Blue Yonder builds model factory with NVIDIA

Blue Yonder said on May 18 that it is building a “Model Training Factory” with NVIDIA to create specialized AI models for supply-chain work, rather than relying only on general-purpose frontier models. The announcement came at ICON 2026, Blue Yonder’s annual customer conference in San Diego, where the company said the factory is built on NVIDIA Nemotron and will be used to fine-tune and test domain-specific models for warehouses, planning and transportation. Blue Yonder said the system is intended to support what it calls an autonomous supply chain, with AI agents trained on operational workflows and enterprise data. CEO Duncan Angove framed the effort as a push for “owned intelligence, not rented intelligence.” ### Why is Blue Yonder building its own supply-chain model factory? Blue Yonder said the factory is a repeatable system for fine-tuning and testing highly specialized supply-chain models, with the goal of producing agents that can handle multi-step workflows at the level of subject-matter experts. The company said those models are designed to work alongside human operators and then be graded for quality before broader deployment. (businesswire.com) Duncan Angove said supply chain is not a generic reasoning problem but an operational one shaped by hard constraints, real-time execution and physical consequences. In the company’s description, that means models need to be trained on the workflows, telemetry and decision logic that run warehouses, planning systems and transport networks, instead of depending only on external model APIs. (businesswire.com) ### What is NVIDIA providing in the partnership? NVIDIA’s role, according to Blue Yonder, is the model and tooling layer. Blue Yonder said it is using NVIDIA Nemotron open-source models and NeMo AI tools, along with NVIDIA AI Enterprise, to build and deploy the training system. (businesswire.com) NVIDIA has separately been promoting supply-chain AI systems that combine language models with optimization engines. In a May 4 technical post, NVIDIA said its cuOpt agent skills let AI systems translate natural-language supply-chain problems into mathematical models for production planning, inventory management and routing. That aligns with the kind of constrained decision work Blue Yonder is targeting. (businesswire.com) ### What problem is Blue Yonder trying to solve? Blue Yonder said the economics of running large frontier models in production are becoming harder as inference demand rises. The company said its answer is a hybrid model: use frontier models where broad reasoning is needed, and use custom supply-chain models where precision, latency and cost matter more. (developer.nvidia.com) Diginomica reported that Blue Yonder executives tied that argument to tasks such as inventory positioning, transfer decisions and other operational recommendations that depend on internal demand data, lead times, item masters and business constraints. Chief Product Officer Gurdip Singh told the publication that frontier models are not the right answer for every problem. (businesswire.com) ### How does this fit into Blue Yonder’s broader product strategy? Diginomica reported that Blue Yonder has spent roughly four years rebuilding its technology stack under Panasonic ownership and has invested more than $2.5 billion in research and development to create a more interoperable platform. The publication said Blue Yonder introduced domain agents at ICON 2025 and is now focusing on the model layer that powers those agents. (diginomica.com) ICON 2026 is scheduled for May 17-20 in San Diego, according to Blue Yonder’s event page. The company said the conference includes more than 160 sessions and presentations from Blue Yonder, customers and partners, providing the venue for additional details on the NVIDIA partnership and related agentic AI products. ### What comes next at ICON? Blue Yonder’s event agenda shows ICON continuing through May 20, with general sessions, breakout sessions and partner programming in San Diego. (diginomica.com) Blue Yonder said the conference includes customer, expert and partner sessions, and the company is expected to use those sessions to expand on how the model factory will be applied across warehouse management, planning, merchandising and network operations. (blueyonder.com)