.16) and roughly 73 points of the 99% commercial RPO growth: the same loss-making, part-owned customer does double duty as earnings flatterer and demand backstop. The counter-fact sits in the same disclosure — the funding is capped at a disclosed $13 billion, Microsoft voluntarily publishes an adjusted EPS that excludes OpenAI, and the equity-method losses are immaterial against net income above $100 billion, disclosure hygiene rather than concealment. Reported earnings have nonetheless become a weaker guide to run-rate profit than at any recent point.

A stake that reversed the sign on earnings

For two years the OpenAI investment was a drag on reported profit. Microsoft accounts for it under the equity method, recognizing its share of OpenAI's operating losses through the "Other, net" line of other income. That line deepened from a $223M loss in FY2023 to $1,319M in FY2024 and $4,725M in FY2025, and management attributes it primarily to net losses on equity-method investments, including OpenAI [1]. In the nine months to March 2025 the OpenAI drag alone was $2.7B [2].

The recapitalization reversed that. On October 28, 2025 Microsoft signed a new agreement, OpenAI formed a public benefit corporation, and Microsoft's holding was set at approximately 27% on an as-converted diluted basis. The step-down in proportionate ownership produced a dilution gain [3]. In the December 2025 quarter, investments in OpenAI added $10.0 billion to other income, and $5.9 billion for the six months — figures the filing states relate primarily to that dilution gain [4].

Source: Q3 FY2026 Form 10-Q, Note — Other Income (Expense), Net [5].

Because the gain landed in one quarter, it dominates the year. The March 2026 quarter carried a negligible $14M OpenAI drag on net income [6], so essentially all of the nine-month swing is the single recapitalization event. Microsoft itself now publishes an adjusted diluted EPS that strips OpenAI out. On that measure, the December quarter's headline 60% EPS growth becomes 24%, and the OpenAI dilution gain accounts for $1.02 of the $5.16 reported [7].

Source: Q2 FY2026 Form 10-Q, Non-GAAP Financial Measures reconciliation [8].

The sign flipped. A year earlier OpenAI was a small drag, so adjusted EPS sat above reported; after the recapitalization it is a boost, so reported sits above adjusted. Neither direction reflects Microsoft's operating performance.

How the number is built

The gain is an accounting re-measurement of a private stake, not cash and not operations. OpenAI is a variable interest entity, but Microsoft has determined it is not the primary beneficiary, so the entity is not consolidated and is carried under the equity method [9]. Microsoft does not simply book 27% of OpenAI's results. It uses the hypothetical liquidation at book value method — recognizing income or loss based on the change in what it would receive if OpenAI's net assets were distributed at book value — because its liquidation rights and priorities differ from its ownership interest [10].

The recapitalization also changed the basis of that calculation. As CFO Amy Hood described it, Microsoft now records gains or losses "based on our share of the change in their net assets on their balance sheet as opposed to our share of their operating profit or losses from their income statement" [11]. A change in a book-value measurement drove other income to $10 billion; adjusted for OpenAI, the same line was slightly negative [12].

The stake is small on the balance sheet relative to its earnings effect. Equity-method investments — of which OpenAI is the dominant part — rose from $6.0 billion at June 2025 to $11.5 billion at December 2025, against total funding commitments of $13 billion, $11.7 billion of it funded [13]. An investment carried near $11.5 billion generated a $10 billion swing in reported other income in a single quarter. That mechanics — a private, unobservable book value driving a nine-figure earnings line — is why Microsoft reports an adjusted number that excludes it.

The circular demand

The same counterparty sits on both sides of Microsoft's ledger. Microsoft funds OpenAI and holds roughly 27% of it; OpenAI is at the same time one of Microsoft's largest cloud customers.

Sources: Q1 FY2026 Form 10-Q, Note 17 [14]; Q2 FY2026 Form 10-Q, Note 1 [15].

The link is visible in the bookings line. In the September 2025 quarter commercial bookings rose 112%, which management attributed to Azure commitments from OpenAI alongside growth in large contracts — and that figure predated the new deal [16]. The October agreement then added an incremental $250 billion of contracted Azure purchases by OpenAI [17]. That commitment flows into commercial remaining performance obligation — $392 billion and up 51% at the September quarter, on a weighted-average duration of about two years [18]. The contracted backlog that underwrites the capital program is therefore concentrated in part on a counterparty Microsoft part-owns and funds.

The exclusivity that once justified the investment has loosened as the dollars have grown. The FY2025 10-K described OpenAI's API as exclusive to Azure with Microsoft holding a right of first refusal on OpenAI's new capacity needs [19]. Under the October agreement, Microsoft will no longer have that right of first refusal to be OpenAI's compute provider [20]. The relationship is deeper in committed spend and looser in exclusivity than it was a year ago.

The capital that is not on the balance sheet

Much of the datacenter capacity that serves the OpenAI-anchored demand is contracted through leases not yet on the balance sheet — $196.6 billion of not-yet-commenced leases, primarily for datacenters, at March 2026 [21] — committed capital the reported capex and depreciation lines do not yet carry; the Margin Bridge chapter walks that lease book and its finance-lease conversion in full.

What would change the read

OpenAI has turned Microsoft's GAAP earnings into a poor guide to run-rate profit — a multi-billion drag that became a one-time boost — and has tightened the loop between Microsoft's own capital, its equity stake, and the cloud demand that justifies the build, while the not-yet-commenced lease book shows the capital commitment is larger than the reported capex implies.

The strongest fact against reading this as a concern is Microsoft's own transparency. It discloses the HLBV method, the $13 billion commitment, the not-yet-commenced leases, and — unusually — publishes an adjusted EPS that removes the OpenAI noise entirely [22]. The equity-method losses are immaterial against net income above $100 billion, and the $250 billion is contracted revenue from a counterparty whose demand, management says, exceeds available supply [23].

Three things would sharpen the read. First, disclosure of OpenAI-related revenue as a separate line would let an investor judge how much of the RPO backlog is genuinely third-party. Second, the not-yet-commenced lease book flattening or falling would signal demand catching up to committed supply — while continued growth signals the opposite. Third, the pace at which finance-lease interest and depreciation from commenced datacenters land in the income statement will show whether the committed capital compresses the margins consensus extrapolates. Until OpenAI's own funding position and the pricing of its Azure commitments are visible, the circularity can be described but not fully priced.