Back to Blog
Edge AI
2026-06-22
6 min read

Your "edge AI" might just be the cloud with extra steps

Dev Sanghvi
Founder & CEO, DHI

A cloud giant just walked away from the edge

In June 2026, AWS discontinued Panorama, its appliance for running computer vision on existing cameras. If you built your analytics on it, the box on your wall is now on a migration clock you didn't plan for.

This isn't about piling on AWS. Big platforms retire products that don't hit their numbers, and that's normal. What's worth paying attention to is what the shutdown quietly exposes about how most "edge AI" is actually built.

Most "edge AI" isn't really at the edge

The phrase has come to mean almost nothing. A box sits on your site, so it gets called edge. But for a lot of products, that box is leashed to a vendor's cloud for the parts that actually matter: provisioning, model updates, and the license check that decides whether it's allowed to keep running at all.

When the connection to that cloud changes — a price increase, an outage, a discontinued product — the box on your wall changes with it. You didn't buy a capability. You rented one, and the landlord can leave whenever they like.

That's the real lesson of any edge product getting switched off from headquarters. The hardware was never the thing you owned.

The one question that separates real edge from rented edge

If you're evaluating AI for your cameras, there's a single question that cuts through every demo: if my internet dropped for a week, or the vendor folded tomorrow, does the thing on my wall still work, and does my footage still stay mine?

For a surprising number of products, the honest answer is no. The video was quietly traveling to someone's cloud the entire time, and the "edge" was a sticker on a box that's useless without a connection home.

Why we built Dhi the other way

Dhi runs detection locally, on the cameras you already have. The model runs next to the lens, on-site. Raw video never leaves the premises. The only thing that travels is the event itself — the small signal that says someone's on the track, or standing in a restricted zone — never the footage.

That choice has three consequences that matter in the real world.

It keeps working when the internet doesn't

A safety system that goes blind during an outage isn't a safety system. Because detection happens on-site, a dropped connection doesn't stop Dhi from watching. The alerting can degrade gracefully; the seeing doesn't stop.

Your footage was never anywhere else to lose

The most secure version of your video is the one that never left the building. When raw footage stays on-prem and only events leave, an entire category of risk simply doesn't exist. There's no stream of your operation sitting on a third party's servers waiting to become someone else's breach. Dhi is SOC-2-ready and GDPR-aligned by architecture, not by policy promise.

You're not betting your safety on someone's roadmap

When the capability lives on your own hardware and runs on your own site, a business decision three time zones away can't switch it off. Owning your safety system should mean actually owning it, not leasing it until the next end-of-life email.

"Edge" should be a guarantee, not a label

The Panorama shutdown will send a wave of teams looking for somewhere to migrate. The temptation will be to find another cloud-tethered box with a nicer dashboard and call it edge again.

The better move is to ask the ownership question first. Where does the video go? What happens when the connection drops? Who can turn this off, and from where? The answers tell you whether you're buying infrastructure you control or renting a feature you don't.

Dhi runs on the CCTV and the VMS you already have — Genetec, Milestone, Avigilon — so finding out doesn't require ripping anything out. Start with one camera that matters, and see what real edge looks like.

Edge AICloudData OwnershipArchitecture
Share this post: