Microsoft's non-exclusive OpenAI licence signals a shift SMEs should note

Microsoft's revised agreement with OpenAI does not directly hit most small and medium-sized enterprises, but it makes one thing clear: the AI capability you depend on is moving from exclusive supply to replaceable supply. The real task is not to switch vendors, but to use this window to embed AI into your own business.

What changed in the Microsoft-OpenAI deal

The trigger for this judgment is Microsoft's announcement on 27 April 2026 that it was adjusting its cooperation agreement with OpenAI. According to a Tencent News report titled "Who exactly is controlling whom in the global AI industry?", Microsoft remains OpenAI's primary cloud partner under the adjustment, and related products are in principle released on Azure first, but OpenAI can offer products to customers through other cloud service providers. Microsoft's licence to OpenAI models and product intellectual property runs until 2032 and has become non-exclusive.

Most discussion has focused on who has the upper hand, Microsoft or OpenAI. For business owners, the more useful layer is this: in recent years, many companies chose AI tools by "following the strongest model". That logic now has a shorter shelf life. Models can change clouds, clouds can host other companies' models, and contract terms are renegotiated every few years, while your business systems remain welded to one vendor's interface.

When models are replaceable, what becomes more valuable

According to the same report, the world's major AI companies have already crossed the boundaries of chips, cloud, models and applications. Microsoft runs Azure and also serves enterprise customers through Microsoft 365 and GitHub. Google owns the TPU computing system, Google Cloud, the Gemini model, and search, office software and mobile device ecosystems. Amazon has invested in Anthropic, provides computing power through AWS, and also runs a cloud marketplace and its own model business.

These companies cooperate and compete with one another. Microsoft needs competitive models, while OpenAI wants to serve enterprise customers directly. Model companies reach customers through cloud platforms while trying to retain control over product pricing and customer contracts.

When supply relationships upstream can be renegotiated, the real asset for downstream companies is no longer "which model we use" but "which parts of our business, unique to our company, the model handles once it comes in". Tender documents, quotations, customer records and compliance records all have to be fed in again whichever model you use. Whoever turns them into a reusable data library has the lowest reset cost.

The counterargument: small companies need not worry

A common view holds that agreement changes between giants are too far from a company of a few dozen people, and that it is fine to switch when the impact actually arrives.

That view holds under one condition: you use AI only as a search box, asking a couple of questions and writing some copy, in which case one vendor is as good as another. But if AI has entered your delivery process, such as bidding, quotation or filing, then "switch when it affects me" means the process breaks. Changing a model is just changing an interface; replacing a way of working that has already absorbed your company's materials means starting over.

The judgment has its limits. If your industry demands extremely high accuracy from AI output and lacks a human review step, then regardless of whether the supplier is exclusive, you should not let it produce the final version directly. Replaceability addresses the risk of being locked in; it does not address whether the output is correct.

One thing to do now

Organise the materials your company uses repeatedly and that only you have the right to use — qualifications, track record, personnel, past proposals — into a clearly structured master document. This has nothing to do with which model you use; once it is done, connecting to any of them is faster.

Take bidding as an example. SHEYU AIBID parses tender documents page by page into a list of requirements with page references, attaches real evidence from the enterprise data library to each item, and then drafts the document by volume and section. It runs an 18-rule compliance self-check plus a simulated review, and locks the export until the materials are complete. What it cannot do is judge whether a project is worth bidding for, or replace your final check on the authenticity of the evidence.

Policy is pushing in the same direction. In August 2025, the State Council issued the Opinions on Deepening the Implementation of the "AI+" Action. The direction is clear, but for each company the first steps usually come in areas where the materials are already messy and the processes already depend on people watching them.

The Microsoft-OpenAI contract may have a new version next year. The organised business materials in your hands can be revised once and used for many years.