Microsoft CEO Satya Nadella warns businesses adopting artificial intelligence.
In one long post on X, Nadella calls the problem the “Reverse Information Paradox.” Companies pay to access AI, but then have to feed the system with their personal data, internal processes, employee lists, ratings, and institutional knowledge to make it useful.
"You're essentially paying for information twice, once with money and again with something even more valuable: the proprietary knowledge you need to reveal to make that information useful,” Nadella says.
It's a worrying argument, and it's probably true. But it sounds a bit excessive coming from Microsoft, one of the largest sellers of cloud computing, enterprise software, and artificial intelligence services on the planet.
The company wants businesses to move their files, employees, workflows, identities and increasingly decision-making to platforms controlled by Microsoft, and now, its CEO is warning about giving too much knowledge to AI providers.
Nadella isn't wrong, but businesses should pay close attention to who is issuing the warning and what Microsoft might gain from it.
He argues that the real risk goes beyond posting a confidential internal document. AI systems can learn from prompts, tool use, feedback, corrections, evaluations, and decisions.
“Every fix is distilled into institutional expertise,” Nadella says. “It’s the kind of knowledge a competitor could never buy, and the kind that trickles down almost imperceptibly: little by little, fix by fix, assessment by assessment.”
This is perhaps the strongest argument. Companies often treat AI prompts and conversations as expendable, but these interactions can reveal how employees solve problems, what customers want, which exceptions matter, how prices are set, and what the organization considers a successful outcome.
A long trail of prompts and corrections can reveal how the business actually works. This information can be much more valuable than the documents that are uploaded.
Nadella says that the organization that creates this knowledge should own it.
“By consuming intelligence, you create intelligence,” he says. “And what you create should belong to you.”
Logical. The problem is that the solution Nadella proposes seems suspiciously convenient for Microsoft.
He argues that companies should create private assessment systems, maintain ownership of organizational memory, create proprietary learning environments, and use a level of orchestration that is not tied to a single Artificial Intelligence model.
Microsoft “happens” to sell the cloud infrastructure, development tools, security products, databases, AI services, and orchestration technology required to build exactly this kind of environment.
The message is essentially this: Don't trust an AI provider with all your knowledge. Instead, build the infrastructure around that knowledge within Microsoft Azure.
But this is not independence. It is a different version of dependence.
Nadella also recommends that companies have the ability to switch between models.
“Ask yourself: If one of the models you use is removed, will you still be able to operate and optimize your ratings using other models?”
This is good advice, especially as AI companies change prices, retire models, rewrite terms, and introduce new restrictions without warning.
However, Microsoft's concern about model lock-in doesn't necessarily extend to cloud lock-in, software lock-in, identity lock-in, or productivity suite lock-in. Microsoft would obviously prefer that businesses remain flexible about which model they use, provided that the surrounding infrastructure remains within its ecosystem.
Nadella's post reads partly as a warning and partly as a pitch for private enterprise AI running on systems controlled by Microsoft.
Although the press releases will range from very select to rare, I said I'd pass...because sometimes the editors hide.
