OneRail is using Nvidia-powered artificial intelligence to help retailers, wholesalers, and distributors make faster and more cost-effective decisions about last-mile deliveries. Its new platform, OmniSTAR, analyzes different fulfillment options and determines how each order can be delivered while balancing cost and service requirements.

The system can compare options such as company-owned vehicles, courier services, parcel carriers, and other transportation providers. Instead of relying only on fixed delivery rules, OmniSTAR evaluates available choices and identifies an option that can meet the required service level at a lower cost.

What Is OneRail OmniSTAR?

OneRail OmniSTAR is an AI-driven delivery optimization platform designed to improve how businesses plan and execute last-mile fulfillment.

The platform brings together Nvidia’s cuOpt optimization engine, cuDF data-processing technology, Nvidia accelerated computing infrastructure, and OneRail’s logistics data. Together, these technologies allow large numbers of delivery scenarios to be processed more quickly.

According to OneRail, the technology can reduce some computing workloads by up to 10 times. A calculation that previously required around 20 minutes can potentially finish in less than two minutes. For larger workloads that once took about a week, the company says processing time can fall to roughly two days.

This speed is important because delivery planning often changes while orders are being fulfilled.

Why Real-Time Delivery Optimization Matters

Last-mile delivery is one of the most complicated and expensive parts of the supply chain. A company may have several ways to deliver the same order, but the cheapest option is not always suitable.

A delivery decision can depend on:

• Transportation cost
• Distance
• Vehicle availability
• Driver capacity
• Delivery deadlines
• Traffic conditions
• Weather
• Fuel prices
• Shipping requirements
• Product characteristics
• Customer service expectations

OmniSTAR is designed to evaluate these variables before an order is assigned to a delivery method.

OneRail CEO Bill Catania told CNBC that faster decisions can help companies protect margins because last-mile fulfillment can represent a significant operating expense.

How OneRail Uses AI in Delivery Planning

OneRail’s technology does more than calculate routes. The company uses machine-learning models to estimate conditions that can influence fulfillment.

These models can predict factors such as:

• Expected service time
• Risk of late delivery
• Probability of successful first-attempt delivery
• Potential delivery price ranges

Those predictions can then support optimization decisions.

The distinction is important. Prediction estimates what may happen, while optimization determines what action should be taken.

For example, an AI model might estimate that one delivery option has a higher risk of arriving late. An optimization system can then compare that option with alternatives and select a better combination of price and service performance.

Research into dynamic vehicle routing follows a similar approach, separating the prediction of changing travel conditions from the process of recalculating routes when new information becomes available.

How Nvidia AI cuOpt Supports OneRail

A major component of OmniSTAR is Nvidia cuOpt, an open-source, GPU-accelerated optimization library designed for vehicle routing and other mathematical optimization problems.

The technology can consider numerous constraints when developing routing solutions. These may include vehicle capacity, travel time, operating windows, starting points, transportation costs, and other operational limitations.

Its cost models can also evaluate distance, time, monetary expense, or combinations of different factors.

OneRail applies this capability beyond conventional route planning. OmniSTAR can use the optimization technology to compare different delivery methods and determine which one provides an appropriate balance between cost and required service.

Does cuOpt Test Every Possible Route?

Not necessarily.

Rather than checking every possible route combination, Nvidia says cuOpt generates potential solutions and progressively improves them using GPU-accelerated optimization techniques.

This approach is designed to produce high-quality solutions within a practical amount of computing time.

For large logistics networks, this matters because the number of possible combinations can grow rapidly as businesses add vehicles, drivers, destinations, delivery windows, and transportation providers.

Nvidia cuDF Helps Process Delivery Data

OmniSTAR also uses Nvidia cuDF, a GPU-accelerated technology for processing tabular data.

Businesses dealing with large logistics datasets may need to filter, join, sort, aggregate, and analyze huge amounts of information before an optimization problem can be solved.

GPU acceleration can help speed up those data-processing operations.

OneRail combines Nvidia’s technology with its own delivery pricing information, operational models, and performance data. The company says its network data represents millions of deliveries involving more than 12 million drivers and over 1,000 logistics partners.

This information can help the platform compare transportation choices across different delivery modes.

How Delivery Data Can Improve Profitability

Delivery decisions can affect profitability at the individual-order level.

For example, a product with a relatively small profit margin may become unprofitable if it requires an expensive vehicle, a long-distance trip, or a costly delivery provider.

OmniSTAR is designed to identify these types of situations by comparing delivery costs with operational requirements.

OneRail says the system can also reveal delivery rules or patterns that increase expenses. Businesses can then use those insights to adjust pricing, fulfillment strategies, or transportation arrangements.

How Does Real-Time Reoptimization Work?

Delivery conditions rarely remain fixed throughout the day.

A driver may become unavailable. A vehicle can break down. Traffic may suddenly increase. A road can close, or a high-priority order may enter the system.

These changes can make an earlier route or fulfillment plan less suitable.

Because Nvidia cuOpt is stateless, changes in operating conditions require the optimization problem to be modeled and submitted again. Nvidia identifies events such as vehicle failures, driver shortages, road closures, traffic changes, and urgent orders as situations that can trigger another optimization process.

OneRail says OmniSTAR can similarly rerun delivery scenarios when factors such as fuel prices, weather, and shipping conditions change.

The goal is to give logistics teams the ability to reassess delivery choices instead of continuing to follow an outdated plan.

What Makes OmniSTAR Different From Traditional Delivery Planning?

Many logistics operations still depend on predetermined rules, manual planning, or limited delivery combinations.

These approaches can work for relatively predictable operations, but they may become harder to manage as order volumes and transportation choices increase.

OmniSTAR is designed to examine more possible combinations within a shorter period.

Instead of asking only “Which carrier normally handles this type of order?”, an optimization-based system can consider a broader question:

“Which available fulfillment method can deliver this order within the required service level at the best overall cost?”

That difference can become important when businesses operate across multiple delivery networks.

OneRail OmniSTAR Enters Enterprise Operations

OmniSTAR is already being used by selected enterprise customers.

OneRail said US Foods used the platform to identify delivery configurations that were negatively affecting margins. One example involved lower-margin products being transported long distances with more expensive equipment.

According to OneRail, the findings helped US Foods modify pricing and change parts of its delivery strategy.

OneRail also told CNBC that a large tire distributor using the platform achieved $40 million in run-rate savings over three years. The customer was not publicly identified, and the savings figure came from OneRail.

The company also told CNBC that it expects OmniSTAR to surpass $6 billion in gross merchandise volume during the fourth quarter of 2026.

OneRail and Nvidia Worked on the Technology for Years

The OmniSTAR project was not developed overnight.

CNBC reported that OneRail and Nvidia had worked together on the technology for approximately three years before the platform’s launch.

According to OneRail, the collaboration involved direct work with Nvidia’s cuOpt engineering team to address challenges related to last-mile transportation and large-scale logistics optimization.

OneRail has also participated in the Nvidia Inception program, which supports startups working with Nvidia technologies.

The partnership reflects a broader movement toward using accelerated computing and AI for complex logistics decisions.

OneRail’s Relationship With FedEx

OneRail’s technology is also connected to broader delivery-network expansion.

In March 2026, FedEx launched FedEx SameDay Local in collaboration with OneRail, connecting customers with a national network of more than 1,000 delivery providers.

This relationship highlights how delivery companies are increasingly combining traditional carrier infrastructure with technology platforms capable of coordinating multiple transportation providers.

What Does AI Mean for the Future of Last-Mile Delivery?

AI is increasingly being used to support logistics decisions that were previously handled through static rules or manual processes.

The biggest potential advantage is not simply faster route calculation. It is the ability to continuously evaluate changing conditions and make decisions using large amounts of operational data.

For retailers and distributors, this could mean better visibility into delivery costs, improved utilization of transportation resources, and more informed fulfillment decisions.

However, optimization systems still depend on the quality of the data and the assumptions used to model the delivery environment. Faster calculations do not automatically guarantee the best business outcome.

The value comes from combining accurate data, reliable predictions, realistic constraints, and optimization that matches the company’s actual service goals.

Final Takeaway

OneRail’s OmniSTAR represents a shift from fixed delivery rules toward faster, data-driven fulfillment optimization. By combining Nvidia’s GPU-accelerated technologies with OneRail’s logistics data, the platform can evaluate routing and delivery-mode choices at a much larger scale.

The main opportunity is to make delivery decisions faster while considering both cost and service performance. As retailers and distributors manage increasingly complex transportation networks, real-time optimization could become an important part of modern last-mile logistics.