Satya Nadella says you pay for AI twice, and the second bill is your own know-how

In mid-July 2026, Microsoft CEO Satya Nadella wrote and posted a piece called "The Reverse Information Paradox." He put it on his personal blog and shared it on LinkedIn, with a link circulated on X. This was not an off-hand remark at a conference. He sat down and made the argument on purpose, which is what makes it worth a leader's attention.

The core idea is simple enough to explain to your board in one sentence. When your company uses AI, Nadella writes, "you essentially pay for intelligence twice, once with money, and again with something even more valuable: the proprietary knowledge you must reveal to make that intelligence useful."

The first bill shows up on an invoice. The second one is invisible, and it is the one most companies never think to read.

The second bill is the knowledge you hand over to get good answers

Here is the mechanism Nadella describes. AI models get better by learning from what he calls "exhaust," which he defines as "the prompts people write, the tools agents use, and especially the corrections people make when the model is wrong."

Sit with that last part. Every time someone on your team fixes a model's answer, correcting how it priced a job or how it handled a client, they are teaching it something specific about how your business actually works. Nadella calls this "the kind of knowledge a competitor could never buy." And the better you want the tool to perform, the more of it you have to feed in. As he puts it, "the better you want the model to perform, the more of that knowledge you have to feed it!"

So the tool improves. The question is who owns the improvement.

Why he calls it "reverse"

Nadella opens the essay by invoking Kenneth Arrow, the Nobel Prize-winning economist who described the original information paradox decades ago. Arrow's puzzle was about buying information: you cannot know what a piece of information is worth until you have it, but once you have it, you have effectively gotten it for free. In Nadella's paraphrase, information's "value for the purchaser is not known until he has the information, but then he has in effect acquired it without cost."

AI flips that, he argues. In the old paradox, the buyer got the value without paying. In the reverse version, the buyer pays and pays again, because the act of using the intelligence generates new intelligence out of your own operations.

"In consuming intelligence, you are creating intelligence. And what you create should belong to you."

That line is the whole point. He is not telling companies to avoid AI. He is telling them to notice what they are producing while they use it, and to make sure it stays theirs.

The fix he proposes: five decisions, not a data-science project

The reassuring part of the essay for a leader who does not write code is that Nadella's answer is a set of business decisions, not a technical one. He organizes it around five principles, the "Five C's":

• Control. Build your own private evaluations and keep ownership of your organization's memory, its traces, its feedback, its decisions, and its context. The record of how your company works stays inside your walls.
• Capability. Build proprietary learning environments inside your own tenant boundary, so the improvement loop runs on your ground, not a vendor's.
• Choice. Keep the orchestration layer, the part that routes work, separate from any single model. Do not weld your operation to one provider.
• Cost. Combine your context, the models, and the tasks in the most cost-effective way, rather than paying premium rates for everything.
• Compound. Make sure your human capital and token capital compound inside the trust boundary, so the value builds up on your side rather than a vendor's.

His conclusion is that "a company should be able to use a model without giving up the knowledge that makes it unique." For that to happen, he says, "enterprises need a real trust boundary for their human capital and token capital to compound."

Strip the phrasing down and it is an ownership argument. Use the tools. Keep the know-how.

What this means for a leader who feels behind on AI

If you have been worried mostly about the price tag on AI subscriptions, this essay recasts the whole question, and in a useful direction. The question was never only which tool to buy. It is also how to use these tools without training someone else's product on the way your business runs.

You do not need a data science team to act on that. You need to make ownership a deliberate choice instead of a setting you surrender by default. That is a judgment call, and judgment is the thing that separates the levels in The 7 Levels of AI Proficiency. Early on, most people and most companies treat an AI tool as a magic box: type a question, take the answer. Higher up the climb, the question changes. You start asking what the tool is doing with your inputs, where your corrections are going, and what compounds inside your own walls versus someone else's.

That is exactly the move Nadella is describing. Control over your evaluations, your memory, and your context is the difference between renting intelligence and building a business asset while you rent it.

For an owner, the practical version looks like a few questions you can raise this quarter without hiring anyone:

• Where does the record of how our team uses these tools actually live, and can we get it back?
• If we corrected a model a thousand times this year, do we own those corrections, or did they improve a product we do not control?
• Are we locked into one vendor, or can we move our workflow if the terms change?

None of those require a keyboard. They require an owner who has decided that the knowledge the company generates belongs to the company.

A practical next step: pick one AI tool your team already uses every day. Before you renew it or roll out another, ask those three ownership questions about that one tool: where the record lives, who owns the corrections, and whether you could move if you had to. If you cannot answer all three, that is where your attention goes first. The tools will keep getting cheaper. The know-how your team pours into them is the part worth protecting.

Related reading: Level 5: The Captain (Design Thinker).

Sources

1. The Reverse Information Paradox - Satya Nadella, SN scratchpad
2. Satya Nadella on X - post linking to the essay, July 12, 2026
3. Satya Nadella on LinkedIn - cross-post of the essay
4. Satya Nadella flags AI's 'reverse information paradox' risk for enterprises - Business Standard

Harrison Painter

Executive AI Advisor. Founder, LaunchReady.ai and AI Law Tracker.

Harrison is an Indiana AI Advisor who helps business owners and executives get their time back by building AI systems that run the work for them. Nearly 20 years in business and author of You Have Already Been Replaced by AI. Creator of The 7 Levels of AI Proficiency.

Connect on LinkedIn