What the AWS Panorama shutdown teaches safety technology buyers

On May 20, 2025, AWS announced that Panorama, its service for running computer vision on an on-site appliance, would stop accepting new customers and would end on May 31, 2026 ^[1]. AWS said that after that date, applications that try to access Panorama would no longer work, and Panorama devices would no longer work either ^[1]. Google followed a similar path: its documentation says Vertex AI Vision was deprecated on June 15, 2026 and reaches end of life on September 30, 2026 ^[2].

Neither product was a safety system as such, but both were the kind of platform that video analytics run on. The lesson for safety buyers is that the edge hardware and the contract matter as much as the detection features. This article sets out what happened, where the large cloud providers went next, and what to ask before you sign for any camera-based safety system.

What happened to Panorama customers

Video safety analytics usually need an edge device: a computer on site that takes the streams from existing cameras and runs the models locally. The Panorama appliance was that device, and AWS describes it as based on the NVIDIA Jetson Xavier platform ^[1].

When the service ended, the appliance stopped working with it ^[1]. The AWS migration guide asks customers to do three things: replace the hardware, migrate the application off the Panorama service, and set up their own edge management and security ^[1]. For the hardware, AWS suggests an off-the-shelf device on a current NVIDIA Jetson platform or, for sites with many cameras, an edge server ^[1]. Applications have to be rewritten to remove Panorama-specific API calls ^[1].

For a safety team, that means a working detection system became a replacement project with a fixed deadline, driven by a supplier's product decision rather than by anything on site. Google's notice for Vertex AI Vision likewise points customers to other Google services to migrate to ^[2].

Where the large cloud providers went

The cloud providers have not left video AI. They now supply the parts that other vendors build on.

• Blueprints. In May 2025 NVIDIA released its AI Blueprint for Video Search and Summarization, a toolkit for building video analytics agents. NVIDIA says manufacturers and warehouses use such agents to help increase worker safety ^[3].
• Reference designs. In April 2026 the AWS Architecture Blog published a design for detecting missing PPE (personal protective equipment) and housekeeping problems from camera images ^[4]. The authors say it blurs faces before analysis and scales to more than 10,000 cameras, a figure they validated by processing 10,000 images at the same time ^[4]. These are the authors' own claims.
• Models inside EHS suites. In December 2025 Cority launched Cortex AI, including an agent that classifies incident images, and announced a collaboration with Google to integrate Gemini models ^[5].

Analysis: more safety products will be built on the same few models and toolkits, so detection quality alone will separate vendors less over time. Workflow, validation on your sites and integration with your EHS (environment, health and safety) system will matter more. A published reference design also gives large employers with their own data teams a credible build-it-yourself option to bring to a negotiation.

The edge hardware is changing too

The chips inside edge devices are also moving. NVIDIA's Jetson Thor developer kit went on sale in August 2025 at $3,499, and NVIDIA says it delivers up to 2,070 FP4 teraflops ^[6]. Investing.com reported that Microchip completed its acquisition of the edge AI chip maker Hailo on September 21, 2026, with terms not disclosed ^[7]. Analytics also increasingly run inside cameras: Ambarella, which makes camera chips, reported fiscal 2026 revenue of $390.7 million, and its chief executive said edge AI chips made up 80% of revenue ^[8].

For buyers, the point is not which chip is best. It is that the box on your network has its own supply chain, and a supplier change can force a refresh. Ask what silicon the edge device uses, when the vendor expects to replace it, and who pays.

Camera platforms now bundle basic detections

At the same time, the camera and video management platforms you may already own are adding analytics:

• Axis says AXIS Object Analytics is preinstalled at no extra cost on compatible cameras and now includes hard hat detection (beta) ^[9].
• Hanwha Vision's AIB-800 AI box adds detection of people and vehicles, line crossing and loitering to up to eight existing cameras that support the ONVIF or SUNAPI protocols, and it works with Milestone, Genetec and Wisenet Wave video management software ^[10].
• Milestone Systems reported net revenue of DKK 2 billion for 2024 and acquired BriefCam analytics that year ^[11].
• Verkada said in December 2025 that it had passed $1 billion in annualized bookings, at a $5.8 billion valuation ^[12].

Before paying for a specialist subscription, price basic PPE and zone detection against what your camera maker or video platform already includes. Specialist vendors are easier to justify for things these platforms do not do, such as vehicle and pedestrian interaction analysis, ergonomics, or feeding findings into your EHS workflow (see how AI video safety works).

Pricing, expansion and switching costs

Specialist pricing is hard to compare from public pages. Intenseye's homepage describes one subscription per camera with every use case included, without a public price ^[13]. Visionify lists an edge server, a Mac mini that handles up to 10 cameras, at $1,299, and sends the subscription price by quote ^[14].

Vendors with installed hardware earn much of their growth from existing customers. Blackline Safety, which sells connected gas detectors and lone worker devices, reported fiscal 2025 revenue of $150.5 million (Canadian), of which $90.5 million was service revenue, and net dollar retention of 128% in the fourth quarter ^[15]. Samsara reported $2.125 billion in annual recurring revenue in its second quarter of fiscal 2027, and said customers with more than $1 million in recurring revenue accounted for over $500 million of it, growing over 50% ^[16]. Analysis: once devices are installed and integrated, adding cameras, sites or users is where the vendor has pricing power, so fix those prices at the start.

Ownership changes are platform risk too

A product can also change because its owner changes. Verdantix counted more than 50 transactions in EHS software in the two years to January 2023 ^[17]. Sphera, for example, was sold by Genstar Capital to Blackstone in a 2021 deal valuing it at about $1.4 billion ^[18], and in 2022 Apax made a cash offer for EcoOnline, whose largest shareholders included funds advised by Summa Equity ^[19]. A new owner may change roadmaps, pricing or support. Our article on private equity in EHS software covers the contract terms in more detail.

Edge architecture and portability

The table below compares common ways to run video analytics. The risk ratings are our assessment, not measurements.

Questions to ask before you sign

• Edge dependence. Who owns the edge device? Does it need the vendor's cloud to keep running? Can it run other software, and what happens to it if the vendor ends the service?
• Model dependence. Which AI models does the product use, who hosts them, and what changes if the model provider changes its terms or price?
• Notice. How much notice will you get of an end of service or a change of control, and what migration help is included?
• Data. Can you export alerts, images, video clips and reports in open formats, during the contract and at the end?
• Price. What does each added camera, site or user cost over three years?

For a fuller list of vendor questions, see questions to ask AI safety vendors.

What to do next

• Ask each current and shortlisted vendor for a written answer on what happens to the edge device and your data if the product is discontinued.
• Check which detections your existing cameras or video management platform already include before renewing or buying a specialist subscription.
• Add data export in open formats, termination assistance and notice of end of service or change of control to your contract template.
• Fix per-camera and per-site expansion pricing in the first contract.
• Prefer edge hardware you own or could reuse, and record the device model and support end date in your asset register.