A fine-tuned 30-billion-parameter model outperformed a 550-billion-parameter general-purpose model on supply-allocation tasks inside Nvidia's own operations, scoring 86.7% accuracy versus 55.5% for the model roughly 18 times its size. Nvidia and Palantir announced Thursday that they're testing sovereign AI technology on Nvidia's sprawling supply chain, turning the chipmaker's internal operations into a proving ground for systems other companies can adapt. The partnership builds on work the duo began last October, combining Nvidia's AI computing with Palantir's software to help organizations make complex operational decisions.
The companies fine-tuned Nvidia's Nemotron 3.5 Lightning model using decisions from Nvidia's supply-chain operations team, with Palantir's Foundry and Artificial Intelligence Platform bringing together the data behind those choices. Palantir's Ontology functions as a live map connecting components, factories, capacity and production commitments, while Nvidia's cuOpt software calculates how scarce parts could be distributed. Nemotron weighs the broader context and recommends actions for planners. Nvidia's supply chain spans millions of parts, thousands of suppliers and a global manufacturing network, with a single Vera Rubin rack alone containing roughly 1.3 million parts. The companies plan to extend lessons from Nvidia's deployment to firms in manufacturing, energy, healthcare, automotive and aerospace sectors.
According to Nvidia founder and CEO Jensen Huang, "Supply chains are the operating system of the physical economy, and AI factories are among the most complex systems ever built." Palantir co-founder and CEO Alex Karp goes further, stating that "Nvidia has arguably the most valuable, intricate, and complex supply chain in the world." In a technical blog post published Thursday, Nvidia solutions architects note the result demonstrates how far specialization can go: on a tightly defined allocation task, the 30B model was able to surpass a general-purpose model more than an order of magnitude larger. However, they caution that "this doesn't mean the smaller model is more capable overall," with gains concentrated in the domain it was post-trained on.
The deployment reflects Nvidia's broader push into what it calls sovereign AI, allowing organizations to run and customize Nvidia's AI models inside tightly controlled environments while keeping sensitive data and model weights under their own control. Palantir customers will be able to build versions tailored to their own supply chains by training Nemotron on their proprietary data using Foundry and AIP, then run the resulting system on-premises or through cloud and colocation providers. The approach centers on Nvidia's open-model strategy: the company now publishes weights and, for many models, training data and recipes so developers can customize themselves. Nvidia's Nemotron 3.5 Lightning, released in August, is the 30B model the companies fine-tuned for this supply-chain deployment. For companies considering following Nvidia's blueprint, the technical blog post offers a notable caveat: future production risk forecasting remained difficult despite fine-tuning, and specialization improved the decision task but failed to solve every prediction problem attached to it. The takeaway is that a smaller open model, taught the specifics of a business, can sometimes prove more useful than simply reaching for the biggest model available. The sovereignty pitch may resonate particularly well with organizations that need to keep proprietary operations data internal while still benefiting from advanced AI capabilities, though the complexity of assembling the necessary data infrastructure and domain expertise shouldn't be underestimated.