Kinaxis GPU acceleration changes the economics of large-scale planning scenarios

Supply Chain Digital's George Cooper reported on 18 March 2026 that Kinaxis is extending Maestro with GPU-accelerated optimization using NVIDIA cuOpt.

In testing disclosed by Kinaxis on a semiconductor planning model with nearly 50 million decision variables, covering more than 40,000 SKUs across a six-quarter daily planning horizon, total end-to-end calculation time was reduced by up to 12X, from more than three hours to approximately 17 minutes. Kinaxis separately states that core optimization solve time improved 23X, reducing compute time by more than 95% while maintaining comparable solution quality. These are supplier-reported benchmark results and should not be treated as independently validated general performance.

The decision implication is more important than the raw benchmark. Faster end-to-end planning can materially expand the number of scenarios that planners can evaluate inside a real planning cycle, particularly when models span very large variable sets and multiple horizons. But compute speed only creates value when scenario definitions, constraints and decision ownership are sufficiently stable for the organization to act on the additional iterations.

Dataleo perspective: GPU acceleration can move planning from one expensive batch run toward iterative decision exploration. The limiting factor then shifts from runtime to the quality of the decision model and the operating process around it. The main failure mode is faster production of scenarios that the organization cannot consistently interpret, compare or execute.

Source: George Cooper, Supply Chain Digital.