Supermicro's Marc Del Vecchio - From pilot to production: infrastructure strategies for retail AI

By Marc Del Vecchio, Retail/ Staff Solutions Manager, Supermicro

Artificial intelligence (AI) is rapidly becoming a cornerstone of modern retail.

As the industry moves from pilot projects to full scale deployment, the question is no longer whether to use AI, but how to implement it efficiently and reliably across stores and operations. For many retailers, the answer lies in Edge AI, deploying intelligent systems directly at the point of data generation, where real-time responsiveness and operational realities are critical. 

Unlike traditional cloud-based AI, which requires data to be sent to distant data centres, Edge AI processes information locally, right where it is generated. This shift brings several tangible benefits: it eliminates network latency, enabling real-time decision-making for loss prevention, inventory management, and customer engagement; it reduces bandwidth and energy use, as only relevant insights, not raw data, need to be transmitted; and it keeps sensitive customer and operational data on-site, supporting privacy and compliance. 

Unlocking practical value

Retailers are putting Edge AI to work to address a wide range of operational needs and to enhance both efficiency and the customer experience. One of the most impactful applications is in video analytics for loss prevention. Loss includes innocent mistakes made at checkout counters as well as criminal activity, and is a global and widespread problem.  

Edge AI is invaluable for analysing data from security cameras to identify suspicious behaviour on retail floors, at Point of Sale (PoS) stations, and in stockrooms. This functionality can be enhanced through the use of Visual Language Models (VLMs). VLMs are designed specifically to understand and analyse video content, much like large language models (LLMs) do for text.

By leveraging VLMs, Edge AI systems are able to interpret and make sense of activities recorded on video. As a result, retailers receive near-instant insights and can respond to incidents or anomalies as they happen. 

Edge AI also plays an important role in optimising inventory management. By leveraging technologies such as RFID tags or Bluetooth Low Energy transponders, retailers can achieve greater accuracy in tracking stock levels. Edge AI can be set up to monitor shelf availability and automatically notify staff when items need to be restocked.

In addition, it can significantly improve the customer experience. For example, retailers can deploy chatbots powered by LLMs to answer product questions or suggest suitable items. Video analytics can also be utilised to monitor in-store shopper behaviour, enabling retailers to deliver targeted promotions on digital displays or kiosks at the point of engagement.

From pilot to production: infrastructure strategies for retail AI

Edge AI enables precise and efficient inventory management in retail. By integrating technologies such as RFID tags or Bluetooth Low Energy transponders, retailers can achieve real-time tracking of stock movements and maintain accurate inventory levels.

Right-sizing and real-world challenges

Right-sizing plays a critical role when introducing Edge AI. Rather than adopting a one-size-fits-all approach, retailers should assess the operational requirements of each location, considering factors like store size, expected data volume, and available infrastructure.

A large flagship store may require a high performance, multi-GPU system, while a small-format shop or kiosk might benefit from a compact, fanless server. This tailored approach ensures that resources are used efficiently, costs are controlled, and future scalability is supported. 

Data management is another key consideration. The sheer volume and diversity of information generated in retail, ranging from video streams to transaction logs, can quickly overwhelm traditional IT systems. Efficient local processing, secure storage, and robust data protection are essential to keep operations running smoothly and to safeguard sensitive information.

Finally, integrating Edge AI with existing systems can be complex, especially in environments where legacy PoS, inventory, and signage solutions are still in use. Leveraging modular hardware, software development kits (SDKs), and pre-built modules can simplify integration, accelerate deployment, and help retailers realise value from their AI investments more quickly.

Solution spotlight

Supermicro’s modular server design philosophy has enabled the company to assemble one of the most comprehensive portfolios in the industry, spanning both cloud and edge environments.

This flexibility allows retailers to select from a wide array of standard systems or work with Supermicro to develop tailored solutions that precisely match their operational needs and growth plans. The company’s deep expertise in AI infrastructure has also fostered a strong partnership with NVIDIA, the global leader in AI acceleration. 

A prime example of this collaboration is the integration of the NVIDIA Jetson Orin NX platform into Supermicro’s compact, fanless ARS-E103-JONX system.

The Supermicro ARS-E103-JONX, featuring the NVIDIA Jetson Orin NX platform, provides high performance AI inferencing in a compact, fanless design with minimal power requirements. Supporting up to 157 TOPS for simultaneous AI workloads and equipped with a wide range of high speed I/O interfaces, this system is exceptionally well-suited for demanding edge applications across retail, manufacturing, logistics, and healthcare environments. Source: Supermicro.

This solution delivers high performance AI inferencing at low power, supporting up to 157 TOPS for multiple concurrent AI pipelines. With a broad set of high speed I/O options, including 5G and Wi-Fi, this platform is ideally suited for advanced edge applications in retail, as well as in manufacturing, logistics, and healthcare.

Deploying AI at the edge requires specialised hardware, and AI accelerators are at the heart of these systems. NVIDIA’s portfolio of AI optimised CPUs and GPUs, including the Jetson Orin NX for embedded edge and the RTX PRO 6000 Blackwell Server Edition and H200 NVL GPUs for enterprise edge, provides scalable performance for various retail workloads.

To further streamline the adoption of AI, platforms like NVIDIA Metropolis offer retailers a head start. By providing pre-trained models, software development kits, and an optimised infrastructure, Metropolis enables rapid development and deployment of intelligent video analytics.

This means retailers can quickly implement solutions for monitoring foot traffic, managing inventory, and analysing shopper behaviour - without the need to build everything from scratch. The result is a faster time to value, reduced development costs, and the ability to transform everyday video data into actionable business insights.

Supermicro