The Edge AI Promise Still Comes With a Cloud Asterisk

AI is moving closer to the systems it watches. Andrew Nassar, CTO of IC Realtime, has spent more than 20 years in the surveillance industry and has helped guide the company through the shift from analog cameras to digital recording, remote monitoring, and AI-assisted video analysis.
IC Realtime designs video surveillance systems and software for residential and commercial environments. Its Bunker platform adds AI analysis to compatible camera systems through a dedicated NVIDIA Jetson-powered gateway, placing the main computation at the property instead of sending every task to the cloud.
That design supports contextual alerts and natural-language video search. A compatible camera sends video to the Bunker Gateway, core AI analysis happens on that local gateway, and the system links detected activity to video events before sending an alert with a snapshot and plain-language description.
Bunker Keeps Its Brain Local, Not Its Entire Operation
The user can then review the video in the app. Bunker combines local analysis with cloud-connected services for remote access and video history, creating a hybrid architecture rather than a dramatic cloud-free fantasy.
That distinction matters. Local NVIDIA Jetson processing can cut cloud dependence and speed alerts, but cloud connectivity still supports remote access, subscription services, and video history. It would be misleading to say that no video leaves the property.
Edge AI is not a privacy policy by itself. It describes where core analysis happens; it does not settle what a service stores, what it sends remotely, or how users access past events.
Nassar also refuses to build biometric identity features. His position is that surveillance should describe events rather than profile people, keeping the system focused on what happened instead of attempting to identify who appeared in a camera view.
That boundary gives Bunker a clear purpose: detect activity, connect it to video, explain the event in ordinary language, and let the user inspect the recording. The approach still depends on cloud services, but it avoids treating identity recognition as the inevitable destination for every camera system.
AI Infrastructure Is Expanding Beyond Cameras
The same split between local systems and connected services appears in a separate announcement from IC Manage. The company announced immediate availability of a Model Context Protocol, or MCP, server for IC Manage GDP-AI, allowing semiconductor and system companies to connect their choice of AI agents directly to GDP-AI.
GDP-AI is described as a high-performance design and IP management system and as the single source of truth for near real-time design and IP data across design, IP, and silicon lifecycles. As AI agents take on more engineering tasks alongside designers, companies are building agentic AI workflows for electronic design automation with preferred orchestration frameworks, using MCP as the dominant protocol.
Those workflows need instant design data access so agents can make accurate decisions. The MCP server targets that connection between AI agents and the design and IP information those agents need, rather than forcing companies to abandon their existing agent choices.
The pattern is familiar: keep important processing close to the system, then connect it to cloud services, applications, and outside agents where access adds value. In surveillance, that means a local gateway paired with remote access and video history. In semiconductor design, it means AI agents connected to design and IP data through MCP.
Neither approach removes the need for careful boundaries. A local gateway does not erase cloud exposure, and an AI agent with data access still depends on the quality and control of that data.
That concern sits beside warnings about technology outages. Cybersecurity expert Carmi Levy says there could be more outages, while Offline Research’s X. Eyee says the industry is entering a new era of cyber defense. Separate coverage also discusses cooling computer chips in data centers, another reminder that the systems supporting AI have physical limits as well as software ones.
The takeaway is less glamorous than the usual AI pitch, which is probably healthy. Useful systems need clear data paths, defined privacy limits, reliable infrastructure, and enough restraint to avoid turning every technical capability into a surveillance feature.
Based on
- Andrew Nassar, CTO of IC Realtime – Interview Series — unite.ai
- IC Manage Announces MCP Server for GDP-AI Design and IP Management System | Markets Insider — markets.businessinsider.com
- Tech outages: Cybersecurity expert says there could be more — ctvnews.ca
- We’re entering into a new era of cyber defense says, Offline Research’s X. Eyee — cnbc.com
- Cooling down computer chips – CTVNews — ctvnews.ca




