Retail Networks, Edge Computing and the In-Store Technology That Got Removed

How CodoraTech is funded: CodoraTech is supported by advertising and, in some articles, by affiliate links. Where an article contains affiliate links, we say so at the top of that article.

Kroger’s store circuits ran at 3 Mbps until a rebuild that Fierce Network described in May 2026 lifted them to an average of 20 Mbps. That single figure is a corrective to the way in-store technology is usually written about. Alongside the retailers pushing compute into stores sit the ones that pulled technology back out: Amazon removed Just Walk Out from its US Amazon Fresh grocery stores, Walmart cancelled its shelf-scanning robot contract, and the Federal Trade Commission barred Rite Aid from facial-recognition surveillance for five years.

Key facts

• Walmart’s own engineering blog reports more than 100,000 routers and switches and 450,000 wireless access points
• Kroger store circuits moved from 3 Mbps to an average of 20 Mbps, per Fierce Network, May 2026
• Chick-fil-A put Kubernetes in restaurants on roughly $1,000 of commodity hardware per site
• Amazon began removing Just Walk Out from US Amazon Fresh stores in April 2024, NPR reported
• The FTC’s December 2023 order bans Rite Aid from facial-recognition surveillance for five years

Inside Walmart’s own account of its store network

Walmart Global Tech publishes the most detailed public description of a retail network at scale. It covers more than 10,000 stores, 1,000 supply-chain locations and e-commerce sites across 19 countries, serving about 270 million customers a week. The network runs more than 100,000 routers and switches and 450,000 wireless access points, supporting five million associate devices, and Walmart says it handles approximately 10 petabytes of customer-interaction data per day. Walmart also says it is placing edge compute closer to stores and distribution centres specifically to cut latency for real-time inventory applications, naming its Walmart Cloud Native Platform and a Triplet Model. Its first store network went in during 1988.

Every one of those numbers is Walmart’s own, published in an engineering and recruiting blog. None of it has been independently audited, and the blog reports capacity rather than outcomes: it does not say what the network changed about sales, waiting times or stock accuracy. Treat it as a credible description of scale from the organisation best placed to know, and as nothing more than that.

A thousand dollars of hardware per restaurant

Chick-fil-A’s technology team has published the clearest engineering account of retail edge computing available. It runs Kubernetes clusters inside individual restaurants, and at full scale anticipated more than 2,000 clusters with tens of containers per cluster, on approximately $1,000 of commodity hardware per site. The business reason it gave was blunt: many restaurants were operating at greater than three times the volume of their original design.

The stated engineering drivers are worth separating from the marketing version of edge computing, because they are about failure rather than speed alone.

• Low-latency processing inside the restaurant rather than a round trip to a data centre
• Applications that keep working when the wide-area link drops, which is the point Chick-fil-A stresses most
• High availability across many small sites rather than a few large ones
• Horizontal scaling as device counts per restaurant grow

Kroger’s stores were running on 3 Mbps

Fierce Network reported in May 2026 on a multi-year rebuild of Kroger’s store network, covering 2,730 stores in 35 states, 11 million customers a day and 63 million households a year, with an IT organisation of 6,000 and CIO and Group Vice President Jim Clendenen named as leading it. The number that matters is the circuit speed: stores went from 3 Mbps to an average of 20 Mbps, scaling toward 50 Mbps, with gigabit as a long-term goal. Kroger segments its Wi-Fi into employee, guest and device traffic.

That baseline is the useful part. A grocery chain of that size was operating stores on 3 Mbps links until very recently, which sets a realistic floor under any claim about artificial intelligence at the shelf edge. Note also what the reporting does not say: neither the trade coverage nor the equipment supplier’s own material states which Wi-Fi generation was deployed or how many access points were installed, so this piece does not describe Kroger’s rebuild as a Wi-Fi 6 deployment.

Just Walk Out left the grocery aisle

In April 2024, NPR reported that Amazon had begun removing its Just Walk Out system from its US Amazon Fresh grocery stores, while keeping it in UK locations. More than half of the 40-plus Fresh stores had been using it. The replacement is the Dash Cart, a trolley with scanners, produce scales and a screen showing a running receipt.

• Shoppers were reported as weary of passing through entry gates and being tracked by cameras and sensors
• Customers wanted a running price tally while shopping rather than a receipt after leaving
• The system was expensive and complex at grocery scale
• It required behind-the-scenes human involvement

Shelf robots, false matches and a five-year ban

On 3 November 2020, Retail Dive reported that Walmart had ended its contract with Bossa Nova Robotics. Roughly 500 stores had the shelf-scanning robots and Walmart had planned to reach 1,000 stores by the end of that year, nearly a quarter of its approximately 4,500 US locations. Walmart’s stated reason is the interesting part: it had found different ways to track inventory, including other automation and its own employees, who were by then walking the aisles more frequently to fill online orders. A human process created by e-commerce growth displaced the robot.

The sharper case is Rite Aid. On 19 December 2023 the Federal Trade Commission announced an order banning the retailer from using facial recognition for surveillance for five years. The FTC’s account describes a deployment running from 2012 to 2020 across hundreds of stores, matching shoppers against a persons-of-interest database assembled from security cameras, employee phone photographs and news stories. The system generated thousands of false matches, and staff acting on false alerts followed, searched and ejected customers, called police and publicly accused people of theft, the FTC says sometimes in front of friends or family. The order also found that the technology was more likely to generate false positives in stores located in plurality-Black and Asian communities than in plurality-White communities, and required deletion of the collected images and the algorithms derived from them, consumer notification, a comprehensive data-security programme, annual certification by the chief executive and independent third-party assessments.

No retailer has published a Wi-Fi 6 outcome

The gap in this field is specific and worth naming. No retailer identified for this article has published a measured business result attributable to Wi-Fi 6 or Wi-Fi 6E, such as a conversion uplift, a dwell-time change or a stock-accuracy improvement. Suppliers publish their own examples; independently reported figures do not exist. The same applies to return on investment for in-store edge computing, where the available numbers come from suppliers.

The older precedent explains why the binding constraint may not be technical at all. Fox 13 Seattle reported that Nordstrom ran Wi-Fi smartphone-tracking sensors supplied by Euclid in Seattle-area stores from 2012, and stopped in 2013 after the practice became public and customers complained. A spokeswoman, Tara Darrow, said it had been a test and that no data was retained. Between that and the FTC’s Rite Aid order, the pattern across more than a decade is that store-side sensing is limited less by bandwidth than by what customers will tolerate and what regulators will permit.

Sources: Walmart Global Tech · Chick-fil-A Tech Blog · Fierce Network · NPR · Retail Dive · Federal Trade Commission · Fox 13 Seattle