Microsoft Booked a $5.9 Billion Gain on a Company That Loses Billions

---

Begin with the accounting, because it is the cleanest entry point into a relationship that has become one of the
most important and least understood in the market. Microsoft holds roughly 27% of OpenAI's for-profit entity,
accounted for under the equity method — meaning Microsoft records its share of OpenAI's results in its own income
statement. For most of the relationship, that line was a drag: in the quarter around October 2025, Microsoft's
share of OpenAI's enormous losses cut its net income by about $3.1 billion, exactly what you would expect from
owning a slice of a company spending far more than it earns. OpenAI loses staggering sums, and Microsoft, owning
a quarter of it, had to report a quarter of the pain.

Then the sign flipped. Over the nine months ending March 2026, Microsoft recorded $5.9 billion of net gains
from the same investment — a swing of billions, from loss to gain, on a company whose underlying economics did
not improve. What changed was not OpenAI's profitability but its valuation. When OpenAI restructured into a
public-benefit corporation in late 2025, the transaction reset the accounting, and because OpenAI's implied
valuation had soared even as Microsoft's ownership percentage fell, Microsoft was able to book the difference —
the rising value of its stake — as income. The gain is real in the sense that the stake is worth more on paper.
It is unreal in the sense that no cash changed hands, OpenAI did not become profitable, and the "earnings" depend
entirely on a private valuation that could move the other way as easily as it moved this one.

This essay is about that loop — the circular financing and circular accounting that now sit inside the most
trusted earnings in technology — and about why a company as genuinely strong as Microsoft has nonetheless woven a
meaningful, opaque, OpenAI-dependent thread through both its growth story and its profits, at the precise moment
the AI capital binge is compressing its margins to multi-year lows.

The loop, traced

To see the circularity clearly, follow a dollar around the circle, because once you have traced it you cannot
unsee how much of the AI boom's apparent vigor is the same money passing the same checkpoints.

Microsoft has committed $13 billion to OpenAI, of which about $11.8 billion had been funded by March 2026. That
money does not sit in OpenAI's bank account; OpenAI spends it, overwhelmingly, on computing power — and the
computing power it buys is Microsoft's Azure cloud. As part of its 2025 recapitalization, OpenAI committed to
spending a staggering $250 billion on Azure over time. So the sequence is: Microsoft gives OpenAI cash; OpenAI
hands much of it back to Microsoft as Azure payments; those payments are booked as Azure revenue; and Azure
revenue growth — running at a spectacular 39–40% — is the single most important number in Microsoft's bull case,
cited everywhere as proof that the AI cloud era is real and that Microsoft is winning it.

Some meaningful portion of that headline Azure growth is, therefore, Microsoft's own investment dollars cycling
back to it as revenue. This is not fraud and not even unusual in the AI economy — the same circular pattern runs
through Nvidia's relationships with the labs it both supplies and invests in, and through the broader web of
cross-investments binding the AI industry together. But it changes what the growth means. Revenue that comes
from an independent customer making an arm's-length decision to buy your product is worth more, and is more
durable, than revenue that comes from a company you funded spending your money on your product. The first is
demand. The second is, in part, recirculation. The market is pricing Azure's growth as though it were entirely
the first, when a portion of it is the second — and the portion that is recirculation depends entirely on OpenAI
continuing to raise and spend enormous sums, which depends on OpenAI's valuation continuing to rise, which is the
same valuation generating Microsoft's $5.9 billion accounting gain. The loop closes on itself.

Earnings quality, quietly diluted

The equity-method gains deserve their own scrutiny, because earnings quality is the entire reason Microsoft
commands the premium it does, and these gains are the lowest-quality earnings the company reports. When an
investor pays a high multiple for Microsoft, they are paying for the durability and predictability of its
profits — the recurring, cash-generative, enormously high-margin streams from Office, Windows, and enterprise
software. Those are the gold standard of earnings quality. The OpenAI equity-method line is the opposite: a
non-cash, mark-to-model entry that swings from a $3.1 billion loss to a multi-billion-dollar gain based not on
operations but on the shifting paper valuation of an unprofitable private company.

The danger is that the market does not distinguish between these two kinds of earnings when it looks at the
headline number. A dollar of Office profit and a dollar of OpenAI valuation gain both land in net income, but
they are not worth the same, and a valuation gain that can reverse is being counted by the market as though it
were as solid as a software subscription. Worse, the direction of the OpenAI line is now a source of earnings
volatility for a company whose entire appeal is the absence of volatility — one period it subtracts $3 billion,
another it adds $6 billion, and the swing has nothing to do with how Microsoft's actual business performed. The
most predictable earnings in technology now have an unpredictable, valuation-driven component bolted to them, and
in a downturn for AI valuations, that component flips from tailwind to headwind precisely when everything else is
also under pressure.

The capex bill is compressing the best margins in software

Underneath the OpenAI loop sits a more mundane but equally important pressure, and it is the same one documented
across this series at every hyperscaler: the staggering cost of building AI infrastructure is eating into
margins. Microsoft guided full-year 2026 capital expenditure to roughly $190 billion — up about 61% from the
prior year, and inflated further by a roughly $25 billion hit from soaring memory and component prices. That is
an almost incomprehensible sum, and it does not disappear when spent; it lands on the income statement as
depreciation for years afterward.

The effect is already visible. Microsoft's gross margin fell to 67.6% — its narrowest since 2022 — as
depreciation from the data-center build-out mounted. For most companies a 67.6% gross margin would be a triumph;
for Microsoft, whose software-driven margins have long been among the highest in the world, the compression is a
meaningful and ongoing degradation of the very profitability that justifies its valuation. And the trajectory
points the wrong way: as the $190 billion of capex is spent and depreciated, and as next year's capex likely
climbs again, the depreciation drag intensifies, pressing on margins for years. The AI build-out is not free, and
Microsoft is paying for it in the currency that matters most to its valuation — the quality and level of its
margins.

This is the quiet trap shared by Microsoft, Amazon, and Google alike: each is spending hundreds of billions to
win the AI cloud race, and each must depreciate that spending against AI revenue that, while real and growing,
remains a fraction of the capital being poured in. Microsoft is better positioned than almost anyone to absorb
it, because its underlying franchise is so strong. But "better positioned to absorb it" is not the same as "it
costs nothing," and the 67.6% gross margin is the receipt.

The other half: is the AI demand real at the seat level?

There is a second question lurking beneath the Azure-growth headline, and it concerns where the AI revenue
actually comes from. Broadly, Microsoft monetizes AI two ways: by renting raw computing capacity through Azure
(much of it, as we have seen, to OpenAI and other AI labs), and by selling AI features to its enormous base of
software customers — chiefly Copilot, the AI assistant bolted onto Office. The first is infrastructure revenue,
capital-intensive and, in OpenAI's case, partly circular. The second is the high-margin, durable,
demand-driven revenue that would truly validate the AI thesis — businesses paying recurring fees because the AI
makes their workers more productive.

The honest state of play is that the infrastructure side is unambiguously booming while the genuine product-
demand side, Copilot adoption, is real but more gradual and harder to read than the bulls imply. Enterprises are
experimenting with Copilot, some are deploying it widely, and the seat counts are growing — but the question of
whether AI assistants deliver enough measurable productivity to justify their per-seat price, durably and at
scale, is precisely the question the entire enterprise-AI industry is still answering, and the answer is not yet
a resounding, universal yes. This matters because the cleanest, highest-quality version of Microsoft's AI revenue
is the Copilot-seat version, and the degree to which the AI story rests on that versus on capital-intensive,
partly-circular infrastructure rental determines how durable and how profitable the AI growth ultimately proves.
The market has tended to treat all AI revenue as equally golden. It is not, and the mix matters enormously.

The partnership is loosening, in both directions

It is worth noting that the Microsoft-OpenAI relationship is no longer the tight, exclusive marriage it once was,
and the loosening cuts in ways that are double-edged for Microsoft. The 2025 restructuring removed Azure
exclusivity, which means OpenAI is now free to buy computing power from other providers — and it has, becoming a
cornerstone of enormous infrastructure arrangements with Oracle and others, spreading its colossal compute
spending beyond Microsoft. For Microsoft, this reduces capacity-commitment risk, which is genuinely good, but it
also means the $250 billion Azure commitment is a ceiling shared with rivals rather than an exclusive lock, and
that some of OpenAI's future growth will enrich Oracle and others instead.

Microsoft, for its part, is hedging too. It has been building its own in-house AI models and capabilities,
reducing its dependence on OpenAI's frontier models over time, and positioning itself to compete in a world where
OpenAI is a partner rather than the sole engine of its AI strategy. This mutual diversification is rational for
both parties — neither wants to be wholly dependent on the other — but it also quietly undercuts the simplest
version of the bull narrative, which was that Microsoft and OpenAI were a single, unbeatable, vertically
integrated AI champion. They are increasingly two large companies with overlapping and partly competing
interests, bound by contracts and a 27% stake but no longer joined at the hip. The relationship that produced the
loop is itself evolving, and an investor pricing Microsoft on the strength of an exclusive OpenAI alliance is
pricing a marriage that has already become an open one.

What the bulls genuinely get right

In fairness, the bull case on Microsoft is overwhelming, and a forensic piece that pretended otherwise would
forfeit its credibility — this is one of the great businesses in the history of capitalism, and the OpenAI bet
has been, by most measures, a strategic triumph. Several points deserve real weight.

First, the overwhelming majority of Microsoft's business has nothing to do with OpenAI. Office, Windows, the
enterprise software empire, the security business, LinkedIn, and the large portion of Azure driven by ordinary
enterprise cloud migration are all genuine, durable, cash-generative franchises that would thrive if OpenAI
vanished tomorrow. The OpenAI loop, for all this essay's scrutiny, is a thread in a vast tapestry, not the whole
cloth. Second, the OpenAI partnership has been a genuine masterstroke: it put Microsoft at the center of the AI
revolution, drove Azure to 40% growth, and gave it Copilot and a frontier-model relationship that competitors
would kill for. Third, the 2025 restructuring actually reduced several risks — it removed Azure exclusivity
(lowering Microsoft's capacity-commitment exposure), eliminated certain payment obligations, extended the IP
licensing through 2032, and removed the awkward AGI-termination clause, all while preserving Microsoft's roughly
27% stake and its 20% revenue share through 2030. And fourth, that 27% stake could be worth a colossal sum if
OpenAI eventually goes public at the valuations being discussed — the equity-method gains, paper though they are,
may presage a genuinely enormous realized windfall.

The honest synthesis is that Microsoft is a fortress with a loop running through one wing of it. The fortress is
real and the loop is real, and the mistake the market is making is not overvaluing Microsoft wholesale but
failing to discount the specific, lower-quality, circular, OpenAI-and-valuation-dependent component of its growth
and earnings — treating recirculated revenue as pure demand and mark-to-model gains as pure profit, while the
capex quietly erodes the margins underneath. None of that threatens Microsoft's survival or dominance. It does
mean the pristine quality the premium assumes is a little less pristine than the headline suggests.

When the loop runs in reverse

The most important question about any circular arrangement is what happens when it stops going around — and here
the symmetry that flatters Microsoft today is the same symmetry that would hurt it in a downturn. Consider the
scenario, not as a prediction but as a stress test. If AI valuations were to cool — if OpenAI's next funding
round came at a flat or lower valuation, or if the broader AI-investment fervor broke — several of Microsoft's
tailwinds would reverse at once. The equity-method line would swing from gain back toward loss as OpenAI's paper
value stopped rising. OpenAI's ability to fund its $250 billion Azure commitment would come into question, since
that commitment depends on OpenAI continuing to raise tens of billions from investors betting on the valuation.
And the portion of Azure growth that represents recirculated investment rather than independent demand would
soften just as the depreciation from $190 billion of capex reached full weight.

None of these is likely to be catastrophic for a company of Microsoft's strength, and each is survivable in
isolation. The point is that they are correlated — they would tend to arrive together, in an AI downturn,
because they are all expressions of the same underlying bet on ever-rising AI valuations and ever-growing AI
spending. The loop that adds to growth and earnings on the way up subtracts from both on the way down, and a
market that has priced only the upside of the circularity has not priced its reversibility. Microsoft built a
brilliant machine for converting AI enthusiasm into reported growth and profit. Machines that run on enthusiasm
run in both directions.

The kicker

There is a reason Microsoft has long been the stock that cautious investors buy when they want technology
exposure without drama: its earnings were boring in the best sense, a metronome of recurring software profit you
could set your watch by. The OpenAI era has changed that in a way the headline numbers obscure. Microsoft is
still a fortress, still dominant, still extraordinarily profitable — but its growth now contains a loop of its own
money coming back, its profits now contain billions of non-cash gains tethered to the paper valuation of a
company that loses money by the tens of billions, and its famous margins are being slowly compressed by the
largest capital-spending program in its history. The $5.9 billion gain is the tell: it is a number that exists
because a private valuation went up, sitting in the income statement of the most trusted company in technology,
counted by the market as though it were as solid as a Windows license. It is not. It is the most modern kind of
earnings — a bet on a valuation, dressed as a profit, in the accounts of the company least expected to make one.

Microsoft gave OpenAI the money, OpenAI gave it back as cloud revenue, the unprofitable startup's valuation
climbed, and Microsoft booked the climb as a six-billion-dollar gain — and somewhere in that elegant circle is a
question the market has decided not to ask, which is what any of it is worth if the valuation that powers the
whole loop ever stops going up — and what the most trusted earnings in technology look like on the day it does.