OneRail Deploys Nvidia-Powered AI to Cut Last-Mile Delivery Calculations from Hours to Minutes

OneRail has introduced an AI-driven delivery optimization platform built on Nvidia technology that assists retailers, wholesalers, and distributors in determining the most efficient way to fulfill individual orders. The platform, branded OmniSTAR, weighs various fulfillment channels—including company-owned delivery fleets, courier services, parcel carriers, and alternative shipping methods—before recommending the most economical choice that satisfies service-level requirements.

The solution integrates Nvidia's cuOpt decision optimization engine and cuDF data processing framework with OneRail's proprietary delivery pricing and performance information. Nvidia's accelerated computing infrastructure powers the underlying calculations for routing and delivery-mode selection.

According to OneRail, the platform achieves a tenfold reduction in processing duration. Tasks that previously required 20 minutes now complete in under two minutes, and week-long calculations finish in approximately two days. This acceleration enables optimization to operate within active delivery workflows, permitting evaluation of multiple fulfillment options before an order receives its final assignment.

If you don't have the ability to make lightning-fast decisions, you're giving up margin. Last-mile fulfilment is expensive.

Catania, in an interview with CNBC

From prediction to delivery decisions

OneRail's broader AI infrastructure employs prediction and optimization across distinct phases of the delivery workflow. The company's machine-learning algorithms forecast elements such as service duration, lateness probability, likelihood of successful first-attempt delivery, and anticipated cost ranges.

These forecasts inform optimization engines that establish how an order should be processed. Concurrently, OmniSTAR evaluates alternative fulfillment approaches and selects one based on expense and service criteria. Academic research on dynamic vehicle routing similarly distinguishes between forecasting evolving circumstances and recalculating operational choices as fresh data emerges. A 2024 analysis published in the European Journal of Operational Research identified travel-time prediction and real-time re-optimization as distinct research domains within time-dependent routing.

Nvidia cuOpt handles route optimisation

Nvidia characterizes cuOpt as an open-source, GPU-powered optimization toolkit designed for vehicle routing and broader mathematical optimization challenges. According to Nvidia's technical documentation, cuOpt accommodates vehicle expenses, load limits, travel durations, time windows, departure points, and various operational constraints during route computation. Its pricing mechanisms can incorporate distance, duration, financial charges, or weighted combinations thereof.

OmniSTAR applies cuOpt to both routing calculations and fulfillment-mode assessment. This dual application enables the system to contrast accessible fulfillment pathways for a given order and pinpoint the most cost-effective option that maintains service standards. OneRail notes that numerous retailers presently depend on predetermined guidelines or human-driven planning for such determinations, whereas OmniSTAR is engineered to examine substantially more delivery scenarios within compressed operational windows.

Rather than testing all conceivable routes exhaustively, Nvidia's solver creates candidate solutions and progressively refines them through GPU-accelerated heuristics to yield superior outcomes within a predetermined timeframe. The system additionally employs Nvidia cuDF, a GPU-powered library for processing structured data, encompassing data filtering, merging, and summarization operations.

OneRail pairs these tools with its proprietary delivery information and operational frameworks. The underlying dataset originates from millions of completed deliveries spanning a network encompassing over 12 million drivers and more than 1,000 logistics collaborators. This data encompasses pricing details and performance metrics across distinct transportation categories. OneRail indicates that OmniSTAR leverages this information to uncover delivery procedures that inflate expenses and measure how delivery selections influence profitability at the item level.

OmniSTAR's disclosed design emphasizes GPU-accelerated data handling and mathematical optimization, with cuOpt serving as the optimization module for scenarios including vehicle routing. Since cuOpt maintains no persistent state, shifts in operational parameters necessitate reformulating and resubmitting the optimization problem. Nvidia identifies vehicle malfunctions, driver unavailability, road obstructions, traffic congestion, and urgent new orders as instances prompting dynamic reoptimization. OneRail reports that OmniSTAR can reprocess delivery scenarios as factors like fuel expenses, meteorological conditions, and transportation circumstances fluctuate, and has stated that its cuOpt implementation permits assessment of additional routing scenarios and faster route recalculation relative to its prior methodology.

OmniSTAR moves into live operations

OmniSTAR is currently operational with select major clients. At US Foods, OneRail identified delivery arrangements that were eroding profitability, specifically low-margin goods being transported substantial distances using premium transportation equipment. Following these insights, US Foods modified its pricing approach and reorganized certain delivery routes. OneRail also disclosed to CNBC that an undisclosed major tire distributor leveraging the system achieved $40 million in annualized savings over a three-year span, though the company did not identify the customer or independently verify the figure. Additionally, OneRail communicated to CNBC its projection that OmniSTAR will surpass $6 billion in gross merchandise volume in the fourth quarter of 2026.

CNBC reported that OneRail and Nvidia invested three years in development prior to the platform's introduction. OneRail indicated the partnership encompassed direct collaboration with Nvidia's cuOpt development staff on last-mile delivery and broad-scale logistics optimization, as well as OneRail's enrollment in the Nvidia Inception programme. In March of this year, FedEx unveiled FedEx SameDay Local in partnership with OneRail, granting customers entry to a nationwide community of more than 1,000 delivery operators.