SHEYU.AI舍予基业
AI Daily Brief

Microsoft's AI revenue leans on OpenAI, and what it means for smaller firms

Microsoft's AI income is largely compute sales and revenue share from OpenAI, a cost structure that shapes what smaller firms actually get when they buy AI tools.

·4 min read
Microsoft's AI revenue leans on OpenAI, and what it means for smaller firms
AI-generated illustration, not a news photograph

Microsoft has for the first time set out in a financial filing who its AI business actually earns from. According to Cailianshe, Microsoft booked $24.1 billion in revenue from OpenAI in the fiscal year ending June 2026; on the assumption that its AI business held its earlier growth rate, one institution estimates that this accounts for more than half of its AI sales, and possibly close to 70%.

That is not just gossip about a giant. It is a reminder for smaller companies: who stands behind the AI you buy, and how they make their money, directly affects what it costs you and how reliably it keeps working.

The big platforms earn from compute and revenue share, not from saving you steps

According to Cailianshe, the agreement between Microsoft and OpenAI requires OpenAI to pay Microsoft for the use of Azure cloud computing and inference capacity and for development costs, and to pay Microsoft a share of its revenue. A Microsoft spokesperson said the company's total AI revenue includes all sales and revenue share from OpenAI.

In other words, the bulk of Microsoft's AI business is selling compute to the companies that build models, plus taking a slice of those model companies' revenue. That is a different business from a small company buying an AI tool in the hope of hiring one fewer person or making one fewer wrong order.

Why does this matter to teams going global and to small business owners?

Because when the money upstream comes mainly from compute rental and equity ties, the pricing, free allowances and feature boundaries of downstream products follow that upstream cost structure. Free today does not mean free next year; the capability a product leads with today may be redirected tomorrow to serve larger customers. Betting your operating processes on a cost structure you can neither see nor influence carries hidden risk.

Some will say this is overthinking it: just compare features and price. That is half right. Tools should be judged on features and price. But it misses one thing: whether the features will last.

If an AI product earns mainly by selling compute to model companies, its small-business features are naturally a lower priority. The feature list can be long while updates are slow; the free allowance can be generous and withdrawn at will. What you see as "complete" and what you can rely on over time are not the same thing.

When does this judgment not hold? If a vendor's small-business arm already sustains itself with a clear base of paying customers, the upstream structure matters far less. Or if what you use is a light, replaceable tool with low switching costs, there is no need for undue worry.

Judge AI by whether it reliably removes one step

What small business owners should really do is shift AI selection from comparing features to asking whether a tool can reliably save them one step.

Shu Yi, executive director of Sheyu Jiye, puts it this way: whether software is well chosen depends not on how long the feature list is, but on whether it lets frontline staff do one step less. At this moment, that carries more weight than usual.

Take a concrete case. For companies doing government procurement and bidding, the most repetitive and error-prone daily task is trawling announcements for tenders that concern them. Miss one and you may lose an order. This is exactly the kind of work an AI subscription tool can take on reliably: SHEYU AIBID (舍予AI投标官) aggregates national government procurement and bidding information, lets users subscribe by keyword, region and industry, and pushes new matches. What it cannot do is bid for you or make your judgments for you — relationships beyond the announcement, qualifications and pricing strategy remain human work.

That is the boundary of "one step less": AI puts the numbers you should look at in front of you, and the decision stays with the owner. Shu Yi puts it more bluntly: AI will not make decisions for the boss; it puts the numbers the boss should see in front of him first.

What the $24.1 billion really says

Back to Microsoft's $24.1 billion. What it tells smaller companies is not that AI is very profitable, but that AI money is being reallocated.

Upstream earns from compute, revenue share and equity ties; downstream earns from whoever can save frontline staff that one step. You have no hand in the former; you live the latter every day.

So the thing to do now is not to study which large model has been upgraded again, but to find the task in your company that is repeated every day and goes wrong every day. Shu Yi's advice is to start digitalisation there. Find it, then see whether an AI tool can reliably catch it for you — and write down clearly who still has to handle the part it cannot.

Once that account is clear, what you are buying is not a feature list but a position from which you can reliably do one step less.

閱讀繁體中文版 →