Microsoft and OpenAI: The Economics of the Deal

The cleanest way to misread the Microsoft and OpenAI partnership is to call it a $13 billion investment and stop there. The Microsoft OpenAI deal economics are not one number. They are a loop: Microsoft funds OpenAI partly in Azure compute credits, OpenAI commits to spend on Azure, that spend books back as Microsoft cloud revenue, and Microsoft also holds equity and a slice of OpenAI’s revenue.

Reading the deal means separating the cash that actually leaves Microsoft from the credits that recirculate, then asking who depends on whom. After OpenAI’s October 2025 restructuring and the April 2026 amendment, those dependencies changed in ways the headlines mostly skipped.

This piece reads the structure through the filings and the dated announcements, not the keynotes. Every figure below is tagged either to a Microsoft SEC filing and fiscal period, or labeled as a reported estimate when it comes from announcements and press rather than a disclosure line. The framing is analytical: how the money moves and where the structure is fragile, not what to do about any stock.

Key takeaways

• A large share of the “investment” recirculates. Reporting indicates roughly half of Microsoft’s earlier $10 billion 2023 commitment was structured as Azure credits, not cash (reported, not a filing line). Credits committed to compute book back as Azure revenue, so the cash permanently leaving Microsoft is smaller than the headline.
• The stake is large but volatile. Microsoft holds roughly 27% of OpenAI on a fully diluted basis after the October 2025 restructuring, at a reported implied value near $135 billion (Bloomberg, CNBC, Fortune, Oct 28, 2025). It is an equity-method holding whose carrying value swings with OpenAI’s valuation.
• The accounting whipsaws. Microsoft recognized a $4.1 billion equity-method loss in Q1 FY2026 (Form 10-Q, ended Sept 30, 2025), then $5.9 billion of net gains over the nine months ended March 31, 2026 (Form 10-Q, Q3 FY2026), driven mainly by non-cash dilution gains.
• Exclusivity ended. The April 2026 amendment removed Microsoft’s exclusive cloud rights and capped OpenAI’s revenue-share payments at a reported $38 billion through 2030 (CNBC, Apr 27, 2026), converting an open-ended claim into a bounded one.
• The $250B Azure number is a floor, not a fence. OpenAI’s reported $250 billion incremental Azure commitment (Oct 28, 2025) is a guaranteed revenue floor for Microsoft, not a guarantee of exclusivity.

---

What do the Microsoft OpenAI deal economics actually look like?

The Microsoft OpenAI deal economics resolve into four moving parts: cash and credits flowing into OpenAI, a 27% equity stake, OpenAI’s committed Azure spending flowing back to Microsoft, and a capped revenue share. The headline commitment overstates the cash at risk, because much of it recirculates as compute that becomes Microsoft revenue.

That last clause is the whole story. A dollar of Azure credit is not a dollar out the door. It is a claim on Microsoft’s own data center capacity that, once consumed, shows up as cloud revenue. So the deal is less a one-way investment and more a closed loop with an equity option attached.

The point of this analysis is not to score the partnership. It is to read each flow, mark what is filed versus reported, and find where the structure is load-bearing. The same logic that makes a single owned input decide the economics of everything above it, the pattern at the center of the AI infrastructure market map, operates here too: Microsoft owns the compute layer OpenAI rents.

---

The Microsoft-OpenAI Deal Map: what moves, which direction, and what is filed

Here is the central framework for this piece, which I will call the Microsoft-OpenAI Deal Map. It is the analytical asset the rest of the analysis builds on, and it is meant to be citeable on its own: a one-screen read of what moves between the two companies, in which direction, and whether the figure comes from an SEC filing or from reporting.

Read it as a flow table. The “Filed vs reported” column is the discipline most coverage skips.

Flow	Direction	Magnitude	Filed vs reported
Total committed investment	Microsoft to OpenAI	$13 billion total; $11.8B funded as of Mar 31, 2026	Filed (Microsoft Form 10-Q, Q3 FY2026)
Funding split: cash vs Azure credits	Microsoft to OpenAI	~50% of earlier $10B as credits	Reported (press), not a filing line
Equity stake	OpenAI to Microsoft	~27%, implied value ~$135B	Stake reported (Bloomberg/CNBC/Fortune); equity-method treatment filed
Equity-method result	OpenAI results to Microsoft P&L	-$4.1B (Q1 FY2026); +$5.9B net (9 mo to Mar 31, 2026)	Filed (Microsoft Forms 10-Q)
Incremental Azure commitment	OpenAI to Microsoft	~$250 billion	Reported (announcement Oct 28, 2025)
Revenue share	OpenAI to Microsoft	20% of revenue, capped ~$38B through 2030	Reported (CNBC, Apr 27, 2026)
IP / model rights	Microsoft access to OpenAI IP	Non-exclusive, reported expiry ~2032	Reported

The Microsoft-OpenAI Deal Map. Filing-sourced figures are tagged to Microsoft’s 10-Q filings and fiscal periods; everything else is labeled reported because it comes from announcements and press, not a disclosure line.

Read the directions, not just the numbers. Money and credits go up to OpenAI; equity, committed compute spend, and a revenue share come back to Microsoft. The flows that are most certain (the equity-method results, the funded amount) are the filed ones. The flows everyone quotes (the $250B, the $38B cap, the 27%) are reported. That distinction is the difference between analysis and stenography.

---

Cash out vs. credits recirculated: how much actually left Microsoft?

Less than the headline. Microsoft’s total committed investment is $13 billion, of which $11.8 billion was funded as of March 31, 2026 (Microsoft Form 10-Q, Q3 FY2026). But reporting indicates roughly half of the earlier $10 billion 2023 commitment was structured as Azure credits rather than cash, so a meaningful share of the funding is compute that recirculates back as Microsoft revenue.

This is the part of the Microsoft OpenAI deal economics that the round-number coverage flattens. Treat the two paths separately.

• Cash out. Actual cash transferred to OpenAI is gone in the ordinary sense: it funds OpenAI’s operations and is not recoverable as Microsoft revenue.
• Credits recirculated. Azure credits are a commitment of Microsoft’s own capacity. When OpenAI consumes them, the consumption books as Azure usage. The credit is an expense to Microsoft, but the activity it funds runs on Microsoft infrastructure and supports Microsoft’s cloud revenue line.

The honest caveat: Microsoft’s SEC filings do not break out the cash-versus-credit split inside the full $13 billion. The roughly half-as-credits figure is reported, tied to the earlier $10 billion tranche, not a disclosure line. So the precise cash-at-risk number is not derivable from public filings. Directionally, the lesson holds: the cash that permanently left Microsoft is smaller than $13 billion, because a large portion is committed compute, and committed compute on your own cloud is closer to a transfer between pockets than a write-off.

This recirculation is the same dynamic that makes owning the rented layer so valuable. The tenant who pays you rent with credits you issued is a different kind of customer than one paying cash. It is the platform-ownership advantage that runs through Microsoft Copilot and enterprise lock-in: control the surface, and even your outflows tend to flow back.

---

The 27% stake and the $135B paper value: equity-method mechanics

Following OpenAI’s October 2025 for-profit restructuring, Microsoft holds roughly 27% of OpenAI on a fully diluted, as-converted basis, reported at an implied value near $135 billion (Bloomberg, CNBC, Fortune, Oct 28, 2025). That stake is accounted for under the equity method, which is why Microsoft’s reported “gain” on OpenAI can swing violently quarter to quarter without a dollar changing hands.

The 27% and the $135 billion are reported figures, drawn from the restructuring announcements rather than a single disclosure line. What is filed is the accounting treatment and its results. Under the equity method, Microsoft recognizes its share of OpenAI’s gains and losses in other income (expense), net.

That treatment produces two very different-looking quarters from the same investment:

• Q1 FY2026 (ended Sept 30, 2025): Microsoft recognized a $4.1 billion net loss from its equity-method OpenAI investment (Microsoft Form 10-Q, Q1 FY2026). OpenAI was burning capital; Microsoft booked its share of the loss.
• Nine months ended March 31, 2026: Microsoft recorded $5.9 billion of net gains from OpenAI (Microsoft Form 10-Q, Q3 FY2026), driven mainly by dilution gains.

A dilution gain is the counterintuitive one. When OpenAI raised capital at a higher valuation during the October 2025 restructuring, the implied value of Microsoft’s existing stake rose. Accounting recognizes part of that step-up as a gain, even though Microsoft sold nothing and received no cash. It is a mark, not a payment.

The takeaway for anyone reading Microsoft’s income statement: the OpenAI line in other income is not operating performance and not cash. It is a valuation echo. A big “gain” can mean OpenAI raised money expensively; a big “loss” can mean OpenAI is spending to grow. The discipline of separating real cash generation from accounting marks is the same skill in how to read a tech S-1 like an operator.

Methodology: reading the equity-method swings

• Inputs: Q1 FY2026 equity-method loss of $4.1B (Form 10-Q, ended Sept 30, 2025); $5.9B net gains over the nine months ended March 31, 2026 (Form 10-Q, Q3 FY2026); reported ~27% stake and ~$135B implied value (Oct 28, 2025).
• Assumption: the nine-month net gain is driven primarily by a non-cash dilution gain from OpenAI’s higher-valuation raise, partially offset by Microsoft’s share of operating losses.
• Sensitivity: if OpenAI’s next round prices flat or down, the dilution-gain mechanism reverses, and the same stake can produce a reported loss in a quarter with no change in the underlying business.
• What this misses: the filings disclose the net result, not a clean decomposition into “dilution gain” versus “share of operating loss,” so the split is inferred from the restructuring timing, not read directly off a line.

---

The $250B Azure commitment and the end of exclusivity

The reported $250 billion incremental Azure commitment (announced Oct 28, 2025) is the loop closing: OpenAI agreeing to spend back, on Microsoft’s cloud, an amount that dwarfs what Microsoft put in. But the same restructuring removed Microsoft’s exclusive cloud rights, which changes what that $250 billion actually guarantees.

Hold both facts at once. The $250 billion is a floor: a contractual minimum of OpenAI spending that Microsoft can count as guaranteed Azure revenue over the life of the agreement. That is real and large. It is also reported from the announcement, not a filing line, so treat the figure as a reported commitment rather than a disclosed obligation.

What the announcement also did was end exclusivity. Microsoft’s cloud rights to OpenAI became non-exclusive, with reporting putting the broader IP arrangement on a roughly 2032 horizon. OpenAI is now free to split workloads with other providers. So the structure shifted from “OpenAI runs on Microsoft” to “OpenAI must spend at least $250B on Microsoft, and may spend more elsewhere.”

That is a weaker form of lock-in than exclusivity, and it matters. A guaranteed floor protects Microsoft’s downside. The loss of exclusivity caps Microsoft’s upside, because OpenAI’s incremental growth above the floor can now flow to a competing cloud. Microsoft traded exclusivity for a contractual minimum and a cleaner equity position. Whether that trade is a strengthening or a concession depends entirely on how fast OpenAI grows past the floor.

---

Revenue share through 2030: the $38B cap

The April 2026 amendment is the most underrated change in the deal. Per CNBC (Apr 27, 2026), OpenAI’s revenue-share payments to Microsoft were reported as 20% of revenue but capped at $38 billion total through 2030, replacing what had been an open-ended percentage arrangement. A cap converts an uncapped claim on OpenAI’s growth into a bounded one.

Walk the logic. Under an uncapped 20% share, Microsoft’s payments scale forever with OpenAI’s revenue: if OpenAI becomes enormous, so does the share. A $38 billion cap severs that link past a point. Once cumulative payments hit the ceiling, OpenAI’s further growth no longer enriches Microsoft through the revenue-share channel; it only shows up in the equity stake.

That is a deliberate restructuring of who captures OpenAI’s upside:

• Below the cap: Microsoft participates in OpenAI’s revenue directly, payment by payment.
• At and above the cap: Microsoft’s exposure to OpenAI’s growth runs almost entirely through its 27% equity, not through the revenue share.

So the April 2026 amendment pushes Microsoft from being OpenAI’s metered partner toward being primarily its shareholder. The reporting also indicates the bilateral payment arrangement was simplified, with Microsoft no longer paying a revenue share to OpenAI. All of these are reported terms from the amendment coverage, not filing lines, so the precise mechanics should be read as reported.

For Microsoft, the cap is a known maximum on what it pays out, but it is also a ceiling on a once-uncapped inflow. Capping a revenue share is the kind of structural choice that looks small in a press release and reshapes the deal’s long-run economics, the same way pricing structure quietly decides outcomes in usage-based pricing vs seat-based pricing.

---

The circular compute model: why the stake’s value depends on the spend

Step back and the deal forms a circle. Microsoft funds OpenAI partly in Azure credits. OpenAI commits to spend on Azure. That spend books as Microsoft cloud revenue. Microsoft holds 27% of OpenAI, whose valuation is propped up in part by its access to that same Azure capacity. The value of Microsoft’s stake and the size of Microsoft’s cloud revenue are not independent; they reinforce each other.

This is the structural heart of the Microsoft OpenAI deal economics, and it cuts both ways.

• The virtuous read: every dollar Microsoft commits supports Azure utilization, which supports Azure revenue and margin, which supports the AI buildout, which makes OpenAI more capable, which raises OpenAI’s valuation and the carrying value of Microsoft’s stake. The loop compounds.
• The fragile read: the loop depends on OpenAI continuing to spend on Azure and continuing to raise at higher valuations. Weaken either, and the same reinforcement runs in reverse. A slowdown in OpenAI’s Azure spend pressures Microsoft’s cloud revenue; a down round pressures the stake’s carrying value through the dilution-gain mechanism working backward.

The cloud-revenue side of this circle is why the partnership shows up in Microsoft’s growth narrative far beyond the equity line. It is the same capital-intensity story reshaping cloud economics across the industry, dissected in AWS margin pressure and the cloud reset: the spend builds the moat, and the moat only pays off if utilization holds.

What makes the circular model defensible rather than merely circular is that Microsoft owns the layer in the middle. The compute is a real asset with real demand from outside OpenAI. The risk is concentration: a large, identifiable share of that demand sits with one counterparty.

---

The bear case: what the skeptics get right

A deal this favorably read deserves the strongest argument against it. The bear case is that the circular structure flatters Microsoft’s reported economics, hides single-counterparty risk, and rests on terms that just moved against Microsoft.

Start with the recirculation framing. The skeptic’s version is that calling Azure credits “not really an outflow” is too generous. Credits commit real capacity that has a real cost to build and power, and booking the funded spend back as revenue makes the relationship look more self-financing than it is. If OpenAI’s economics never reach durable profitability, Microsoft has financed a customer whose payments are partly funded by Microsoft’s own credits. That is closer to vendor financing than to a clean investment, and vendor financing flatters revenue until the customer falters.

Then the equity gains. The $5.9 billion of net gains over the nine months to March 31, 2026 is mostly a non-cash dilution mark (Form 10-Q, Q3 FY2026). The skeptic is right that this is not cash and reverses on a down round. Treating it as evidence the deal is “working” confuses an accounting echo for performance, and the Q1 FY2026 $4.1 billion loss (Form 10-Q, ended Sept 30, 2025) shows how fast the sign can flip.

Finally, the terms. The April 2026 amendment removed exclusivity and capped the revenue share at a reported $38 billion. The bear reading is that both changes moved against Microsoft: OpenAI can now diversify away from Azure, and Microsoft’s claim on OpenAI’s growth is now bounded. A partner that negotiates away your exclusivity and caps your upside is a partner with the upper hand, and bargaining power tends to compound.

The honest weighing: the bear case is right about the mechanics and mostly about direction, but it overshoots on certainty. The recirculation does flatter optics, yet Azure demand from outside OpenAI is real, so the credits are not pure round-tripping. The equity gains are non-cash, yet a 27% stake in a leading AI lab is a genuine option with real value if any reasonable scenario plays out. The terms did move toward OpenAI, yet a guaranteed $250 billion floor and a capped, known payout are also a de-risking of Microsoft’s downside. What the skeptics get right is that the deal is more contingent and more counterparty-dependent than the headline numbers suggest. What they overstate is that contingency equals weakness. It is exposure, priced into a structure that still protects Microsoft’s floor.

---

Where this is vulnerable: diversification and the cap

A credible read names where it breaks. The Microsoft OpenAI deal economics are vulnerable on two specific seams: OpenAI’s new freedom to diversify cloud providers, and Microsoft’s now-capped claim on OpenAI’s growth.

The diversification seam is on OpenAI’s side. With exclusivity gone, OpenAI can route incremental workloads to other clouds and raise from new capital sources at higher valuations. Each of those choices weakens a different part of Microsoft’s position. Workloads elsewhere shrink the share of OpenAI’s spend that recirculates as Azure revenue. New outside capital dilutes Microsoft’s 27% over time and can reset the valuation that the stake’s carrying value depends on. The more independent OpenAI becomes, the less the circular model favors Microsoft specifically.

The cap seam is on Microsoft’s side. The reported $38 billion ceiling means that past a point, Microsoft stops capturing OpenAI’s revenue growth through the share channel. If OpenAI grows far beyond expectations, Microsoft’s participation is increasingly limited to equity it cannot easily monetize and to whatever Azure spend OpenAI still chooses to direct its way. The cap protects Microsoft from paying an unbounded amount in the bilateral arrangement, but it also fences in the upside on the inflow side.

There is also a concentration seam that sits underneath both. A large share of the deal’s value to Microsoft rides on one counterparty’s continued spend and continued fundraising success. That is the single-name version of the same filing-level risk every analyst should already check, the customer-concentration risk that turns a strength into a fragility the moment the counterparty wobbles. The public filings do not isolate OpenAI’s contribution to Azure revenue, so the exact concentration is not disclosable from this vantage. Directionally, the dependency is real and now less locked in than it was before October 2025.

---

What operators should take from this

The deal does not tell you whether to admire Microsoft or OpenAI. It tells you how to structure a capital-intensive relationship so that most of your “investment” recirculates and your downside is floored. Here is how to act on that if you build software rather than fund AI labs.

1. Separate cash out from credits recirculated in every deal you model. When you fund a partner or customer in your own product credits, that is not the same as cash leaving. Model the two paths separately, and never let a recirculating-credit commitment masquerade as a clean investment in your own analysis.
2. Floor your downside before you chase the upside. Microsoft’s $250 billion commitment is a guaranteed revenue floor first and a growth bet second. When you sign a large counterparty, negotiate the minimum that protects you before you price the optionality, because the floor is what survives a bad scenario.
3. Treat non-cash marks as marks, not money. Equity-method dilution gains feel like wins and reverse like losses. If your business carries equity stakes or convertible positions, separate the cash you can spend from the carrying value that only swings, and report them to yourself that way.
4. Watch what exclusivity costs you, both ways. Microsoft traded exclusivity for a floor and a cleaner equity position. Exclusivity is a bargaining chip you can spend once. Decide deliberately whether you are buying lock-in or selling it, and price the trade instead of backing into it.
5. Stress-test single-counterparty dependence. A loop that compounds in your favor compounds against you if the counterparty diversifies or stalls. Quantify how much of your revenue or asset value depends on one partner continuing to spend, and cap that exposure before it caps you.

As a small, illustrative analog from running a product like PDF9to5: when a single integration partner drives a large share of usage, the difference between a handshake and a contracted minimum-commitment floor is the difference between revenue you can forecast and revenue that can vanish in a quarter. The structure Microsoft used at $250 billion scale is the same structure a solo operator should want at the scale of one large account: a floor that survives the partner changing its mind.

---

Analysis, not investment advice. Filing-sourced figures are drawn from Microsoft’s public SEC filings, cited inline by fiscal period (Microsoft Forms 10-Q for Q1 FY2026 and Q3 FY2026; Form 8-K, FY2025). Deal terms such as the ~27% stake, the ~$135B implied value, the ~$250B Azure commitment, the ~$38B revenue-share cap, and the cash-versus-credit split are reported figures from announcements and press, not disclosure lines, and are labeled as reported throughout. Frameworks here, including the Microsoft-OpenAI Deal Map, are for understanding business structure and tradeoffs, not for making buy or sell decisions.

Want the full toolkit for reading deals like this, the cash-vs-credits worksheet, the equity-method swing model, and the Microsoft-OpenAI Deal Map used above? It’s in the Tech Business Analysis Playbook.