With the launch of Moonshot AI's Kimi K3 open-weight AI model recently, it’s become easier to envision a future where the cost of AI becomes drastically cheaper, and nowhere is that more interesting to watch than in developments around AI ‘on the edge’. This is an arena where small models can run efficiently on local devices while still performing highly relevant tasks.
Further evidence of just how much this is taking off comes today in the form of London-based Edgify's $9 million Series A+ raise to expand its AI-on-the-edge platform.
Edgify runs machine-learning models directly on retailers’ existing hardware, like a security camera or a POS self-checkout, rather than shipping data back and forth to the cloud, which, in an AI-era of tokens, could rack up high costs.
Rank Ventures and Mangrove Capital Partners backed the A+ round, bringing Edgify’s total funding since its launch in 2019 to $25 million. Edgify had previously raised a Series A round of $8m from Octopus Ventures in Mar 2018. Edgify is an evolution of an earlier startup, Pixoneye, which used AI to understand photos on smartphones.
Edgify’s platform connects hardware already sitting inside stores, including self-checkouts, cameras, weighing scales and point-of-sale systems, and turns it into what the company calls a distributed AI network.
Its pitch is that retailers shouldn’t need racks of new servers or expensive cloud infrastructure to run increasingly sophisticated computer-vision systems.
“The ability to do AI in a really low-cost manner, on-device, where you don't have to go to data centres or buy servers, makes so much more sense,” Edgify COO Mitchell Goldman told Pathfounders over a call.
Edgify has built its own proprietary stack, which, Goldman said, can perform both inference and model training at the edge.
Individual retailers can then run models fine-tuned around their own products and environments.
As odd as this might sound in an era where millions of GPUs are deployed for frontier models, Edgify wants to make use of the potential for ‘compute’ that already exists inside stores.
“Whether it's a self-service scale where you put your bananas or tomatoes, all the way through to a self-checkout, those systems already have a lot of compute inside them,” Goldman said. “That compute is sufficient to run what we believe is AI on the edge.”
Rather than treating every checkout, camera or scale as an isolated endpoint, Edgify effectively pools and orchestrates the available computing capacity.
That means accounting not only for how much compute is available, but also for where devices are physically located, their bandwidth constraints, power availability, and even their thermal limits.
For computer vision, Edgify can also work with cameras retailers already have installed. That might be an IP security camera, a CCTV system, a USB camera, or one incorporated into a checkout itself.
“We act as the bridge between a store surveillance camera and the point-of-sale system,” Goldman said.
Keeping data processing within the store cuts latency and cloud infrastructure costs while allowing sensitive customer and operational data to remain on-site.
Edgify is initially putting that technology to work on one of retail’s biggest problems, where people check out expensive items as if they were cheaper ones (see video above).
The company says its computer-vision systems are already being deployed by grocery retailers in the US and Europe to recognise produce and detect behaviours including scan avoidance, product switching and items leaving stores without being properly scanned.
The company works with hardware providers including Zebra Technologies and Bizerba, allowing its software to operate across equipment from multiple manufacturers rather than requiring retailers to rip out existing systems.
The new capital will be used to expand beyond individual loss-prevention applications and build what Edgify describes as an orchestration layer for AI across the store.
But retail is not where this ends.
“The DNA of what we've achieved is running AI solutions directly on edge devices, low-powered barcode scanners, printers and so on,” Goldman said.
Retail provided Edgify with a relatively forgiving environment in which to develop the technology, he added. Misidentifying one piece of produce for another is considerably less consequential than getting an AI decision wrong in aviation, manufacturing or healthcare.
Having proven the underlying approach, Edgify now wants to apply it to industries where human inspection and legacy processes remain common but are often expensive and time-consuming.
Goldman pointed to aviation as one example.
“You don't need a person to go around an aircraft wing to see if there's a crack in it,” he said. “A camera can do it far quicker, far more efficiently and far more accurately.”
Transportation, logistics and manufacturing are also on the company’s target list.
The global edge AI market size is projected to grow from $46.96 billion in 2026 to $445.75 billion by 2034, exhibiting a CAGR of 32.5%, according to Fortune Business Insights.
Rank Ventures managing partner Rajan Dosanjh said in a statement that Edgify’s attraction to the firm was its ability to train and run AI models on ordinary in-store hardware, with the longer-term opportunity of becoming infrastructure for edge computing.
