AI proves itself on the battlefield first, then exposes which enterprises and societies can bear the load.
Karp says off-the-shelf models won't work: they are a commodity and not precise enough for regulated work like underwriting. Borrowing a large language model and putting it on your stack fails. Only once you build a software layer like ontology, orchestrating the models in a language your enterprise understands, can you create value.
Karp says in the areas Palantir enters it can take out up to 80% of cost and improve the top line dramatically, though it depends on the use case. Speed has changed too: work that five years ago would have taken a year can now take a week.
Karp calls the Western narrative that AI will destroy humanity's jobs unfortunate. He says there will be more than enough jobs for a nation's citizens, especially those with vocational training. Technicians building batteries in America, high school educated, now do roughly what Japanese engineers do and are nearly irreplaceable.
Karp gives enterprise software two roles on a battlefield: first, make sure all the underlying things actually work, since parts of an enterprise exist only on a PowerPoint and fail in rough conditions; second, raise the level to one no one else in the world has.
Ukraine essentially started from nothing, so it never had to rediscover mid-fight that its enterprise did not work, unlike Western countries whose pieces exist on a PowerPoint. Its small team of very courageous, highly technical soldiers built proprietary things on top of Palantir's product that Palantir itself does not understand.
America and China understand different versions of making this work, and both work at scale; Karp expects that to accelerate far beyond what people believe, with Russia possibly following because they are so good at fighting. Europe's tech adoption he calls a serious, very structural problem no political leader admits.
Because AI is a low-trust environment: people have tried lots of things and much of it hasn't worked. If you deliver things that do work, you don't need a sales force. Palantir's keeps getting smaller every time Karp visits, not to save on unit economics, and sometimes he says don't talk to us.
Exporting is hard because Palantir must train people and has limited bandwidth. Building something like Project Maven into a country's architecture needs somebody with the highest level of security clearance who is also technical; most technical people do not pursue that clearance, so the resource is scarce, and training takes a while.
Hospitals, many of which Palantir powers, all have an intake problem, a shortage of doctors and nurses, and low margins, yet each processes patients differently according to its specialty. Structuring that intake flow lets them process 10 to 15 times faster than before, saving a lot of lives and leaving nurses happier.
Karp says AI pen tests: it load-bears on societies, organizations and companies, so those that can bear the load gain a huge advantage, while those pretending to bear it collapse and must start there. Over the next three years he expects market-value honesty that political structures cannot obfuscate.
“Since Alex at Palantir went public, his compounded return is 73%.”
“It's actually because it is a low trust environment in AI.”
“but there will be more than enough jobs for the citizens of your nation, especially those with vocational training.”
“Well, I think the obvious first imbalance is it seems like America and China understand versions of making this work, and they're different.”
“But the tech adoption in Europe is a serious and very, very structural problem.”
“one way to look at the unfairness of AI is it pen tests, meaning it load-bears on things.”
“Conversation with Alex Karp, CEO and Co-Founder, Palantir Technologies | WEF Annual Meeting 2026.” https://www.youtube.com/watch?v=dRgPbNJ5RJc. Transcript summary by WhipScribe, https://whipscribe.com/transcript-pages/conversation-with-alex-karp-ceo-and-co-founder-palantir-tech.html.