Microsoft (OpenAI stake + Azure)
Satya Nadella (chairman/CEO); the OpenAI relationship is a corporate strategic investment, not a fund.
Microsoft's AI exposure is dominated by its ~27% economic interest in OpenAI Group PBC (~$135B) plus the Azure compute contract that monetizes it: OpenAI is committed to purchasing an incremental $250B of Azure cloud services. Microsoft retains exclusive IP rights and Azure API exclusivity until AGI, but gave up cloud right-of-first-refusal in the April 2026 terms reset.
- Not open — Not an open-ended vehicle — analysis only, not an available allocation.
- Reconstructed — Figures are third-party estimates rebuilt from filings, not reported results.
Top holdings
~$135B (~27%) — Largest external shareholder. Stake diluted from ~32.5% pre-recap as new investors entered and valuation scaled (~$500B at recap, $852B by March 2026).
$250B incremental — OpenAI contracted to buy $250B of incremental Azure services — the revenue side of the relationship; not an equity holding but the strategic payoff.
Recent moves
Oct 28, 2025: OpenAI completed for-profit restructuring; Microsoft's stake formalized at ~27% / ~$135B with a $250B Azure contract. April 27, 2026: partnership terms reset — revenue-share payments capped, Microsoft loses cloud right of first refusal but keeps IP + API exclusivity to AGI. Sources: Microsoft blog, OpenAI, CNBC, DCD.
Our take
The lowest-volatility way to own a top-tier frontier lab inside an investment-grade balance sheet — you get OpenAI economics plus the Azure consumption that the lab is contractually obligated to drive. Edge: the stake and the cloud revenue compound together. Caveats: the equity is private (no mark-to-market, but quarterly equity-method losses hit reported EPS, e.g. the $3.1B drag), the relationship is less exclusive after April 2026, and 'until AGI' is a legally fuzzy trigger. Exposure is real but buried inside a $3T+ diversified franchise.
The reported P&L line tells you almost nothing about OpenAI.
FY26 Q1 showed a $3.1B net-income hit; grossing that up implies roughly $10B of OpenAI quarterly losses, but that is a derived estimate on an unstable denominator — the quarter closed before the Oct 28 recap, when the economic interest was 32.5%, not 27%. By the nine months to Mar 31, 2026 the line had swung to $5.9B of net gains against $2.7B of net losses in the prior-year period. Almost none of that is OpenAI's operating economics improving. Two mechanical drivers explain it: equity-method loss recognition is bounded by carrying value, so once cumulative losses exhaust the funded basis the pickup collapses toward zero; and the October 2025 recap produced a dilution/remeasurement gain because implied valuation rose faster than ownership fell. The vanishing EPS drag is an accounting artifact — a non-cash, unmarked private position priced by OpenAI's own primary rounds, with cash economics that are Azure revenue from a counterparty Microsoft helped capitalize.
- 2026-08OpenAI has since filed confidentially for an IPO, reportedly targeting a 2027 listing near $1T. If it lists, the position becomes marked and eventually realizable — so 'unmarked' reads as timing, not structure.
Thesis
Not a fund — a vendor-financing structure wearing an equity stake. Microsoft's ~27% as-converted interest in OpenAI Group PBC (~$135B implied at the Oct 28, 2025 recap, cut from 32.5% of the old for-profit) is the equity leg of a commercial deal. That recap surrendered Microsoft's cloud right of first refusal for a $250B incremental Azure commitment; the April 27, 2026 amendment then ended model exclusivity and deleted the AGI clause, leaving a non-exclusive IP license to 2032 and capped OpenAI-to-Microsoft revenue share through 2030. The moat now carries expiry dates.
Capitalize a frontier lab, then monetize it as a customer. Microsoft committed $13B ($11.8B funded at Mar 31, 2026) for ~27% as-converted under the equity method, and books the real return via Azure consumption plus an IP license embedded in Copilot, Foundry and Office. Post-April-2026 the flows run one way: OpenAI keeps paying capped revenue share through 2030 while Microsoft stops paying revenue share to OpenAI, and the IP license continues to 2032 non-exclusively.
Assessment
- Structurally it sits inside an investment-grade, cash-generative balance sheet: no fee layer, no gate, no capital call, no forced-seller mechanic if sentiment turns.
- Microsoft monetizes twice — equity economics plus Azure consumption booked at its own margin — so the commercial leg pays even if the equity mark is never realized.
- The Oct 2025 recap traded the cloud right of first refusal for a contracted $250B incremental Azure purchase, converting an option into a stated obligation.
- Equity-method accounting, unlike fair-value marking, forces investee losses through reported income rather than burying them in other comprehensive income.
- Related-party circularity: much of the $250B Azure commitment is spending by an investee Microsoft capitalized, funded from capital markets, booked as cloud revenue with no disclosed decomposition.
- Two amendments, both downgrades dressed as continuity — Oct 2025 gave up the right of first refusal, April 2026 gave up model exclusivity and the AGI clause. A structural lock became a preferred-vendor relationship.
- Time-bounded moat: revenue share caps and ends 2030, the IP license expires 2032. A durable strategic asset does not come with two termination dates inside seven years.
- Disclosure is thin by design — no OpenAI segment, no broken-out carrying-value trend, and no way to separate OpenAI-driven Azure revenue from organic Azure.
- The 27% is as-converted and diluted by every subsequent OpenAI raise; the ~$135B implied value is a primary-round print, not a marked or realizable price.
- 2026-03OpenAI's $122B round closed 31 Mar 2026 at $852B post-money; at ~26.8% as-converted the same primary-round arithmetic implies ~$228B, not the ~$135B Oct-2025 print.
Record
There is no standalone return series for this stake and never will be — Microsoft reports none. The verifiable trace is accounting: FY26 Q1 (quarter to Sep 30, 2025) a $3.1B net-income decrease; the nine months to Mar 31, 2026 a $5.9B net gain versus $2.7B of net losses in the comparable prior period, driven primarily by the October 2025 recap dilution gain. A near-zero FY26 Q3 quarterly pickup has been reported secondhand; we have not confirmed it against the 10-Q itself. Read the sequence as mechanics — carrying-value exhaustion plus a one-time remeasurement — not as evidence OpenAI turned a corner. Attribution of MSFT shareholder returns to this stake is not separable from Azure, Copilot pricing, Office attach and the capex cycle. Anyone quoting a 'return on the OpenAI stake' is quoting a valuation Microsoft itself did not set and cannot sell into.
- 2026-07FY26 (to 30 Jun) is now disclosed in full: +$4,963M net income, +$0.67 EPS, against FY25's -$3,620M / -$0.49. The full-year gain sits below the nine-month $5.9B cited, so the two are on different bases.
Risks & fit
- Carrying-value floor: with $11.8B of $13B funded, further OpenAI losses may not flow through income at all — reported EPS stops reflecting investee burn.
- Post-ROFR and post-April-2026, OpenAI can buy compute from and serve customers on any cloud, eroding the Azure leg while the equity stake stays illiquid.
- A down round or IPO priced below current private marks would force a visible remeasurement against a stake with no offsetting hedge.
- AI capex intensity is the larger balance-sheet exposure; the stake is small next to the datacenter buildout it helps justify.
- Counterparty concentration: governance friction or key-person loss at OpenAI has no substitute inside the current structure.
The skeptical read breaks if Microsoft begins disclosing OpenAI-attributable Azure revenue separately and it proves large and growing at company-average-or-better margin — showing the commercial leg stands on its own rather than being funded in a circle. It also breaks if OpenAI reaches positive operating cash flow before 2030, making the capped revenue share and 2032 license expiry moot. Conversely, the constructive read breaks if multi-cloud migration shows up as decelerating Azure growth while the $250B commitment is restructured or quietly allowed to lapse.
Analytically, this is the least-volatile and least-transparent way frontier-lab economics surface inside a public security: a small contributor buried in a multi-trillion-dollar franchise, valued by a private round rather than a market, with no separable performance series. It rewards readers who care about accounting mechanics and the contract-expiry schedule; it frustrates anyone seeking clean pure-play AI exposure or a markable position.
No management or performance fee — a corporate balance-sheet holding, not a managed vehicle. The implicit cost: shareholders fund the capex and the commitment while having no ability to size, hedge, or exit the OpenAI exposure independently of MSFT itself.