OneRail has launched OmniSTAR, an AI-powered platform that helps retailers, wholesalers and distributors choose how each order should reach its destination.
The system compares company fleets, couriers, parcel carriers and other delivery modes, then selects what OneRail describes as the lowest-cost option that still meets the required service level. It combines Nvidia’s cuOpt optimization engine and cuDF data-processing software with OneRail’s delivery prices, performance records and operational models.
OneRail says GPU acceleration can reduce some calculation times by as much as tenfold. A scenario that previously took 20 minutes may finish in under two, while a week-long calculation may fall to about two days. Those performance figures come from the company and were not independently tested in the source.
Prediction feeds a separate optimization layer
OneRail uses machine learning to estimate service time, lateness risk, first-attempt delivery success and expected price ranges. Those predictions describe what is likely to happen. OmniSTAR then performs a different task: deciding how an order should be fulfilled under cost and service constraints.
That distinction matters in live operations. A prediction can warn that a delivery may be late, but an optimization system must compare alternatives and choose a response before the order is assigned.
Nvidia describes cuOpt as an open-source, GPU-accelerated library for vehicle routing and mathematical optimization. It can account for vehicle capacity, operating windows, travel time, starting points and different forms of cost. Rather than exhaustively testing every possible route, it generates candidate solutions and improves them with accelerated heuristics within a set computing period.
CuDF handles tabular data operations including filtering, joining and aggregation. OneRail combines these components with information drawn from millions of deliveries across a network it says includes more than 12 million drivers and over 1,000 logistics partners.
Live conditions require repeated decisions
CuOpt is stateless, meaning an optimization problem must be modeled and submitted again when operating conditions change. Nvidia cites vehicle breakdowns, driver absences, blocked roads, traffic and urgent orders as examples that can trigger recalculation.
OneRail says OmniSTAR can rerun scenarios as fuel costs, weather and shipping conditions change. Faster processing is therefore important not only for producing an initial plan but for keeping that plan useful during execution.
The platform can also examine how delivery choices affect item-level profitability. OneRail says many retailers still depend on static rules or manual planning, which can limit the number of combinations assessed before dispatch. OmniSTAR is intended to evaluate more alternatives while the decision can still influence the order.
The architecture disclosed by the company centers on accelerated data processing and mathematical optimization. Although the product is described as AI-powered, the source distinguishes its routing solver from the predictive machine-learning models that estimate operational outcomes.
Early deployments test the savings claims
OmniSTAR is already in use with selected enterprise customers. At US Foods, OneRail says it identified delivery arrangements that reduced margins, including low-margin products transported long distances with expensive equipment. The customer then adjusted pricing and restructured some delivery patterns.
OneRail also told CNBC that an unnamed large tire distributor achieved $40 million in run-rate savings over three years. The customer was not identified, and the savings figure was supplied by OneRail. The company projects that OmniSTAR will process more than $6 billion in gross merchandise value during the fourth quarter of 2026; that is a forecast rather than a completed result.
OneRail and Nvidia reportedly worked on the project for three years, including direct collaboration with Nvidia’s cuOpt engineers. The delivery company also participates in the Nvidia Inception program.
OmniSTAR’s practical test will be whether faster calculations translate into sustained savings without sacrificing service. Its launch shows how GPU computing is moving into operational choices that must be refreshed continuously, where the value of an optimized answer can decline within minutes as roads, prices and available capacity change.
