Latest / The Cloud Business Podcast with Fexingo: AWS, Azure, GCP, and Enterprise Infrastructure / The Edge Computing Reality Check for Enterprise Cloud
Transcript
- Lucas: So a global logistics company — I'm not naming them, but they run hundreds of warehouses worldwide — decided last year that they needed real-time inventory tracking. Every pallet scanned, every forklift move, reflected in the system within a second. Their central cloud, in a major AWS region, was giving them three to five seconds of lag, which was creating bottlenecks at shipping docks. Luna: And that's a classic use case for edge computing, right? Process data closer to where it's generated. Lucas: Exactly. So they went with AWS Outposts. Put a mini data centre—essentially a rack of compute and storage—right in each warehouse, connected to the cloud but running some workloads locally. Sounded great on paper. Luna: How did it actually work? Lucas: Mixed. The latency dropped to under 200 milliseconds, which solved the immediate problem. But they ran into two new ones. First, cost: each Outposts rack, fully loaded with support and software, ran them about one hundred ten thousand dollars a year per warehouse. They have two hundred warehouses. That's twenty-two million dollars annually. Luna: Just for the edge hardware. That's not including the network or the central cloud bill. Lucas: Right. And the second problem was operational complexity. They needed local IT staff—or trained warehouse managers—to handle hardware failures, software updates, security patches. Things the cloud normally abstracts away. So now they had a hybrid team model they hadn't budgeted for. Luna: I've heard that story before from other companies. The edge promise is compelling, but the reality often feels like running a mini data centre in a dusty warehouse. Lucas: And that's the tension. We're in mid-2026 now, and every major cloud provider has an edge offering. AWS has Outposts, Azure has Azure Stack Edge and the new Azure Local, Google has Distributed Cloud. But adoption is still surprisingly low for workloads that aren't absolutely latency-critical. Luna: What does the data say? I remember a Gartner survey from earlier this year—they polled enterprise IT leaders. Lucas: Yeah, the numbers are striking. About eighty percent of respondents said edge computing is a strategic priority. But only fifteen percent said they have a workload in production on an edge platform. That's a huge gap between interest and real deployment. Luna: So what's causing that gap? Is it cost, complexity, or just not finding the right use case? Lucas: All three, honestly. But I think the biggest issue is that many enterprises don't actually need true edge. They need what I'd call 'near-edge'—running workloads in a regional data centre that's closer than their primary cloud region, but still fully managed. Luna: So like using an AWS Local Zone instead of putting an Outposts rack in your building. Lucas: Exactly. For the logistics company, if they had put a Local Zone in each major metro area where their warehouses cluster, they'd have gotten latency under one second—maybe five hundred milliseconds—at a fraction of the cost, and zero on-site hardware management. Luna: But there are use cases where even five hundred milliseconds is too slow. Autonomous vehicles, real-time industrial control, some healthcare imaging. Lucas: Absolutely. And for those, you do need on-premise edge. But I'd argue that's maybe ten percent of the workloads people talk about as 'edge.' The rest can be served by a well-placed regional cloud endpoint. The problem is that providers market all of these as 'edge,' muddying the decision. Luna: So what's the decision framework? When should a company actually pull the trigger on true edge? Lucas: I think there are three tests. First, latency requirement: does the workload break if response time exceeds ten milliseconds? If yes, you likely need true edge. Second, data volume: are you generating terabytes per day that would be too expensive to send to the cloud? Autonomous vehicle fleets generate petabytes—edge processing makes sense. Luna: And the third test? Lucas: Regulatory or data residency constraints. Some industries—finance, healthcare, government—are required to keep certain data within the country or even within a specific facility. Edge can solve that. But for the vast majority, near-edge is probably the smarter choice. Luna: That framework is really helpful. I think a lot of companies get pulled into edge because it sounds cutting-edge, no pun intended, but they haven't done that analysis. Lucas: If today's episode was useful to you and you want to keep it ad-free, buy me a coffee dot com slash fexingo helps. We don't run ads—that's a deliberate choice. If you want to support that, buy me a coffee dot com slash fexingo is the place. Luna: Yeah, it really does make a difference. Even a small contribution helps us keep these conversations going. Lucas: So back to the logistics company—they ended up doing a hybrid. They moved most warehouses to AWS Local Zones, but they kept true edge in three high-volume distribution centres where they were doing real-time robotic sortation. That saved them about fifteen million dollars a year compared to the full Outposts rollout. Luna: And that's the kind of pragmatic approach I think more enterprises need. It's not edge or cloud—it's a spectrum. Lucas: Exactly. And the providers are slowly adapting. Microsoft just launched Azure Local last month, which is essentially a lighter version of Azure Stack Edge—smaller footprint, lower cost, designed for single-site retail or branch offices. And Google's Distributed Cloud has a 'software-only' tier that runs on standard x86 servers. Luna: So the market is clearly moving toward offering more granular options. But that also means the onus is on the buyer to understand what they actually need. Lucas: Right. And I think a lot of them are still in the experimentation phase. The Gartner survey showed that of the fifteen percent with production edge, most were in retail or manufacturing—specific use cases like inventory tracking, predictive maintenance, or video analytics. Luna: What about telco edge? That's been a big push from AWS and Azure, putting compute at 5G base stations. Lucas: That's a separate animal, honestly. Telco edge is more about network function virtualization—replacing proprietary hardware with virtualized software. It's still early, and the business case is murky. Most enterprises aren't going to touch that directly; they'll consume it through a carrier. Luna: So for the typical enterprise listener—head of infrastructure or cloud architect—what's the one takeaway from this episode? Lucas: Run the latency experiment first. Measure your actual workload response time end to end before you decide you need edge. I've seen companies spend millions on Outposts or Azure Stack only to discover that their real bottleneck was a slow database query, not network distance. Fix that first, then consider if true edge is necessary. Luna: Good advice. And if you do decide you need edge, start with a single site pilot, not a hundred. Lucas: Hundred percent. The logistics company pilot was three warehouses, and that's how they learned the cost and complexity before scaling. I think that's the model going forward.