NVIDIA and Palantir Revolutionize Supply Chain with AI
How AI transforms complex supply chain decisions at NVIDIA
1The Challenge of NVIDIA's Supply Chain
NVIDIA manages one of the world's most complex supply chains, spanning from wafer production to final token generation. The process involves thousands of suppliers and millions of parts globally. Each component, like Grace CPUs and Blackwell GPUs, has dynamic availability, making weekly allocation decisions critical. The supply chain's complexity resembles a daunting combinatorics problem, requiring precise coordination across multiple manufacturing sites and suppliers.
2Time of Ownership Metric
NVIDIA measures supply chain performance using Time of Ownership (TOO), tracking the interval from material receipt at a manufacturing site to its departure as part of a product. This metric is crucial for minimizing delays. Components often arrive at different times, causing early arrivals to wait for late ones. TOO helps NVIDIA optimize this process, ensuring efficient material flow and reducing bottlenecks in the production timeline.
3Critical Material Allocation Problem
Every week, NVIDIA faces the critical material allocation problem: deciding how much of each constrained material goes to which manufacturing sites. This involves balancing current and future needs, with near-term weeks already committed. The allocation spans the current and next quarters, requiring constant re-evaluation. Human planners incorporate qualitative factors like emails and geopolitical events, which traditional quantitative models often miss.
4Digital Supply Chain Intelligence
NVIDIA and Palantir created a Digital Supply Chain Intelligence command center in Palantir Foundry. This unified platform integrates material, capacity, and qualitative signals into a single governed data layer. The Ontology connects materials, sites, commits, and unstructured data, providing a complete operational picture. This enables planners to simulate scenarios and make informed allocation decisions, laying the groundwork for AI-driven improvements.
5NVIDIA cuOpt for Quantitative Solutions
NVIDIA cuOpt addresses the quantitative side of material allocation using mixed-integer linear programming. It minimizes Time of Ownership across manufacturing sites, identifying binding constraints each period. While effective, cuOpt lacks the ability to incorporate qualitative factors like human planners. This gap led NVIDIA to explore AI solutions to enhance decision-making accuracy and efficiency in the supply chain.
6Nemotron 3.5 Lightning
NVIDIA post-trained Nemotron 3.5 Lightning on allocation decisions, rationales, and outcomes using NeMo tools. The 30B model achieved 86.7% accuracy in allocation decisions, outperforming larger models by 31.2 percentage points. Nemotron captures human expertise, enabling governed retraining runs that compound institutional knowledge. This AI flywheel shortens onboarding and improves future decisions, transforming supply chain management.
7Human vs. AI Performance
Human planners consistently outperformed quantitative models by incorporating emails, weather forecasts, and supplier debriefs. Nemotron bridges this gap by codifying human expertise into AI. The model's recommendations, whether accepted, edited, or overridden, feed back into the Ontology. This continuous learning loop enhances decision-making over time, combining the best of human intuition and AI precision.
8Operational Transformation
The most significant transformation occurs in codifying human expertise. Nemotron captures nuanced decision-making factors that quantitative models miss. This shift enables NVIDIA to compress time-to-rack, the interval from silicon leaving the fab to system arrival in data centers. By integrating AI, NVIDIA achieves faster, more accurate allocations, reducing delays and improving supply chain efficiency.
9Global Impact
NVIDIA's AI-driven supply chain solutions have global implications. The Vera Rubin supply chain is twice as large as Grace Blackwell, highlighting the scalability of these tools. By optimizing material allocation, NVIDIA ensures timely delivery of critical components like GPUs and CPUs. This approach sets a benchmark for other tech companies facing similar supply chain challenges in a dynamic global market.
10Future Steps
NVIDIA encourages exploration of the Nemotron LoRA customization cookbook for adapting open-weight models to agentic workflows. Open Nemotron models are available on Hugging Face, and NIM microservices can be deployed via build.nvidia.com. Subscribing to NVIDIA generative AI news and following their channels ensures updates on advancements in AI-driven supply chain solutions.
11The AI Flywheel Effect
NVIDIA's AI flywheel compounds institutional knowledge, improving supply chain performance over time. Each decision, whether AI-generated or human-overridden, enriches the Ontology for future retraining. This creates a virtuous cycle of continuous improvement, reducing onboarding time and enhancing accuracy. NVIDIA's approach demonstrates how AI can transform complex, global supply chains into agile, efficient systems.
Original article: Nvidia Developer Blog — NVIDIA and Palantir Revolutionize Supply Chain with AI