Nvidia and Palantir Deploy Fine-Tuned 30B Model That Outperforms Massive 550B Competitor in Supply Chain Work

Nvidia and Palantir revealed Thursday a joint initiative to deliver what they call "sovereign AI to critical supply chains," beginning with deployment across Nvidia's own global operations. The collaboration demonstrates how Nvidia's open-model strategy can operate within enterprise environments, with Palantir customers positioned to replicate the same methodology for their own supply chain needs.

This announcement extends a partnership that began last October, when the two companies committed to merging Nvidia's AI infrastructure and models with Palantir's software platform to enable organizations to make sophisticated operational decisions using artificial intelligence. By June, they had broadened their collaboration to encompass sovereign AI capabilities, permitting organizations to execute and modify Nvidia's models within isolated, controlled settings while maintaining ownership of sensitive information and model parameters. The current deployment applies this framework directly within Nvidia, where a smaller, customized model is already delivering superior results compared to substantially larger alternatives.

A proving ground for sovereign AI

Nvidia and Palantir have adapted Nvidia's 30-billion-parameter Nemotron 3.5 Lightning model using historical decisions from Nvidia's supply-chain operations team. Palantir's Foundry and Artificial Intelligence Platform (AIP) consolidate the data underlying those decisions, while its Ontology functions as a dynamic framework linking parts, production facilities, available capacity and manufacturing commitments. Nvidia's cuOpt software determines optimal distribution of limited components, with Nemotron providing context-aware analysis and operational recommendations to supply planners.

The companies intend to "extend the learnings from Nvidia's deployment" to organizations across manufacturing, energy, healthcare, automotive and aerospace sectors. Palantir's customer base will have the ability to construct customized versions for their respective supply chains by training Nemotron on their confidential information through Foundry and AIP, then deploying the resulting system in on-premises infrastructure or through cloud and colocation services.

In essence, Nvidia and Palantir are operationalizing the sovereign AI partnership they announced in June within Nvidia itself, simultaneously using that implementation as a reference architecture that other organizations can modify for their distinct operational requirements.

Nvidia as a test case

With a market capitalization of $5.4 trillion, making it the world's most valuable publicly traded company, Nvidia has compelling reasons to pilot the technology internally. The company's supply chain encompasses millions of individual parts, thousands of supplier relationships, and a worldwide manufacturing ecosystem. A single Vera Rubin rack contains approximately 1.3 million parts, according to the company. Precise timing and coordination are essential: the absence of even one component can halt assembly operations while other delivered materials accumulate.

Supply chains are the operating system of the physical economy, and AI factories are among the most complex systems ever built.

Jensen Huang, Nvidia founder and CEO

Nvidia founder and CEO Jensen Huang identifies this very complexity as the reason supply chains represent an ideal application for the technology. He contends that contemporary AI system development increasingly demands orchestration across an intricate network of organizations and components, spanning semiconductors, memory systems, manufacturing, networking infrastructure, electrical systems and thermal management.

Nvidia has arguably the most valuable, intricate, and complex supply chain in the world.

Alex Karp, Palantir co-founder and CEO

Palantir co-founder and CEO Alex Karp emphasizes that Nvidia's operational environment presents an exceptionally rigorous testing ground for validating the companies' methodology.

The sovereignty selling point

Nvidia has positioned itself prominently in discussions surrounding open-model artificial intelligence. In July, Huang made his inaugural post on X platform to endorse an industry communication urging U.S. policymakers to champion frontier open-weight models, contending they provide enterprises and nations greater autonomy over their AI systems.

During early September, Nvidia executed a $12.9 billion acquisition of Hugging Face, the platform widely recognized as the "GitHub for AI models." Following apprehensions that ownership by the world's preeminent AI chip manufacturer could compromise Hugging Face's impartiality, Huang committed that the platform would maintain its open character, persist in hosting models from diverse industry participants and extend support for computing hardware beyond Nvidia's own portfolio.

Nemotron occupies a central position in Nvidia's open-model initiative. The designation traces to 2023, when Nvidia introduced its inaugural Nemotron-3 8B models designed for enterprise customization and fine-tuning. These initial models were accessible via Hugging Face and Nvidia's NGC catalog, though Nvidia's proprietary community license governed and restricted access. While the models permitted customization and provided available weights, the broader "open model" characterization Nvidia employs currently emerged subsequently.

Nvidia released the Nemotron 3 series in December, initially offering Nano, Super and Ultra variants targeting distinct agentic AI applications. The company now distributes model parameters and, for numerous models, includes training data and implementation guides enabling developer customization. Nemotron 3.5 Lightning, introduced in August, represents the 30-billion-parameter model that Nvidia and Palantir have customized for this supply-chain application.

This transparency forms the foundation of the sovereignty concept: organizations can refine Nemotron using proprietary information while maintaining that information, the model parameters, and processing operations entirely within their controlled infrastructure.

Specialization over size

Nvidia's implementation produces measurable outcomes for external evaluation. The customized 30-billion-parameter Lightning variant achieved 86.7% accuracy on its supply-allocation assignment, compared to 55.5% for the 550-billion-parameter Nemotron 3 Ultra—a model approximately 18 times larger.

In a technical publication released Thursday coinciding with the primary announcement, Nvidia solutions architects Nell Barber, Rana Haber, and Aastha Jhunjhunwala highlight how substantially domain-specific training can enhance performance. When applied to a narrowly focused allocation assignment, the 30-billion-parameter model surpassed a general-purpose model substantially larger in scale.

This doesn't mean the smaller model is more capable overall. Its gains are concentrated in the domain it was post-trained on. Future production risk forecasting remained difficult despite fine-tuning. Specialization improved the decision task but failed to solve every prediction problem attached to it.

Nell Barber, Rana Haber, and Aastha Jhunjhunwala, Nvidia solutions architects

For enterprises evaluating Nvidia's methodology, the more significant insight may be straightforward: a more compact open model, refined using business-specific information, can frequently deliver greater practical value than selecting the largest available model.