SHEYU.AI舍予基业
AI Daily Brief

Microsoft's $24.1bn OpenAI revenue shows why SMEs must map their AI dependence

Microsoft disclosed $24.1bn in revenue from OpenAI for the year to June 2026. For small firms, the lesson is to separate replaceable AI tools from business data that is hard to move.

·5 min read
Microsoft's $24.1bn OpenAI revenue shows why SMEs must map their AI dependence
AI-generated illustration, not a news photograph

Microsoft has for the first time disclosed how much money it makes from OpenAI. In its annual filing with the US Securities and Exchange Commission, the company reported $24.1bn in revenue from OpenAI for the fiscal year ending June 2026. According to Cailianshe, analysts extrapolating from Microsoft's earlier AI growth rates estimate this accounts for more than half — possibly close to 70% — of its AI sales.

For small and medium-sized business owners, this is not a spectator story. It shows that the AI tools you use today may depend heavily on a single model company. Whether that supplier is profitable, and how deeply the two are tied, eventually feeds through to your renewal prices, service stability and where your data sits.

Our view: SMEs do not need to build their own models, but they must start asking which layer of AI has a hold on their business.

Why Microsoft put this number on the record

According to Cailianshe, the $24.1bn figure did not come from Microsoft's fourth-quarter earnings release on 29 July, but from the 10-K annual regulatory filing submitted to the SEC the same day. It is the first time Microsoft has published commercial revenue from OpenAI.

The move itself is worth examining. Cailianshe reported that Olga Usvyatsky, founder of data analytics firm Nonlinear Analytics, wrote that the disclosure may be connected to OpenAI's preparations for an IPO. In other words, Microsoft is telling the capital markets how much of its AI revenue is genuinely customer money and how much is revenue share from a deep tie-up with OpenAI.

The report contains another key detail: Microsoft's total AI revenue includes all sales and revenue share from OpenAI. Under their agreement, OpenAI pays Microsoft for the use of Azure cloud computing and inference capacity and for development costs, plus a percentage of revenue share. Microsoft holds about 27% of OpenAI on a diluted basis, with economic rights only and no operating voting rights.

Once that structure is clear, so is the picture: Microsoft and OpenAI are linked by equity, compute and revenue share. Money moves between the two, but the companies ultimately paying are the enterprise customers using AI services.

What this means for teams going global and for smaller firms

The first effect is on price expectations. When a cloud provider's AI revenue depends heavily on a single model partner, its pricing cannot be fully independent. A tool that is cheap today may cost more next year if upstream compute costs or revenue-share terms change. Teams going global should be wary of treating AI tool spending as a fixed cost in annual budgets.

The second effect is data and compliance. If your customer lists, order records and quotations all run on a foreign model service, cross-border data transfer and supplier lock-in will have to be faced sooner or later. For small teams doing cross-border business in particular, where customer data sits is not a technical question but a commercial one.

The third effect is more practical. Many owners buy AI tools on the strength of a feature list, rather than on whether the tool helps them do one step less. Shu Yi, executive director of Sheyu Jiye, put it plainly: the test of software is not how long its feature list is, but whether it lets frontline staff do one step less.

The same holds for AI procurement. Which company built the model, and how many parameters it has, matters little to a market-stall owner or an export salesperson. What matters is whether orders are written quickly, whether accounts reconcile clearly, and whether the month-end can be closed without two late nights.

What smaller firms should do now

This is not a call to build your own large model, which is neither realistic nor necessary. What is needed is to split "AI dependence" into two layers.

One layer is general capability: writing copy, translating emails, making images. These tools are cheap to switch, so use whichever is cheaper and better, with no need for loyalty.

The other layer is operating data: customer credit, purchasing costs, order flows. Once this is handed to a platform, migration costs rise. When choosing, ask clearly: can the data be exported? If we change systems, can the accounts come with us?

Shu Yi makes a further point: when going digital, start with the thing that repeats every day and goes wrong every day. For owners of agricultural wholesale stalls, that thing is often credit bookkeeping — who owes how much, how much has been repaid, which entry does not match — all held in memory and on scraps of paper, with profit quietly leaking away over the year.

General-purpose large models cannot help with this kind of scenario. They do not know how many truckloads left your stall today, or which regular customer has been three days late again.

One product option, and where it stops

If the problem happens to be writing orders and collecting payment at a stall, SHEYU WANGPU (舍予AI旺铺搭子) is one option to look at. It is an order-and-collection app for agricultural wholesale stalls: weight-based orders can be entered with one hand, customer credit and repayments are automatically kept as a single ledger, reconciliation statements are generated in one tap and sent via WeChat, inbound freight is allocated by weight to calculate costs, and closing stocktakes, consignment sellers and vehicle-trip settlement are all included.

It addresses the "three ledgers of a stall" — order ledger, credit ledger and purchasing cost ledger — when they are unclear or do not reconcile.

What it does not do: it does not make business decisions for the owner, and it does not handle finance and tax matters outside the stall. It is a mobile app, not a general-purpose large model; writing copy or translating is beyond it. Shu Yi has said that AI will not decide for the owner, but will put the numbers the owner should see in front of them. That sentence marks this product's boundary.

Back to Microsoft's $24.1bn. It is a reminder to every small business owner: the deeper AI goes, the more concentrated suppliers become, and the more clearly you need to know which part of your business you have handed over. General tools can follow the crowd; operating data must keep a way out.

In the AI era, real operating security is not how much AI you use, but whether your books can be balanced whoever leaves.

閱讀繁體中文版 →