Light is fast, but not infinitely fast. In optical fibre it travels at roughly 200,000 kilometres per second, about two-thirds of its speed in a vacuum. That works out to one millisecond for every 200 kilometres, one way. Send a request from Halifax to a cloud data centre in Montreal and back, and physics alone costs you about eight milliseconds before a single server does any work. Route it through real-world networks, with their detours and switching equipment, and the number climbs. Shrinking that trip is the whole premise of edge computing.
For most of what you do online, that doesn’t matter. Your email can wait 50 milliseconds. A robot arm on an assembly line that needs to stop before it hits a worker cannot. Neither can a self-driving car deciding whether that shape ahead is a plastic bag or a child.
That gap between what the cloud can deliver and what some machines need is the problem the edge is built to solve. Here’s what it is, how it differs from the cloud, where it’s actually used, and what it costs you in complexity.
Edge computing explained in one paragraph
Edge computing means running computing work close to where data is created or used, rather than shipping everything to a big centralized data centre. IBM describes it as a framework that brings applications closer to data sources such as sensors, devices or local servers. “The edge” isn’t one place. It can be a ruggedized server in a factory, a box in the back room of a grocery store, a small data centre inside a telecom carrier’s network, or the computer inside a car. What these spots have in common is distance: they’re metres or a few kilometres from the action, not hundreds.
Edge vs cloud: not a rivalry
It’s tempting to frame edge as the cloud’s competitor. In practice they work as a team, and nearly every edge deployment reports back to a cloud somewhere.
| Centralized cloud | Edge | |
|---|---|---|
| Where it runs | Large regional data centres (AWS has two Canadian regions, in Montreal and Calgary) | On site, in a nearby telecom facility, or on the device itself |
| Typical latency | Tens of milliseconds, depending on distance | Single-digit milliseconds or less |
| Strength | Huge scale, cheap storage, training AI models, analytics across all your data | Real-time response, working offline, filtering data before it’s sent |
| Weakness | Distance, bandwidth costs, depends on the connection | Many small sites to manage, physical security, limited capacity |
A common split: the edge handles the split-second decisions and trims the raw data down to what matters, and the cloud does the heavy lifting, the long-term storage and the model training. A factory camera might inspect 30 frames a second locally and send only the defect images, plus a daily summary, to the cloud.
The latency physics, and why it isn’t the only reason
Latency is the headline reason for edge computing, so it’s worth understanding where delay comes from. A round trip to a server includes the time light takes to cross the distance, time spent in routers and switches along the way, any queueing when networks are busy, and the server’s own processing time. Edge computing can’t speed up light. It shortens the trip.
Some quick arithmetic shows why this matters for machines. A car at 100 km/h covers nearly three metres in 100 milliseconds. An industrial control loop may need to react within a few milliseconds. Even a well-connected cloud region a few hundred kilometres away can’t promise that, especially when a network hiccup adds unpredictable delay.
But latency isn’t the only argument. In practice, three others often matter just as much:
- Bandwidth. High-resolution cameras and sensors generate enormous streams. Sending all of it to the cloud is expensive and often pointless. Statista projects the number of connected IoT devices will roughly double from 19.8 billion in 2025 to 40.6 billion by 2034, which is a lot of data to move.
- Resilience. A mine, a ship or a factory can’t stop because its internet link went down. Microsoft’s Azure IoT Operations, for example, is designed to keep running offline for up to 72 hours, according to its documentation.
- Data residency and privacy. Hospitals, governments and regulated industries may need certain data to stay on site or in a particular jurisdiction. Processing locally can satisfy rules that a foreign cloud region can’t.
Where edge computing is actually used
Manufacturing
Factories are the most natural fit. Machine-vision systems inspect parts for defects in real time, vibration sensors flag a motor that’s about to fail, and production lines track overall equipment effectiveness minute by minute. Siemens sells an Industrial Edge platform for exactly this: run AI models for quality inspection and predictive maintenance right on the shop floor, while connecting the results to cloud and enterprise systems. For Canadian manufacturers in Ontario’s auto corridor or Quebec’s aerospace cluster, this is where edge spending tends to start.
Retail
Stores use local servers to run point-of-sale systems that keep working when the internet doesn’t, along with camera analytics for shelf stock, self-checkout and loss prevention. Video is the big driver: analyzing it locally avoids streaming every camera to the cloud. AWS offers compact 1U and 2U Outposts servers aimed at locations such as retail stores and branch offices, and Microsoft pitches its Azure Local platform for local AI inference in retail.
Autonomous vehicles
A self-driving car is the ultimate edge device. It can’t wait for a data centre to decide whether to brake. Waymo, which by early 2026 was running roughly 3,000 robotaxis and about 500,000 paid rides a week across U.S. cities, packs lidar, radar and cameras into each vehicle along with powerful onboard computers. The cloud still matters: it’s where the driving models are trained on fleet data, and Waymo uses remote human assistants who can advise a car when it’s stuck. But the life-or-death decisions are made in the car.
5G and multi-access edge computing (MEC)
Telecom carriers have their own version, called multi-access edge computing, standardized by the European standards body ETSI. The idea is to put computing inside the mobile network itself, so a phone or connected device reaches an application without its traffic leaving the carrier’s network for a distant data centre.
The best-known product is AWS Wavelength, which embeds AWS compute and storage inside carriers’ facilities. According to AWS’s location list, it’s available in 31 cities worldwide, with partners including Verizon, Vodafone, KDDI and, in Canada, Bell in Toronto. Pitched uses include cloud gaming, live video production and running AI inference for things like medical diagnostics. Here’s the catch: MEC has grown more slowly than the 5G hype of 2019 and 2020 suggested. Wavelength’s footprint is a few dozen cities, not every cell tower, and for many apps a well-placed regular cloud region or content delivery network is close enough.
The products: who sells the edge
The big cloud providers and network companies have each built their own on-ramp. A quick map of what’s available in 2026:
- AWS Outposts puts AWS hardware in your building, fully managed by Amazon. It comes as standard 42U racks (scalable up to 96 racks) running services like EC2, S3 and managed databases, or as small 1U and 2U servers. AWS pitches it for workloads needing single-digit millisecond latency, data residency or local processing, and has introduced second-generation compute racks.
- AWS Wavelength runs inside telecom 5G networks, as described above, and AWS Local Zones place smaller pockets of infrastructure in additional metro areas.
- Azure Local, formerly Azure Stack HCI, runs on customer-owned hardware and is managed through Azure Arc, Microsoft’s control plane for on-premises and multi-cloud resources. Microsoft targets factory floors, stadiums, transit systems and sovereign or disconnected deployments.
- Cloudflare Workers take a different approach. Instead of hardware in your building, your code runs across Cloudflare’s own network, which spans 348 cities in more than 100 countries, including Toronto, Montreal, Vancouver, Calgary, Winnipeg, Saskatoon and Halifax. Cloudflare says 95% of the world’s internet users are within 50 milliseconds of one of its data centres. Workers run in lightweight V8 “isolates” that start about a hundred times faster than a traditional Node.js process, according to Cloudflare’s developer docs, and the company’s Workers AI service now runs open-source AI models on GPUs spread across that network.
Note the two flavours. Outposts and Azure Local bring the cloud to your premises. Workers and similar services (Fastly, Akamai and others offer comparable edge platforms) bring your code closer to your users without you owning anything. Which one you want depends on whether your “edge” is a machine on your property or a customer somewhere on the internet.
The trade-offs nobody puts in the brochure
Edge computing solves real problems, but it creates new ones. Before you buy, weigh these honestly:
- Fleet management is hard. One cloud region is a single thing to patch and monitor. Two hundred store servers are two hundred things, each of which can fail, fall behind on updates or get unplugged by a well-meaning employee. Tools like Kubernetes and Azure Arc help, but they demand skills many teams don’t have.
- Physical security matters again. A server in a back room can be stolen or tampered with. Edge hardware needs disk encryption, secure boot and remote wipe.
- Capacity is fixed. The cloud’s great trick is elasticity. An edge box has the CPU and memory it shipped with. Plan for peaks, or plan to fall back to the cloud.
- Costs move, they don’t vanish. You may save on bandwidth and cloud compute but pay for hardware, power, cooling, site visits and management software. Run the numbers per site.
- Data consistency gets tricky. When the same data lives in many places, keeping it in sync, especially after an outage, is a genuinely difficult engineering problem.
How to decide if you need it
For most software startups and small businesses, the honest answer is that you don’t need your own edge infrastructure. A cloud region in Montreal or Calgary, plus a content delivery network or an edge-functions platform for your website, will get you most of the latency benefit with none of the hardware.
You should take edge seriously if you can answer yes to any of these:
- Does a machine or person need a response in under about 20 milliseconds, every time?
- Do you generate more raw data (usually video or sensor streams) than it makes sense to upload?
- Must the system keep running when the internet connection drops?
- Do regulations or contracts require data to stay on your premises?
If so, start small. Pick one site and one workload, such as a single camera-based quality check on one production line. Measure latency, uptime and cost against a cloud-only version. Choose a platform that lets you manage edge and cloud from the same console, because the operational burden, not the hardware, is what sinks most edge projects.
Edge computing isn’t a replacement for the cloud, and it isn’t a fad. It’s a recognition that some decisions have to be made where the action is, because the speed of light won’t negotiate.
Sources and further reading
- IBM: What is edge computing?
- AWS Outposts product page
- AWS Wavelength locations
- Microsoft Learn: What is Azure Local?
- Microsoft Learn: Azure IoT Operations overview
- Cloudflare Docs: How Workers works
- Cloudflare: Global network
- ETSI: Multi-access Edge Computing
- Siemens: Industrial Edge
- Wikipedia: Optical fiber (signal propagation speed)
