A London startup is betting AI’s next big layer is the supermarket, not the data centre

The Series A+ round, reported by Tech Funding News, came from Rank Ventures and Mangrove Capital, a backer since Edgify’s 2020 seed. It takes total funding to $25 million. EU-Startups pegs the raise at €7.7 million.

The pitch is edge AI. Edgify’s software runs across a store’s existing self-checkouts, cameras, scales and tills, whatever the brand. Each device trains a model on the spot and shares what it learns across the shop, without raw video leaving the building. On the company’s own numbers, it spans 2,042 stores and 9,437 devices.

Loss prevention is the wedge

The first job is catching theft. At the self-checkout, Edgify’s vision models flag scan avoidance, product switching and items that never reach the scanner. They also recognise loose produce that carries no barcode. If one till learns a new trick, that lesson spreads to every other till on the network.

That local-learning method is federated learning, and it is the hard part. Cameras in different shops see different products, light and shoppers, so local models can drift apart. Edgify’s approach keeps them pulling toward one shared model without pooling the footage. It also cuts cloud bills and latency, and keeps customer data inside the store.

“A store is a fleet of machines that can see, decide and learn together, without a single byte leaving the building,” said chief executive Nadav Israel. “Retail is just the beginning.”

A bet on physical AI

Edgify grew out of a failure. Israel’s earlier venture, Pixoneye, analysed phone photo galleries and never found its market. Its engineers repurposed the on-device technology for shops, and Edgify launched in 2019.

It is not alone. Trigo, AiFi and Everseen all sell retail vision systems. Edgify’s claim is that it skips the dedicated servers and long installs rivals need, using spare compute on hardware stores have already bought. Whether that is a lasting edge or just a cheaper on-ramp, no one has proven it at scale yet.

The round joins a European run at physical AI, the same wager behind a wave of startups wiring up factories and warehouses. It also cuts against the cloud. As the data-centre buildout swells, backers of edge AI argue the next value pool sits out where the data is born.

Rank Ventures’ Rajan Dosanjh put it plainly: the winners will own the point where data is created. The retail computer-vision market they are chasing is worth about $15.8 billion.