Alex Karp has spent the past fortnight telling anyone who will listen that Palantir Technologies is indispensable. Retail investors and the German military have arrived at the opposite conclusion, and their timing is almost perfectly synchronised.

In the week through 13 May, individual investors were net sellers of $82 million worth of Palantir shares, according to JPMorgan data cited by analyst Arun Jain. They also dumped $117 million of Microsoft stock. The rotation is not out of technology entirely, it is out of software and into semiconductors. The VanEck Semiconductor ETF and iShares Semiconductor ETF are up 60% and 75% year to date, respectively. The Roundhill Memory ETF, which trades under the ticker DRAM and launched on 2 April, has roughly doubled since inception and attracted more than $5 billion in assets, including $1.1 billion in a single day last week. Memory chips, not enterprise AI software, are now the hottest trade in the retail crowd.

Palantir’s stock is down roughly 20% year to date. That decline has arrived despite first-quarter results that would have been the envy of any software company a year ago: revenue of $1.63 billion, up 85% year on year, with US revenue surging 104% and US commercial revenue climbing 133%. The company raised its full-year guidance to $7.65 billion, implying 71% annual growth. And yet the stock fell after earnings. The market has decided that Palantir’s growth, however impressive, does not justify a valuation of roughly 42 times implied 2026 sales. Michael Burry, the investor made famous by The Big Short, has been bearish. Short seller Andrew Left has called the valuation “absurd”.

Germany says no

The investor scepticism has a geopolitical echo. In an interview with Bild published this week, Karp said he was “surprised” by Germany’s refusal to consider Palantir for military contracts, a stance that has hardened over the past month into something close to official policy. Vice Admiral Thomas Daum, head of the Bundeswehr’s Cyber and Information Domain Service, told Handelsblatt that Palantir “is not being considered at all right now” because the military is not prepared to allow employees of a private American company to access national data. German Digital Minister Karsten Wildberger told Politico that Berlin wants a European alternative.

The Bundeswehr has shortlisted three candidates: Almato, based in Stuttgart; Orcrist, based in Berlin; and ChapsVision, based in Paris. The German domestic intelligence service, the BfV, has already chosen ChapsVision. The evaluation of the military candidates is expected this summer, with a contract award by the end of the year. The decision will shape the digital infrastructure of Germany’s armed forces for years to come, and Palantir is explicitly excluded from the process.

Karp’s response was characteristically blunt. He compared the German debate around his company to “conversations about witchcraft” and argued that Palantir’s technology was being used “on every serious battlefield in the world”. He noted that co-founder Peter Thiel was born in Germany and that he himself studied there and speaks fluent German. “Peter and I are the leading Germanic and/or German-speaking businessmen in the world, and every other country would have found a way to embrace us,” he said. “If we were French, the French would collectively force us to get French passports, speak only French, and change our name to Falantir.”

The Ukraine argument

The Bild interview followed a meeting in Kyiv with Ukrainian President Volodymyr Zelenskiy and Digital Minister Mykhailo Fedorov. Karp described Palantir’s products as “an operating system for war” and argued that Ukraine had demonstrated their value more convincingly than any sales pitch could. He said the Ukrainian military managed the battlefield “the way a technology company manages its customers,” tracking outcomes by square kilometre.

The pitch to Berlin was implicit but unmistakable: buy what has been tested in combat, not what has been tested in PowerPoint. “What products is Europe going to buy to defend itself?” Karp asked. “Products tested in PowerPoint? Or products that stopped a major military power on their own?”

Germany’s counter-argument is sovereignty. The Bundeswehr wants a secure private cloud for data processing and AI applications in which no foreign company has structural access. The concern is not that Palantir’s technology is inadequate. It is that dependence on an American provider creates a vulnerability — a point that carries more weight in 2026, with transatlantic relations under strain, than it might have a few years ago. Europe’s own defence-tech sector is growing rapidly, and alliances between European AI companies and defence startups are producing alternatives that did not exist when Palantir first began courting NATO governments.

Two problems, one company

The investor rotation and the German rejection are different problems, but they point to a shared vulnerability. Palantir has long traded on narrative as much as numbers. It was the company that took over Project Maven when Google walked away. It was the AI stock that retail traders held through every drawdown. It was the defence contractor that Karp positioned, through manifestos and conference speeches, as the indispensable provider for Western democracies at war.

Each of those narratives is now under pressure. Retail traders have found a more exciting trade in memory chips, the physical infrastructure of AI rather than the software layer. Germany, the largest economy in Europe and the country embarking on its biggest military build-up since reunification, has decided it does not need Palantir at all. And the broader market is asking whether a company trading at more than 40 times forward sales can sustain its premium when competitors, from Anthropic to open-source alternatives, are closing the gap on enterprise AI.

Karp’s frustration with Germany is understandable. His company’s software is deployed in Ukraine, integrated across the US military, and embedded in NATO operations. The argument that it has been proven in ways European alternatives have not is factually difficult to dispute. But sovereignty is not a technical argument. It is a political one, and Germany has decided that the risk of depending on an American company for the core of its military data infrastructure outweighs the convenience of buying the best-known product on the market.

Palantir’s financials remain extraordinary by any conventional measure. Eighty-five per cent revenue growth at $1.63 billion in quarterly sales is not a company in decline. But the gap between the company’s operational performance and its market reception, both from investors who are selling and a major ally who is not buying, suggests that Palantir’s problem in 2026 is not execution. It is persuasion. The numbers are working. The story, for the first time, is not.