Microsoft Chief Executive Satya Nadella just gave Palantir bulls a fresh talking point and AI investors a new headache. In a blog post on Sn Scratchpad, Nadella warned that companies may end up paying for artificial intelligence twice — first in cash, then again in the proprietary knowledge they have to reveal to make the software useful. The warning lands days after Palantir CEO Alex Karp made a similar case on CNBC, arguing that enterprise customers are already frustrated by AI tools that add little value while pulling out the data that makes businesses run.
The overlap matters because it pushes a fast-growing market debate into clearer focus: who actually captures the value from enterprise AI, and who gives it away? If Nadella is right, the hidden cost of AI is not just model fees or chip bills. It is the internal know-how, workflows and corrections that make systems smarter over time. That framing echoes Karp’s long-running argument that companies need a layer between their data and the model providers, a pitch Palantir has built around its Ontology product.
Nadella described the problem as the “reverse information paradox,” a twist on the idea that buyers often give up knowledge to obtain the tools they want. He wrote, “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.”
He added that “Models learn from ‘exhaust,’ the prompts people write, the tools agents use, and especially the corrections people make when the model is wrong. Every correction is distilled into institutional know-how.” In other words, the more useful the system becomes, the more it may depend on employee prompts, operational details and corrections that live inside the business itself.
That is exactly the kind of concern Karp has been pressing. On CNBC’s “Squawk Box,” the Palantir chief said, “I am paying for tokens that create no value. These people are stealing the weights and alpha of my business.” He also criticized the pricing model directly: “If I can make you $1 billion tomorrow, wouldn’t I say I’ll make you $1 billion, and I want 30%? Why are they charging for tokens if it’s so valuable?”
For Palantir investors, the irony is obvious. Nadella’s warning strengthens the case for the exact problem Palantir says it solves. The company has positioned Ontology as an application layer that connects models to company operations while controlling what those models can access and retain. Karp has said it makes AI “safe and useful and precise,” limiting the risk that a model caches customer data, copies a business process or exposes sensitive intellectual property.
That message has become more relevant as enterprise customers weigh how much internal information they are willing to share with AI systems. The tension is not just about privacy. It is about leverage. If companies are handing over proprietary knowledge to get better answers, the supplier of the AI tool may be absorbing some of the very advantage the customer expected to keep. Nadella put it this way: “In consuming intelligence, you are creating intelligence, and what you create should belong to you.”
Palantir’s challenge is that the market already prices in a great deal of confidence. The stock is down about 27% over the past six months and more than 26% year-to-date, according to Seeking Alpha data, but it still trades at 88 times non-GAAP forward earnings. That compares with a sector median of around 25 times. So while the company is benefiting from a stronger strategic narrative, the valuation leaves little room for error.
Nadella’s comments also arrive at a delicate moment for the wider AI trade. Wall Street still has not answered the most important question in the sector: who earns enough money to justify the extraordinary spending? Reuters has reported that Amazon, Microsoft, Alphabet and Meta are projected to spend about $630 billion on data centers and AI chips in 2026 alone, more than 4 times their 2023 guidance. That level of capital outlay has helped power the market’s obsession with AI infrastructure, chips and cloud capacity.
At the same time, sentiment is getting more cautious. Reuters also reported that 45% of fund managers in Bank of America’s latest survey view an AI bubble as the market’s biggest tail risk. The warning lights are not limited to one corner of the market. Ray Dalio has said AI is “now in the early stages of a bubble,” Jeremy Grantham says “sooner or later, the bubble will burst,” and Michael Burry has called semiconductor valuations “a pure form of overvaluation” while warning the “end may be near.”
That does not mean the AI boom is over. It does mean the debate is shifting from adoption to economics. The market can handle a lot of spending if the returns are visible. It becomes harder to defend when the returns are diffuse and the costs include handing over the knowledge that makes a company competitive in the first place. Nadella’s framing gives skeptical investors a cleaner way to ask whether the value is staying with the buyer or leaking to the provider.
The argument is also moving beyond Palantir and Microsoft. Former White House AI adviser David Sacks publicly backed Karp, pointing to Anthropic’s expansion into customer verticals such as Claude Design, Claude Code and Claude Legal. Separately, Apple filed a lawsuit against OpenAI on July 10, 2026, alleging trade secret theft by former employees now at OpenAI. Those developments are different disputes, but they point in the same direction: the market is getting more sensitive to who controls the data, the workflows and the intellectual property sitting inside AI systems.
That is why Nadella’s warning is so important for investors. It does not read like a sales pitch, but it reinforces a central investment theme: enterprises may not want raw exposure to large language models if the tradeoff is surrendering institutional knowledge. If that caution spreads, demand could tilt toward private, model-agnostic systems and other architectures that keep more control inside the customer’s walls. That would support Palantir’s strategy, but it could also pressure some of the biggest names in closed-model AI.
For now, the market has a simpler read. Nadella has echoed Karp’s warning in more measured language, and that makes the concern harder to dismiss. Palantir still has to prove that Ontology can translate the strategic fear around data exposure into durable contracts, wider margins and measurable customer returns. Until it does, the stock’s premium valuation will remain vulnerable to the same question Nadella just raised: if enterprises are paying to make AI smarter, how much of that intelligence is really theirs?
