Palantir's reliance on pay-as-you-go service with refunds makes customer cancellations and sudden earnings declines or write-downs a serious risk as AI and high SaaS prices pressure the business. Its 10-Q warns of these risks. Anthropic can and will easily replace Palantir with an all-in-one platform under Democratic control of the legislative branches. x.com  x.com 

Palantir looks exposed as AI deployment gets cheaper and more crowded, because outside government contracts it lacks domain expertise and controls no meaningful infrastructure software, relying on others such as Databricks. Amazon and Microsoft are launching forward-deployed engineering and frontier AI groups backed by deep enterprise relationships, and open-source models gaining traction should bring costs down. IT services firms such as Infosys could do really well instead. x.com  x.com 

Enterprise demand for sovereign AI favors Palantir's transformation model. In a two-pronged AI economy, businesses can retain their IP and use low-cost open models while frontier labs compete for the last increment of intelligence. x.com 

Palantir's long-term moat is hard to assert in a heavily computational space. Clients have an incentive to remove inefficiencies, and it will become increasingly easy for them to drop its value proposition and build their own, so there are better picks. x.com 

Palantir's valuation premium is shrinking because it is no longer the only player in defense AI. Its 2024-25 run at hundreds of times earnings rested on being first and only there, but frontier labs it once partnered with now sign directly with the Pentagon, and OpenAI and Anthropic are moving into forward-deployed engineering for enterprises. CEO Alex Karp's attack on the labs reflects that pressure. Opportunity returns once revenue and profit scale up and the valuation is digested. x.com 

Palantir's ontology can make AI work verifiable beyond coding, helping enterprises adopt it more broadly. Its map of an enterprise's objects, relationships and rules could make even an average Nemotron model state of the art for that enterprise's own work. If results are good enough, firms will use them for sovereign AI; enterprises will not want to blindly run closed frontier models on codified context and give their know-how away. x.com 

Palantir's multiple is too high to pay, so the stock is likely to be a source of funds for two-plus years. That holds even if @DivesTech is right that the narrative of Anthropic eating Palantir's lunch is wrong and Palantir sits at the epicenter of AI leaders. x.com 

Palantir is a worse business than CrowdStrike, Cloudflare and Datadog and is at the tail end of its acceleration cycle, while the cybersecurity supercycle may just be starting. So its fall to half CrowdStrike's and Snowflake's EV-to-operating-income multiple, noted by @CoorsLightCEO, makes sense once anchoring is removed. All of them are overvalued, though. x.com 

Palantir's Army NGC2 win makes it infrastructure rather than a vendor, and the stock isn't priced like infrastructure yet. The Army set Palantir's Foundry as the cloud data layer of NGC2, its highest-priority modernization program, with Anduril's Lattice as the tactical layer, so every future app, AI weapon system and command decision runs through Foundry. x.com 

ServiceNow is cheap because the market still doubts enterprises will pay to control their AI stack, while Palantir is expensive because the market already believes it. Alex Karp said technical customers want control over their compute, models, data and alpha. Renting intelligence by the token hands that alpha to model labs, which ship the verticals built on them, as Figma learned from Anthropic's Claude Design. Control means owned compute, open weights, private data and neutral orchestration. … Show more