Blue Yonder and NVIDIA Open a 'Model Training Factory' for Supply Chain Agents

On May 19, Blue Yonder announced a Model Training Factory built on NVIDIA Nemotron — a repeatable system for fine-tuning and testing supply chain models that perform high-value tasks at the level of subject-matter experts. The models are intended to execute multi-step workflows across warehouse management, supply and demand planning, transportation, merchandising, and network operations. CEO Duncan Angove framed the strategy bluntly: "owned intelligence, not rented intelligence — supply chain models trained on the workflows, telemetry, and decision logic that actually run a warehouse or a planning system." Blue Yonder said the first AI models developed through the factory will enter customer production environments later in 2026.

For operators, this is the clearest contrast yet to the hyperscaler-suite story we covered with AWS Connect Decisions and SAP Sapphire 2026. Generic frontier-model APIs are cheap to start and expensive to control: latency varies, behavior drifts with the vendor's release train, and the model has never seen your slotting logic. Domain-specialized models trained on operational telemetry — receiving exceptions, slot heat maps, route deviations, demand sensing signals — are harder to build but cheaper to predict, easier to govern, and far easier to wire into the closed control loops that warehouse and planning leaders actually run. The decision facing operators is no longer model vs. no model. It is rented vs. owned — and whose data graph the chosen model is allowed to learn from over time.