Two cloud vision platforms, eight years, both gone
AWS Panorama was launched in 2020 as Amazon's flagship answer to the problem of running computer vision on physical sites. It promised a managed appliance, a developer toolkit, and a path from camera feed to operational signal that ran through the AWS cloud. The platform reaches its end-of-life in May 2026, and customers who built on it have spent the last several months migrating their workloads to whatever infrastructure they can stand up before the shutdown date.
Azure Video Analyzer, Microsoft's parallel attempt at the same category, was retired in 2022. The official documentation now directs customers to alternative architectures that the platform was originally meant to spare them from having to build. The two largest cloud platforms in the world both tried to own the vision deployment layer for physical operations, and both have stepped back from the category within the same window of three or four years.
Why the cloud was never going to own the vision layer
The cloud is the wrong place to run real-time perception across a large camera estate, for reasons that became visible as soon as customers tried to deploy at scale. Streaming continuous video from every camera at every site to a cloud region is expensive in bandwidth, slow in latency, and structurally incompatible with the data sovereignty obligations that regulated industries carry. The unit economics of running foundation-scale GPU inference in the cloud against thousands of camera feeds never made sense, regardless of which hyperscaler was billing for it.
The cloud platforms were built on the assumption that compute should be centralised and that data should travel to the compute. Physical operations have the opposite shape, because the cameras are distributed across many sites, the footage stays local for regulatory and operational reasons, and the right place to run perception is on hardware that is already operating on those sites. The hyperscalers eventually arrived at the same conclusion, which is why both flagship platforms are either gone or going.
Where the deployment layer actually belongs
Manako deploys Vision Agents directly onto the commodity computing hardware that operators already run on each of their sites, alongside the cameras that already record there. There is no streaming of continuous video to a cloud region, no GPU rental bill that scales with the number of camera feeds, and no data sovereignty exposure for the operator to defend to the procurement team.
The Vision Agent watches the relevant feeds locally, writes a structured event when it sees a defined situation, and delivers that event through the channels the operations team already uses, including Slack, WhatsApp, Telegram, and email. The footage never leaves the premises, and the operational signal flows directly into the dispatch, ticketing, or compliance systems that the operator was already running before Manako arrived.
Three to four times less than the GPU cloud bill
Vision Agents are sub-fifty-megabyte specialist models that run on commodity CPU hardware, which means the underlying compute cost per camera is a fraction of what running a general-purpose foundation model in the cloud would cost over the same period. For an operator with several hundred cameras across several dozen sites, the difference between the two cost structures is the difference between a deployment that scales to the entire estate and one that stays trapped on a pilot site for budgetary reasons.
The lesson the hyperscalers learned for everyone
The hyperscalers spent several years and substantial engineering effort on the assumption that cloud-hosted vision was the right architecture for physical operations. Both of the largest platforms in the world reached the same conclusion, which is that the deployment layer for vision needs to sit on the operator's site, on the hardware the operator already owns, with footage that never leaves the premises and a cost structure that does not scale with cloud egress bills.
Manako provides that deployment layer on a single platform that runs on any camera across any environment.
Tell Manako what you need. It does the rest.
Join the waitlist to be the first to build with Manako, or get in touch to discuss an enterprise deployment across your operation.