Two-Thirds of Microsoft's AI Revenue Comes From OpenAI: $24.1 Billion

The Line That Appeared in the Annual Report for the First Time

Microsoft's annual report, published after the close on August 5, carried a sentence never seen in a prior filing.

"For fiscal year 2026, we recorded revenue from commercial arrangements with OpenAI of $24.1 billion."

Microsoft reported $331.8 billion of revenue that fiscal year, so $24.1 billion is 7.3% of it.

But the story told to investors for months is not total revenue. It is AI revenue, and there the ratio differs.

Microsoft said in the March quarter that AI was running at "an annualized revenue run rate exceeding $37 billion"; $24.1 billion of that is roughly 65%.

Bloomberg, using a $34 billion base, puts the share closer to 70%.

Both arrive in the same neighborhood: between two-thirds and four-fifths of Microsoft's AI revenue comes from a single customer — one Microsoft owns a piece of.

How the Two Companies Became This Entangled

Microsoft began investing in OpenAI in 2019, much of it in Azure compute credits rather than cash.

Microsoft put money in, it was spent on Azure, and Azure booked it as revenue.

On October 28, 2025 OpenAI converted into a for-profit entity, Microsoft emerged as largest corporate shareholder with roughly 27%, and OpenAI committed to $250 billion of incremental Azure spending.

In April 2026 exclusivity was loosened and the revenue share capped.

An investor pricing Microsoft as a cloud company is pricing the cash burn of one AI lab.

A Third of the Order Book

At the end of June, commercial remaining performance obligations were $678 billion, and CFO Amy Hood said the figure "increased 25% when excluding OpenAI."

Investors reverse-engineered the rest: if the non-OpenAI book grew 25% to roughly $460 billion, the gap is OpenAI's — about $218 billion.

Close to one contracted dollar in three.

Strategic investment cycling back to the supplier is routine in cloud. But once the loop closes: how much of this revenue was paid with money OpenAI earned?

The Mechanism: One Customer Feeding Both the Top Line and the Loss

Microsoft is not merely OpenAI's vendor but a part-owner, and that roughly 27% stake is equity-method accounted: OpenAI's losses come off Microsoft's net income.

In fiscal 2026's first quarter Microsoft recorded a $3.1 billion reduction in net income from the OpenAI investment — implying a quarterly loss near $11.5 billion.

OpenAI's own forecast projects about $14 billion of losses for 2026; the two do not reconcile, and the gap likely reflects one-off restructuring items.

The lesson is accounting, not morality: revenue growth and cash generation differ, and when a customer is also an equity investee one transaction shows up twice, pointing opposite ways. Background: how to read a balance sheet

There is a third line item: capital expenditure.

Microsoft has guided to $190 billion of capex for calendar 2026, taking the total since 2022 to $261.3 billion; lease commitments not yet commenced stand at $329.1 billion.

Much of that capacity serves the $218 billion of commitments delivered over years.

Once a data center is built, depreciation runs — not because capacity gets consumed but because the calendar advances.

What Asia Did Overnight

The disclosure landed after the U.S. close, so Asia paid the first bill: on August 6 the Kospi fell 302.82 points, or 4.59%, to 6,295.44.

Microsoft was not the only cause: SanDisk and Western Digital also reported after the close, feeding doubts about the memory cycle's peak.

A memory maker's order book depends on hyperscaler capex, which depends on cloud contracts.

If the customer at the far end loses $14 billion a year, the whole chain depends on it finding the next round of capital.

What Each Side Says

The analyst split is as clean: Tigress Financial raised its target to $690 and Goldman Sachs added the stock at $640, while estimates near $450 sit at the other end.

The gap is not a valuation argument. It is the same footnote, read two ways. Valuation

What Remains

While a dependency is merely "probably large," everyone assumes their own number; once disclosed, everyone argues the same one.

The practical question: what happens if OpenAI cannot raise its next round?

The contract stays on paper, the data center stays where it is, and depreciation keeps running.

Concentration risk is not about whether a partner is good or bad. It is the oldest subject in risk management: cash flow tied to one counterparty is undiversified.

This piece draws on the disclosures in Microsoft's fiscal 2026 annual report, Bloomberg's derivation of the AI revenue share from those disclosures, the company's quarterly earnings releases, and closing data from the August 6 Asian session. Figures for OpenAI's losses are estimates reverse-engineered from Microsoft's accounting rather than disclosed by OpenAI itself, and they vary across sources; they are presented here as a range.

Correction (September 13, 2026): prose tightened for rhythm; figures and sources unchanged.