Microsoft’s AI revenue still leans heavily on OpenAI

Microsoft booked $24.1 billion from OpenAI in the year to June, according to a filing reported by Bloomberg. That figure suggests the ChatGPT maker accounts for more than half of Microsoft’s actual AI sales — and possibly about 70% of the unit.

The disclosure offers the clearest view yet of Microsoft’s dependence on its most important AI partner. It also complicates the company’s broader AI growth story: Microsoft has invested heavily in its own models, added more than 11,000 models to Azure, and backed alternatives such as Anthropic, but most of the revenue still appears to trace back to OpenAI.

How Microsoft’s 70% estimate is calculated

The 70% figure is an estimate, not a number Microsoft reported directly. The company has disclosed its total AI business only twice and did not update that figure in its latest earnings report.

Recommended reading

Moss raises €30m to build controllable finance AI

Bloomberg assumed Microsoft’s AI business continued growing at the 123% rate reported in March. That would put annual AI revenue at roughly $34 billion. Against that estimate, the $24.1 billion attributed to OpenAI represents about two-thirds of the business — roughly 70% depending on the precise total used.

The filing does not establish that Microsoft sells $24.1 billion worth of standalone AI products to OpenAI. Under their agreement, OpenAI pays Microsoft for Azure computing, model-development costs, and a share of its revenue. Much of the disclosed amount is therefore OpenAI’s own compute bill, routed through Microsoft’s cloud and recorded as Microsoft income.

Measured against Microsoft’s total revenue, OpenAI’s contribution is less than 10%, or closer to 7%. The concentration becomes stark only when Microsoft’s AI business is considered on its own.

Microsoft is also steering its own engineers toward OpenAI

A staff memo provided another sign of that dependence. Jay Parikh, who leads Microsoft’s CoreAI engineering group, told employees to default to OpenAI’s flagship GPT-5.6 Sol in GitHub Copilot and use it most of the time.

The stated reason was financial efficiency: Microsoft wants to extract more value from the tokens it has already committed to buying.

“Shifting more workloads to OpenAI models helps us get greater value from our token investment.”

Microsoft holds intellectual-property rights to OpenAI’s models through 2032, so directing internal workloads toward Sol could help it make better use of that arrangement. The instruction also arrives as developers pull back from the period of “tokenmaxxing,” when AI workloads were expanded without as much attention to cost, while cheaper open-weight models — including many from China — put pressure on spending.

Diversification has not changed the revenue mix

Microsoft’s diversification efforts are substantial on paper. It has developed its own MAI models, offers customers more than 11,000 models through its cloud, and made a $5 billion investment in Anthropic. GitHub Copilot, which now has more than 50 million users, supports models from Anthropic, Google, Moonshot, xAI, and Microsoft itself, in addition to OpenAI.

Those products broaden Microsoft’s model lineup, but they have not yet altered the concentration revealed by the filing. The company’s AI revenue remains overwhelmingly tied to OpenAI, at least under the estimate reported here.

That dependence matters alongside Microsoft’s evolving partnership with OpenAI. The two companies have loosened limits on where OpenAI can sell, while Microsoft’s early stake has become a $228.3 billion paper gain according to a leaked cap table. Microsoft and OpenAI have also reportedly capped partnership revenue-sharing payments at $38 billion. Those arrangements make OpenAI valuable to Microsoft, but they also leave Microsoft exposed to the financial health and computing demands of a single customer and partner.

KeyBanc analyst Jackson Ader said the quality of the revenue depends on how much comes from services sold to OpenAI rather than from the economic benefit of Microsoft’s investment.

“The more of that revenue comes from services to OpenAI, the more favorably I’m going to look at it.”

Ed Zitron takes the opposite view, arguing that Microsoft has spent more than $260 billion on capital projects since 2022, while much of its AI revenue may simply be OpenAI’s compute bill recorded as growth. Under that interpretation, an OpenAI setback could leave Microsoft with both the infrastructure spending and underused data centers.

Microsoft’s filing does not settle the central accounting question: it does not break out how much of the $24.1 billion reflects Azure services, model-development costs, revenue sharing, or other activity. The company also did not provide an updated total AI-revenue figure in its latest earnings report, so the 70% estimate depends on Bloomberg’s assumption that the earlier 123% growth rate continued.

The facts point to a business that is growing quickly but is not yet diversified in the way Microsoft’s product lineup suggests. Until Microsoft’s own models and third-party alternatives materially change the revenue mix, its AI success is still, in financial terms, largely an OpenAI success.