Satya Nadella, the CEO of Microsoft, published a blog post warning companies that when they use AI systems from large tech vendors, they may be inadvertently paying twice — once with money, and once with their own private business information TechCrunch.

The warning centers on something Nadella calls model "exhaust." When you type a request into an AI system, the AI learns from that request. If a company uses an AI tool to solve problems specific to its business — pricing strategies, troubleshooting methods, how they organize workflow — that information can end up in the AI system's training data. The vendor then uses that same AI to serve other companies, potentially including your competitors, so your proprietary knowledge becomes available to them indirectly TechCrunch.

Nadella argues this arrangement is one-sided. Large AI vendors train their systems on publicly available information from the web. They do this legally under a doctrine that allows limited use of public material for research. But then they sell licenses that forbid customers from taking the AI's outputs and using them to train their own competing systems. In other words, vendors are free to learn from public sources, but customers cannot do the same with vendor models TechCrunch.

His solution has two parts. First, companies should keep ownership of all the information they feed into AI systems — their prompts, corrections, and feedback — instead of letting it disappear into a vendor's training pipeline. Second, they should build what he calls switching infrastructure on cloud platforms. This would let a company move from one AI vendor to another without rebuilding everything from scratch, which keeps any single vendor from having too much control over the relationship.

Nadella published a related piece in the Wall Street Journal with a broader message: the concentration of power among a few giant AI companies poses a risk to healthy competition in the economy WSJ.

There is something important to keep in mind here. Nadella runs Microsoft, which sells exactly the kind of cloud services and switching tools he is recommending. Microsoft also has a major business deal with OpenAI, one of the proprietary AI vendors he is implicitly criticizing. His warning is worth listening to for its technical substance, but it also serves Microsoft's own business interests. If large customers take his advice and build flexible, multi-vendor setups on Microsoft's Azure platform rather than locking into a single AI provider, Microsoft benefits.

That said, the underlying technical concern is real. Large companies that use AI heavily generate enormous amounts of interaction data — all the requests sent, corrections made, feedback given. The question of who owns that data, who is allowed to use it, and whether it secretly improves a vendor's AI system has not been clearly spelled out in most business contracts. This is a gap that legal and purchasing teams have been worried about informally for the past couple of years, even if they did not have a clear way to talk about it TechCrunch.

The approach Nadella is pushing — using flexible switching layers to avoid being locked in — is not actually new. Companies did something similar about ten years ago with cloud computing, building systems that let them use AWS, Azure, and Google Cloud without getting stuck with one vendor. Whether the same trick works as well for AI systems is less clear. Switching between AI vendors can mean losing months of work you did to tune one system's behavior to your needs, or having to rebuild tools that worked well with a particular AI's strengths and weaknesses. Nadella's infrastructure approach reduces contractual lock-in but does not solve the deeper problem of being invested in how a specific AI system works.

What Nadella has done, intentionally or not, is give companies a clear way to talk about a concern they were already raising in negotiations with AI vendors. The question now is whether this language actually changes the deals companies strike, or whether it remains just another talking point in the ongoing competition between Microsoft and the independent AI labs it partners with and competes against.