In other words, retail CIOs are rebalancing where data is processed: repatriating specific, latency-sensitive steps from public cloud into regional data centers and store locations, even as cloud remains the hub for training, analytics, and multilocation intelligence. That hybrid continuum is how AI moves from pilot to performance.
Retail is entering a moment when the speed of a single interaction can shape the entire customer experience. Every second at the drive-through, the self-checkout, or the service counter matters. The practical way to win those seconds is to give the tools that drive success — AI-assisted ordering, computer-vision loss prevention, dynamic pricing and inventory management, and more — real-time inference close to where data is created. That means that retailers need to bring compute power into the store, running the right slice of compute locally while keeping cloud for what it does best.
What’s driving the shift
AI makes speed a business metric. Vision analytics, voice ordering, dynamic pricing, and associate copilots feel magical only when they respond immediately, and round trips to the cloud struggle to deliver that consistently. Cost and control are part of the picture as well.
Constant cloud calls can pile up egress and per-transaction fees, while moving specific inference steps onsite can reshape that cost curve without abandoning cloud for scale and cross-store learning. The right approach to data is flexible — process what is time-critical locally, extract only what is useful, then sync signals centrally based on needs.
Redefining the “edge” in retail
At a macro level, many enterprises are rebalancing workloads away from a cloud-only posture — repurposing private data centers as regional hubs and, for the most latency-sensitive work, placing compute in stores. Edge compute is part of that rebalancing, not a destination on its own.
In-store “far-edge” nodes run mission-critical, low-latency tasks, such as computer vision, voice ordering, or queue and kitchen optimization. “Near-edge” regional locations coordinate clusters, stage updates, and support services that benefit from proximity. Cloud remains the center of gravity for fleet analytics, training, long-term storage, and omnichannel orchestration. These are guidelines, not rules; placement should flex by use case.
What runs best, where
In a hybrid model, it’s best to establish a general framing of what workloads should be run where:
- Run in the store when a person is waiting or when the process affects live operations. Examples include drive-through speech-to-order, loss-prevention and safety analytics, and real-time inventory checks. The store is also the right spot to filter and compress heavy streams — such as extracting features from video rather than shipping frames — so you save bandwidth and improve privacy posture.
- Run in the cloud when the job is batch, long-horizon, or cross-store, such as model training, enterprise reporting, forecasting, or experimentation and personalization that benefit from elasticity.
In practice, leading retailers are doing both: keep the data lake and training in the cloud and run shelf scanners, video analytics, or POS adjacencies on an appliance in each location for immediacy.
How this underpins retail’s digital fabric
The digital fabric is the enterprise-wide weave of data, apps, devices, and people — kept available via an intelligent, adaptive network that anticipates issues, routes around them, and enforces policy end-to-end. It connects stores, regional hubs/data centers, and cloud with policy-driven networking and a control plane to place, secure, and update apps wherever they run. It underpins repatriation (datacenter hubs) and store-edge, so teams can move workloads to the right place without reinventing operations each time.
Retailers are accelerating this foundation — IDC notes 61% expect to be extensively connected by 2026; 42% already use SDWAN (another 12% migrating by end2025); and 61% are increasing managed network services. These are the building blocks that make a fabric real across thousands of sites.
The business outcomes of shifting the right slice on-prem
Seconds shaved become throughput gained. By shifting the right slice of work to on-prem within a hybrid model, retailers and QSR operators can achieve meaningful results, including
- Time reduction in critical functions: In environments like QSR, moving critical, repeated functions — like voice-to-order processing in the drive-thru — from cloud to instore compute can help cut average processing time.
- Cloud cost rightsizing: Running inference locally reduces per-call cloud charges and bandwidth, while cloud remains the hub for training and fleet-level learning.
- Operational continuity: High availability firewalls plus triple-path connectivity help keep stores transacting during carrier or power events.
- Fast time to value at scale: Blueprinted deployments and centralized orchestration speed consistent rollouts across footprints.
Real-time retail runs on the edge
Retail’s next wins come from placing the right compute in the right place — and running it on one fabric. For many retailers, that means repatriating select workloads to data-center hubs and bringing the most time-sensitive steps into the store while keeping cloud for scale and cross-store learning.
Learn more about how Comcast Business technology solutions can help power retail operations.