Nvidia report finds AI set to take off in retail, but openness is required for scale

As one would expect, artificial intelligence was a big theme at last week’s National Retail Federation’s annual event in New York City, as it was last year, but there was one subtle difference.

The 2025 edition was focused more on AI education, whereas I felt this year’s NRF focused more on use cases. In fact, one of the speakers I saw said something to the effect that NRF is no longer a technology show but rather a business outcomes event.

Concurrent with the event, Nvidia Corp. released its latest trends report, State of AI in Retail and Consumer Packaged Goods, which highlights how far retailers and consumer packaged goods companies have come in implementing AI. Nvidia surveyed global industry leaders and found that 58% of companies are “actively deploying” AI solutions. This marks a 16-point increase (42%) from the previous year, pointing to a growing level of maturity.

In retail, AI is moving from vision to reality

This indicates retail and CPG companies are starting to move past experimenting with AI and into more practical use cases. Tools that were once limited to pilots are now showing up in regular operations.

Overall, 91% of the retail industry is engaged with AI either by actively using it or assessing it. AI has contributed to revenue growth, according to 89% of the industry leaders surveyed by Nvidia. Most (95%) said AI has helped reduce annual costs. Ninety-two percent of executives plan to increase their AI budgets in the next year.

Customer-facing and back-office use cases are both on the rise. Digital commerce is at the top of the list. AI use across e-commerce, marketing and advertising rose from 57% to 61%. A key development in this space is agentic commerce, where AI agents respond to customer intent rather than offering recommendations. One example is a shopping assistant that provides personalized help to customers.

Agentic AI is gaining traction in retail. Nearly half of respondents (47%) said they are using or evaluating agentic AI. They named three key strategic goals that AI agents can address, which traditional automation can’t. Boosting speed and efficiency came up most often (57%), followed by enhancing customer experiences and personalization (40%), and improving decision-making using real-time data (40%).

Nvidia is focused on openness to accelerate AI adoption

To help accelerate AI adoption, Nvidia has been aggressively pushing an open mandate to democratize access to AI tools. At NRF I met with Azita Martin, vice president and general manager for AI for retail, CPG and QSR. “There are many technology companies at NRF,” she said. “Nvidia is focused on openness and interoperability, so customers have the flexibility to run their AI applications on any cloud or data center of their choice. As retailers start deploying AI agents at scale, the cost of inference can become very high. Nvidia is focused on optimizing inferencing speeds and reducing inferencing costs of open-source models.”

Martin then added, “What we are seeing is that retailers typically start their AI projects in the cloud with frontier models such as OpenAI and Gemini, but as the number of agents increase, the cost of inferencing grows significantly as these models generate a lot of tokens because of reasoning. We believe a combination of open-source and frontier models can bring that cost down as retailers move to build more agentic applications.”

The survey indicates that Nvidia’s message of open is being embraced as 79% said it’s important to integrate open-source models into their technology stack, largely because it gives them more control over how AI systems are trained using their own data.

In addition to the study, the company introduced two open-source AI blueprints designed to modernize different parts of retail without replacing existing systems. The first blueprint, Multi-Agent Intelligent Warehouse, targets supply chain operations and uses AI agents that run on top of existing systems. In essence, it acts as an AI command layer that mirrors real warehouse roles and turns fragmented telemetry into real‑time, evidence‑backed recommendations on issues like bottlenecks, staffing, equipment health and safety.

The second blueprint, Retail Catalog Enrichment, addresses what Nvidia calls a “sparse data” problem. This blueprint taps into generative AI to turn basic product images and data into detailed descriptions, which can be used in marketing. A typical scenario might involve a home goods retailer working with a set of product photos. Using the Nemotron vision language model, which is part of Retail Catalog Enrichment, retailers can create product metadata like color, material, capacity, style and use cases, along with localized titles and categories that feed search, recommendations and marketing.

The blueprints are part of Nvidia’s broader effort to help retailers deploy agentic AI across retail their retail operations. Nvidia said the next phase involves applying AI more directly inside warehouses and stores. So, retailers can interpret what’s happening on the floor and to optimize inventory and supply chain issues with less manual intervention.

Physical AI looks to ease supply chain pain

Supply chain challenges have only increased over the past year, according to 64% of industry leaders surveyed by Nvidia. Many companies are turning to AI to deal with that pressure, most often to improve operational efficiency and throughput (51%). Meeting customer expectations (45%) and improving traceability and transparency (38%) also came up frequently.

Physical AI is likely to address some of those supply chain challenges. It makes warehouses and distribution centers smarter by connecting AI to cameras, sensors and robotics and enabling physics-based simulation to create a digital twin of a warehouse or a store.  The digital twin can predict the throughput of the facility by evaluating new layouts and operations before implementing them in the physical warehouse.

Though adoption is still early — with 17% of companies using or evaluating it — physical AI can potentially provide more than just task automation. Among the early adopters, physical AI is being used for intralogistics simulation and optimization (33%), robotic pick-and-place operations (23%) and smart forklifts and autonomous mobile robots (18%).

Some of the initial hurdles around AI are starting to ease. Data readiness, which was a major issue at the start of the generative AI wave, is less of a concern now. Only 13% of companies cited training data as a top challenge, a drop from 27% the year prior. Concerns over privacy and data sovereignty also declined from 22% to 18%.

Final thoughts

Retail, once thought of as a slow-moving industry with regard to technology adoption, now appears to be one of the fastest-moving on AI. My research has found that 90% of companies now compete on customer experience and 86% of consumers admitted they would leave a brand after one or two bad transactions. For retailers, the stakes are higher than ever, and every brand is looking for a way to serve its customers faster and more personalized.

However, budgets are still a reality, and this is where the work Nvidia is doing is so important. The company often gets called out for high prices, but retail decision makers should look beyond the cost of chips and servers and at the total cost of token generation, and this is where Nvidia and its ecosystem have been focused on.

Zeus Kerravala is a principal analyst at ZK Research, a division of Kerravala Consulting. He wrote this article for SiliconANGLE. 

Photo: NRF