--- title: "Deciding at the edge | Nexus Observatory" description: "A model that detects smoke, flooding or a fault is most useful where the frame is captured. Running inference at the edge cuts the time to a decision and…" url: https://observatory.global/stories/edge-ai/ publisher: Nexus Observatory Inc. updated: 2026-10-31 --- [Stories](https://observatory.global/stories/)System: Edge AI inference # Deciding at the edge A model that detects smoke, flooding or a fault is most useful where the frame is captured. Running inference at the edge cuts the time to a decision and the data on the backhaul; standards for edge computing and benchmarks for edge inference make it comparable12. **Systems**: Edge compute, On-site inference, Event-only backhaul **Toward 2030**: Rules for high-risk AI systems **Figures**: 4, comparison, simulator, systems diagram, timeline to 2030 Go to the references **01** Comparison ## The same street, two architectures. On the left, every camera streams to a distant data centre. On the right, an edge node on the block runs the model and sends only events. The sensors are the same; the decision moves closer and the backhaul carries a fraction of the bytes. Figure as text: Cloud-only and edge inference on one block A split view of the same city block: on the left, links from every pole run to a distant node; on the right, they run to an edge node on the block. **Cloud only**: Every frame crosses the backhaul to a distant data centre. **Edge inference**: An edge node on the block runs the model and forwards only events. **02** Simulator ## Time to decision. Capture, transfer, inference: at the edge, that is the whole budget. Through the cloud, add the backhaul transfer and the round trip. Set the frame size, the links and the model’s inference time; every value is arithmetic on the inputs. Figure as text: Decision latency at the edge and through the cloud Two horizontal bars: the edge decision time in red, shorter, and the cloud decision time below it. | Frame size | 200 kB | | --- | --- | | Uplink | 50 Mbit/s | | Backhaul | 20 Mbit/s | | Round trip to the cloud | 60 ms | | Model inference | 25 ms | | Decision at the edge | 87.0 ms | | Decision through the cloud | 227.0 ms | | Backhaul saved per day at 1 frame/s | 17.3 GB | **03** Systems diagram ## When the backhaul goes. If the backhaul is cut, a cloud-only design stops deciding. An edge design keeps detecting and queues events for later delivery. That property is why edge inference is a resilience choice as much as a latency one. Figure as text: Edge inference with a cloud for training and archive A four-tier diagram: cameras, the access network, the edge node with the model, and the cloud and the operations centre. - **Cameras** : Camera 1, Camera 2, Camera 3 - **Access** : Access - **Edge model** : Edge model - **Operations** : Operations, Training and archive 6 links. Scenarios: Normal operation; Backhaul cut. **04** Timeline to 2030 ## Rules for models that decide. The European Union’s AI Act entered into force in August 2024, with obligations for high-risk systems phased in over the following years3. A model at the edge that informs safety decisions will need its data, its versions and its performance on record. Figure as text: Edge computing and AI governance A timeline from 2016 to 2030 with edge computing standards, inference benchmarks and the AI Act. 1. **2016** Edge computing reference architecture (ETSI GS MEC 003). First edition of the multi-access edge computing framework. 1 2. **2019** Inference benchmarks published. Comparable measurements of inference on edge and data-centre systems. 2 3. **2024-08** AI Act in force (Regulation (EU) 2024/1689). Phased obligations for providers and deployers of AI systems. 3 4. **2027-08** Obligations for high-risk AI fully apply. The last of the Act’s phased application dates. 3 Cloud only Edge inference Figure 1 Comparison Figure 1. Cloud-only and edge inference on one block. A split view of the same city block: on the left, links from every pole run to a distant node; on the right, they run to an edge node on the block. ## References 1. [ETSI GS MEC 003: Multi-access edge computing, framework and reference architecture](https://www.etsi.org/technologies/multi-access-edge-computing) . European Telecommunications Standards Institute, 2022. 2. [MLPerf Inference benchmark suite](https://mlcommons.org) . MLCommons, 2024. 3. [Regulation (EU) 2024/1689 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act)](https://eur-lex.europa.eu/eli/reg/2024/1689/oj) . Official Journal of the European Union, 2024. ## Related stories [System: Private 5G and AI-RAN **Private 5G with the inference in the radio** A private cellular network gives a site its own coverage, its own security and its own priorities. AI-RAN adds a second idea: the accelerators that run the…](https://observatory.global/stories/private-5g-ai-ran/) [Place: Forest and wildfire **Seeing smoke sooner** A fire is cheapest to fight in its first hour. Ridge-top cameras, weather in the stand and satellite hot spots each see it at different times and scales;…](https://observatory.global/stories/wildfire/) [Place: Dense city block **The instrumented street** Four city blocks, read as an observing system: small cells on street poles, air-quality and heat sensors beside them, and an edge node that corrects, times…](https://observatory.global/stories/city-block/) ## Where this applies [Edge compute & AI](https://observatory.global/edge-ai/)[Software](https://observatory.global/software/)[Hardware & power](https://observatory.global/hardware/)[Connectivity](https://observatory.global/connectivity/)[Public safety](https://observatory.global/industries/public-safety/)[Industry](https://observatory.global/industries/industry/)[Cities](https://observatory.global/industries/cities/)[Transport](https://observatory.global/industries/transport/)[Field tour: city block](https://observatory.global/environments/city-block/)[Field tour: wildfire forest](https://observatory.global/environments/wildfire-forest/) © 2026 Nexus Observatory Inc. **Legal** [Privacy notice](https://observatory.global/legal/privacy/) · [Terms of use](https://observatory.global/legal/terms/) · [Cookie policy](https://observatory.global/legal/cookies/) · [Accessibility](https://observatory.global/legal/accessibility/) · [Official languages](https://observatory.global/official-languages/) · [Notices](https://observatory.global/notices/) **Company** [About](https://observatory.global/about/) · [Open governance](https://observatory.global/consortium/) · [Contact](https://observatory.global/contact/)