Palantir and Nvidia: Where AI's Value Migrates After the Chips

On October 28, 2025, at an Nvidia event in Washington, D.C., Jensen Huang and Alex Karp stood on the same stage to announce that Nvidia's compute and open models would be wired directly into Palantir's software. It looked like a photo op between the two loudest CEOs in AI. It is actually something more specific: the clearest public acknowledgment that the AI cycle's value is migrating from the people who build the chips to the people who turn the chips into decisions.

Read it as a division of labor rather than a friendship. Nvidia makes an AI model able to think; Palantir makes it able to act on a business. That is the whole cartography of the partnership. Nvidia contributes the accelerated computing, the CUDA-X libraries, and the open-source Nemotron models; Palantir threads them through the Ontology, the connective tissue at the center of its AI Platform that represents a company's world as interconnected objects, links, and actions. A supply chain, a hospital, a government — rendered as something a model can reason over and, crucially, act on.

The first customer shows why that distinction matters. Lowe's is building a digital replica of its global supply chain with the combined stack, running what amounts to continuous, dynamic optimization across the whole network. It is the difference between rebalancing inventory once a week and letting the system rebalance itself in real time as demand ripples through. Nvidia's cuOpt does the hard optimization math; Palantir decides what to do with the result. Chips alone cannot move a pallet of goods.

Nvidia comes to Palantir for the same reason it courts every software vendor: it wants its silicon inside every AI deployment. But the direction of the dependency is the tell. For Palantir, the partnership is not a dependency at all — it is optionality, the same open models available to any rival, wrapped in a layer Nvidia does not build. Karp and Huang described the goal as fusing Palantir's decision intelligence with Nvidia's infrastructure to deploy AI that delivers "immediate, asymmetric value." Decode that: Nvidia monetizes the compute; Palantir monetizes the outcome.

The alliance deepened in June 2026 toward the customer that matters most on Palantir's profit and margin: the U.S. government. Palantir launched an engine that deploys Nvidia's Nemotron open models inside air-gapped, sovereign environments, letting agencies build custom, frontier-quality models on their own classified data and keep full ownership of the resulting weights. The government is effectively one of the world's largest enterprises — roughly three million civilian employees across commerce, energy, healthcare, and transportation — and it is exactly the segment where Palantir's data-control pitch has the most leverage and the highest margins.

The operating results say the collaboration is more than a deck. In the quarter reported August 2026, total revenue rose 93% year over year to $1.94 billion, and the key number — U.S. commercial revenue, the recurring, high-margin engine the Nvidia stack feeds — grew 149% to $764 million. Customers already on the platform are expanding: net dollar retention ran at 157%, meaning the average existing account is spending 57% more than a year ago, and the Rule of 40 blew out to 155 — roughly triple the growth-plus-margin benchmark of a healthy software company. Adjusted operating margin widened to 62% from 46%, and management raised full-year revenue guidance to about 82% growth.

None of this is free. The stock now trades around $169, roughly where it started 2026 — the multiple has compressed even as the business compounded, which is what a run to a $407 billion market cap costs. The price still carries the thesis several years into the future: about 66 times trailing sales, versus roughly 18 times for Nvidia itself, whose chips power the entire stack and trade at a single-digit fraction of Palantir's earnings multiple.

This is the honest boundary of the story. The collaboration has done what a partnership must do — it reaches named customers, a specific government product, and accelerating revenue in the segments that matter. The long-term case is intact and, if anything, strengthened. The question the collaboration does not answer is the one that decides allocation: whether the opportunity is still compelling here relative to owning the bottleneck further up the stack, which has already delivered similar AI growth at a fraction of the price. On an unchanged thesis at a still-demanding multiple, that is the trade-off worth holding, not the announcement itself.