Palantir chief executive Alex Karp has used the weeks since the company’s Q2 earnings to escalate a sovereign AI argument that directly challenges Microsoft’s cloud-first architecture, even as Wall Street continues to price both stocks as a single AI software trade. On August 12, Palantir and Microsoft fell together while chip and networking names rallied, a correlation that masks a fundamental divergence in how each company expects enterprises to pay for intelligence.
The sovereign AI market size
Karp is targeting a total addressable market McKinsey estimates between $100 billion and $160 billion in 2026, growing at a 35 percent compound annual rate to roughly $600 billion by 2030. The consulting firm defines sovereign AI as systems that maintain independence across data, technology, operations and legal frameworks, a definition that requires stitching energy, compute, models and applications into a single stack rather than renting pieces from vendors. Karp has made that integration the center of Palantir’s pitch.
Karp turns Nadella warning into sales pitch
The argument sharpened on the quarterly call when Karp warned that enterprises leasing their full AI stack from a third party risk surrendering the proprietary alpha that makes them competitive. He was responding to Microsoft chief executive Satya Nadella’s July essay, “The Reverse Information Paradox,” which acknowledged that companies “pay twice” for AI: once in cash, once in the know-how they hand over to make models useful. Karp took Nadella’s admission and extended it, suggesting the second payment could eventually train a competitor. Palantir reinforced the message in July with a nine-point sovereignty manifesto and a white paper titled “Institutional Sovereignty in the Age of AI.”
Microsoft model contradicts its own CEO essay
Microsoft’s AI revenue still flows from the model Karp attacks. Azure sells compute, Copilot sells seats, and OpenAI’s models sell tokens; the data that fine-tunes those systems flows back toward Redmond and its partners. Nadella’s essay proposed a “trust boundary” to guard against extraction, but the commercial engine remains unchanged. The contradiction is not theoretical, it is the difference between a vendor that monetizes the stack and one that sells the tools to keep the stack inside the firewall.
Stocks still trade as one basket
The market has not yet separated the bets. As long as Palantir and Microsoft move in lockstep on macro risk-off days, the sovereign AI thesis remains a narrative overlay rather than a pricing signal. The split will arrive when enterprise budgets reflect a choice between renting intelligence and owning the infrastructure that produces it.