OneRail unveils OmniStar, delivery-decision AI built with NVIDIA

Last-mile delivery company OneRail has launched OmniStar, a new platform built on NVIDIA's AI software, CNBC reported on September 1. When a retailer needs to decide the most efficient delivery method for each order, a judgment call that used to take 20 minutes, OmniStar now finishes it in two and a half minutes, the company says. OneRail CEO Bill Catania framed the cost pressure behind last-mile delivery bluntly: fail to make lightning-fast decisions, and you're giving away your margin.

To put that in context: NVIDIA doesn't build chatbots itself — it sells the chips and software that power AI computation. OneRail built OmniStar by borrowing that hardware and software to run, at scale, a model it trained on its own delivery data.

Three years of collaboration with NVIDIA

As e-commerce grew, retailers had to build supply chains flexible enough to handle rising order volume — but the process of deciding delivery routes remained manual and fragmented between retailers and logistics providers, the report notes. David Daeschler, OneRail's head of AI, said the company began working with NVIDIA on this problem three years ago. "Where AI makes these calls extremely fast, NVIDIA's hardware and software are what actually run the system at scale," he said. Azita Martin, NVIDIA's vice president and general manager for retail and consumer goods, described the result as "a real-time decision layer that matches orders to the right carrier and method at the right cost, without relying on fixed rules or manual planning."

What the numbers from real deployments show

OneRail says OmniStar is already live with some customers. One large tire distributor used resources more efficiently and saved $40 million on a three-year run rate, the company said. OneRail expects fourth-quarter gross merchandise value to exceed $6 billion.

Metric	Value
Delivery driver network	More than 12 million
Logistics partners	More than 1,000
Tire distributor's 3-year savings (run rate)	$40 million
Projected Q4 gross merchandise value (GMV)	More than $6 billion

The company says this data comes from the delivery network OneRail has built up on its own. The larger the pool of drivers and logistics partners, the more route-and-cost combinations the AI has to learn from.

A tool for smaller retailers facing off against Amazon and Walmart

Earlier this year, OneRail announced a partnership with FedEx to extend same-day delivery to all its customers. Catania added that this partnership, combined with OmniStar, lets the company work better with smaller retailers too. Daeschler described OmniStar as doing for delivery "what ChatGPT and Anthropic did for writing" — taking a base model, training it on the company's own data, and using it to give people access to affordable delivery.

Editor's take

What stands out here is NVIDIA sticking to its lane once again. Rather than jumping into the chatbot race, it lent its chips and software to speed up decision-making in an entirely different industry — logistics — cutting that process from 20 minutes to two and a half. It's the same underlying structure as OpenAI or Google training models on NVIDIA GPUs: whoever builds the final service, NVIDIA lays the foundation underneath it.

For anyone who's followed logistics for a while, last-mile route optimization used to be territory only big retailers could afford to invest in with their own systems. Amazon and Walmart built their own logistics networks and algorithms, while smaller retailers had to rely on manual judgment or static rules. If OneRail can train a model on data from its 12 million drivers and 1,000-plus partners and take that judgment call off retailers' hands, that gap starts to narrow.

It's worth noting, though, that all these figures come from the company itself, as reported to CNBC. The tire distributor's $40 million in savings and the projected $6 billion Q4 GMV haven't been independently verified. For logistics and retail companies elsewhere looking at this case, it's probably more useful to benchmark the underlying speed metric — how many minutes a route decision takes — against their own systems, rather than fixate on the announced savings figure itself. The real test for this platform over the coming months will be how many more customers actually adopt OmniStar, and whether that shows up in the numbers when Q4 earnings come out.