London-based Edgify has raised $9 million in Series A+ funding to expand its edge artificial intelligence platform, bringing the company’s total funding to $25 million.
The round was backed by Rank Ventures and Mangrove Capital Partners. Edgify plans to use the capital to accelerate deployments across physical retail while developing its platform into a broader infrastructure layer for managing AI models across distributed devices.
The company has initially found traction in grocery retail, where computer vision can be used to recognize products, identify unscanned items, detect barcode switching, and reduce losses at self-checkout. But the larger ambition behind the funding is to provide a common AI layer connecting devices such as cameras, scales, scanners, self-checkout terminals, and point-of-sale systems.
Rather than continuously sending raw data to centralized cloud infrastructure, Edgify is designed to run, update, and manage AI models directly on edge devices.
Turning Existing Store Hardware Into an AI Network
Physical retailers increasingly operate large networks of intelligent or semi-intelligent devices, but those systems are often deployed independently.
Edgify is attempting to connect that hardware into a coordinated machine learning environment.
Its software can integrate with existing point-of-sale systems and self-checkout machines through APIs, while computer vision models can run on hardware including standard USB cameras, scanners, and scales. The company says its models can also continue learning from data generated during normal store operations rather than depending entirely on periodic centralized retraining.
That architecture is particularly relevant to computer vision because video and image data can be expensive to continually transmit and process in the cloud. Processing information closer to where it is generated can reduce network requirements while also allowing decisions to be made with lower latency.
Edgify says raw customer data can remain within the retailer’s perimeter, an approach that may also make edge processing attractive in environments where privacy or data residency requirements make centralized data collection problematic.
The company currently lists 2,042 stores, 9,437 connected devices, and more than 400 million samples across its platform. Those figures are self-reported by Edgify.
Loss Prevention Provides the First Major Use Case
Retail loss prevention gives Edgify a relatively straightforward starting point because computer vision can be tied directly to specific events occurring at checkout.
At self-checkout terminals, the platform can identify discrepancies between an item and the barcode being scanned, detect products placed into the bagging area without being scanned, and alert employees or trigger an automated intervention.
Its product-recognition models can also identify fresh produce and other items without traditional barcodes. Edgify says this technology can operate directly within existing scanner and scale infrastructure rather than requiring retailers to replace their checkout hardware.
The same computer vision infrastructure can be extended beyond theft prevention. Edgify already lists waste management as another application, using cameras to identify discarded products and improve inventory records.
The new funding will support further expansion of this model lifecycle infrastructure, including how models are trained, deployed, monitored, and updated across fleets of retail devices.
Federated Learning Tackles a Harder Edge AI Problem
One of the more technically interesting aspects of Edgify’s approach is its work around federated learning, where multiple devices contribute to improving machine learning models without requiring their underlying datasets to be centrally pooled.
The problem becomes more complicated in physical environments because data generated by different devices is rarely identical. A camera inside one supermarket may encounter different products, lighting conditions, shoppers, layouts, and behaviors than a camera operating hundreds of miles away.
Edgify’s research has explored this problem of federated learning on non-independent and identically distributed (non-IID) data, where locally trained models can begin moving in different directions. Its researchers proposed a method designed to encourage those local models toward a shared optimum without adding additional privacy risks or significantly increasing communication requirements.
In practical terms, this type of distributed training architecture could allow a network of edge devices to collectively improve without requiring every image or transaction to be uploaded into one enormous centralized training dataset.
That could become increasingly important as AI moves beyond data centers and begins operating across cameras, industrial sensors, vehicles, robots, machines, and other physical infrastructure.
From Retail AI to Physical AI Infrastructure
Edgify ultimately sees retail as an entry point rather than the limit of its technology.
The company is targeting additional environments including quick-service restaurants, distribution centers, apparel retail, logistics, manufacturing, transportation, and warehouse operations.
Those sectors share many of the same technical constraints. They generate significant amounts of local data, often rely on existing hardware that cannot easily be replaced, and require AI systems capable of responding in real time even when cloud connectivity is limited or expensive.
This is where the longer-term implications of Edgify’s platform become more significant.
Much of the current AI infrastructure boom has centered around centralized computing, with increasingly powerful models running inside hyperscale data centers. Physical AI creates a different infrastructure problem. A warehouse camera, checkout scanner, factory sensor, or autonomous machine may need to make decisions immediately, continuously, and potentially without sending its entire stream of data somewhere else first.
Edgify’s edge-first architecture represents one approach to solving that problem: distribute AI execution across the devices already operating in the physical world, while maintaining a management layer capable of coordinating models across the network.
The company describes its broader objective as making AI practical wherever real-world decisions are generated, rather than requiring those decisions to depend on constant communication with centralized infrastructure.
The Next Battleground for AI May Be at the Edge
The $9 million round is relatively modest compared with the enormous capital flowing into AI model developers and data-center infrastructure, but Edgify is operating in a different part of the stack.
As computer vision and increasingly sophisticated AI models spread into retail stores, factories, warehouses, transportation networks, and other physical environments, organizations will need ways to deploy and maintain those models across potentially thousands or millions of heterogeneous devices.
That creates a new machine learning operations challenge: not simply training a model, but deciding where inference happens, where training data remains, how models learn from distributed environments, and how updates are coordinated across entire fleets of machines.
Retail gives Edgify a demanding environment in which to solve those problems. If the underlying platform can generalize beyond checkout and loss prevention, the larger opportunity could be providing part of the orchestration infrastructure required as AI moves out of the cloud and deeper into the physical world.
