Latest / The Edge Computing Podcast with Fexingo: Local Compute, CDNs, and Distributed Infrastructure / How Edge Computing Changes the CDN Business Model
Transcript
- Lucas: There's a number I've been turning over in my head all week. It comes from a Cloudflare investor day deck from late last year. They said the average latency drop when a request hits a local edge node instead of a regional data center is about forty milliseconds. Luna: Forty milliseconds — that's basically imperceptible to a human loading a webpage. Lucas: Right. For loading a static image or a JavaScript bundle, forty milliseconds is nothing. But for a real-time video processing pipeline or an autonomous vehicle communicating with a traffic infrastructure node? That forty milliseconds is the difference between a safe stop and a collision. And that's the whole story of why the CDN business model is being rewritten right now. Luna: Walk me through the old model first. CDNs were basically just caching layers, right? Lucas: Exactly. The classic content delivery network — Akamai, Cloudflare, Fastly, Edgio — they made money by caching static content at hundreds of points of presence around the world. You paid them based on bandwidth. How many gigabytes you pushed through their network. It was a volume business. The more data you moved, the more they charged. Luna: And the cloud providers — AWS, Azure, Google Cloud — they had their own CDN offerings too, but they were mostly add-ons to the core compute and storage business. Lucas: Correct. But here's what's changed. The internet is no longer mostly about serving static files — images, CSS, video-on-demand. Increasingly, it's about dynamic, stateful, real-time interactions. Think of live video processing for a gaming stream. Think of an AI inference request for a recommendation engine. Those can't be cached in advance because every request is different. They need compute at the edge. Luna: Which is why Cloudflare Workers and Fastly Compute@Edge exist. They let you run serverless code on the CDN's edge nodes. Lucas: Exactly. And the business model shifts from bandwidth pricing to compute pricing. You're no longer paying per gigabyte. You're paying per CPU millisecond per request. That changes the unit economics completely. For Cloudflare, their Workers product is growing faster than their core CDN business. They reported in their Q1 2026 earnings that Workers revenue grew 85 percent year-over-year. Luna: Eighty-five percent. So the market is voting with its wallet. But what does that mean for the traditional CDN players that don't have a strong serverless platform? Lucas: That's the tension. Akamai, for example, has been slower to adopt this model. They acquired a company called Linode in 2022 to get into cloud compute, but Linode's edge presence isn't as deep as Cloudflare's. Akamai's core business is still heavily reliant on media delivery, which is under pressure from both cloud hyperscalers and from edge-native upstarts. Luna: So the question becomes: can a company that built its entire infrastructure and billing around bandwidth successfully retool for compute? It's a classic innovator's dilemma. Lucas: I think it's even more fundamental than that. It's about where the value is in the stack. In the old model, the CDN was a pipe. In the new model, the edge node is a computer. And if you're a CDN that can only route traffic, you're competing with AWS Lambda at the edge — which runs on AWS's own backbone — and with Cloudflare Workers, which runs on one of the most distributed networks in the world. That's a tough margin squeeze. Luna: Let's talk about a concrete use case. Autonomous vehicles. That forty milliseconds we started with. Lucas: Perfect example. An autonomous car generates terabytes of sensor data per hour. You can't send all that to a central cloud. You need local processing for immediate decisions — obstacle avoidance, traffic light detection. But there are also coordination tasks that benefit from low-latency edge compute: vehicle to infrastructure communication, traffic flow optimization. Some companies are putting edge servers at traffic intersections. Luna: I've read about that. The city of Columbus, Ohio, has a pilot where they deployed edge compute nodes at intersections to process camera feeds locally and adjust traffic signals in real time. They claim a 15 percent reduction in average commute times during peak hours. Lucas: And that data never leaves the intersection unless it's aggregated later. That's a huge privacy benefit too. But the CDN angle here is that these edge nodes are essentially miniature CDN points of presence that also do compute. The same hardware that caches Netflix streams could be running a computer vision model for a traffic camera. Luna: So the CDN of the future isn't just a network of caches — it's a distributed compute fabric. Lucas: That's exactly the phrase Cloudflare uses — 'the global compute fabric.' And Fastly talks about 'programmable edge.' But the question is whether the existing CDN infrastructure is physically suited for this. A typical CDN point of presence is in a data center near a major internet exchange. That's great for low-latency delivery to a city. But for an autonomous vehicle, you might need a node every few blocks. Luna: We're talking about potentially thousands of edge nodes per city, not dozens. Lucas: Exactly. And that's more like a cellular network deployment than a traditional CDN. So the telecom companies — Verizon, AT&T, T-Mobile — they're also entering this space. They already have cell towers everywhere. They can put a server at the base of a tower. Verizon's 5G Edge platform does exactly that. Luna: So the CDN companies are competing with telecoms, cloud providers, and even hardware vendors like Dell and HPE who sell edge servers directly to enterprises. Lucas: It's a crowded field. And the winner probably isn't the one with the most data centers — it's the one with the best software platform for running and managing distributed compute. Because hardware is becoming commoditized. The edge server in a Cloudflare POP is basically the same as the one in an AWS Local Zone. The differentiation is in the orchestration layer. Luna: Which brings us back to pricing. How do you price compute at the edge when the workload might be a five-millisecond function call or a five-minute video encoding job? Lucas: That's the hard part. Cloudflare uses duration-based pricing — you pay per millisecond of execution time, with a minimum charge per request. Fastly uses a similar model but with different tiers. The challenge is that edge compute workloads are incredibly diverse. A single customer might have a mix of real-time inference, batch processing, and static asset delivery. The CDN needs to price all of that competitively while covering the cost of compute at thousands of locations. Luna: And the cloud providers have an advantage here because they can subsidize edge compute with their core cloud revenue. Lucas: They can. AWS Lambda at the edge is priced the same as Lambda in a region — you pay per request and per duration. But AWS has the margins from EC2 and S3 to absorb losses. A pure-play CDN doesn't have that luxury. So they need to be more efficient. Cloudflare, for example, claims their Workers are cheaper than Lambda at the edge for many workloads because they've optimized their runtime — it's based on isolates rather than containers, which gives them better density per server. Luna: Let's shift gears slightly. I want to talk about the impact on network architecture. When you run compute at the edge, you need to think about data locality, state management, and data consistency. Lucas: Right. If a user's session state is stored at an edge node in New York, and they travel to Los Angeles, how do you migrate that state? Or do you always route them back to the same edge? That's the 'sticky session' problem. And it's not trivial at scale. Luna: Some platforms, like Fly.io, solved this by using a global anycast network that routes users to the nearest edge, but they also replicate state across multiple regions using a distributed database. Lucas: Exactly. And that's another area where traditional CDNs have an advantage — they already have the anycast routing. Cloudflare uses anycast for all their traffic. So routing a user to the nearest edge is something they've been doing for a decade. The hard part is adding stateful compute on top of that stateless routing layer. Luna: I think it's also worth noting that not every application needs edge compute. For many enterprises, a regional cloud deployment is still perfectly fine. The edge is for latency-sensitive or data-intensive use cases. Lucas: Absolutely. If you're running a CRM for a sales team, you don't need edge compute. But if you're processing video from a security camera in real time, or running a recommendation model for a retail app with millions of users, the edge starts to make sense. And the CDN companies are trying to capture that middle ground. Luna: There's also the emerging category of 'edge AI.' Running inference on edge nodes rather than in the cloud. That's a huge growth vector. Lucas: It's enormous. NVIDIA has their EGX platform for edge AI, and they partner with CDNs to deploy it. But I think the real disruptor will be when you can run a small language model — like a distilled version of GPT — on an edge node with sub-ten-millisecond latency. That would enable a whole new class of applications: real-time translation, voice assistants that don't need a cloud connection, personalized advertising that respects privacy because the data never leaves the device. Luna: And that's where the CDN business model fully transforms. You're not selling bandwidth anymore. You're selling compute capacity that happens to be geographically distributed. Lucas: Right. And that's why I think in five years, the term 'content delivery network' will feel outdated. We'll just call them 'edge networks' or 'distributed compute platforms.' The content delivery part becomes table stakes. Luna: I want to ask about the economics for the listener. If I'm a startup building a real-time application, should I go with a CDN's edge compute or a cloud provider's? Lucas: It depends on your latency requirements and your data gravity. If you need sub-twenty-millisecond latency globally, Cloudflare or Fastly are your best bets because they have the most points of presence. If you need deep integration with other cloud services — like a database or a message queue — then AWS or GCP might be easier, even if latency is slightly higher. The key is to benchmark your own workload. Don't trust marketing numbers. Luna: And what about pricing? Is there a rule of thumb? Lucas: Cloudflare Workers has a generous free tier — 100,000 requests per day. After that, it's about fifty cents per million requests for the first tier. Fastly's Compute@Edge is a bit more expensive per request but offers lower latency for certain workloads because of their different architecture. AWS Lambda at the edge has a higher per-request cost but integrates with CloudFront. So the answer is: it depends on your traffic patterns. Luna: That's a helpful framework. So the CDN business model is shifting from selling a commodity — bandwidth — to selling a differentiated product — low-latency compute. Lucas: Exactly. And the winners will be the ones that make that compute platform easy to use, reliable, and cost-effective. It's a software platform play, not a network infrastructure play. Luna: Speaking of software platforms, there's a fascinating open-source project called 'OpenEdge' that's trying to create a standard for edge compute. Do you think that will disrupt the proprietary platforms? Lucas: OpenEdge is interesting, but adoption is still early. The challenge is that edge compute requires tight integration with the underlying network. A CDN like Cloudflare controls both the network and the compute runtime. OpenEdge tries to abstract that, but you still need a physical network to run on. I think it's more likely that we see a few dominant proprietary platforms, with open-source being used for internal edge deployments by large enterprises. Luna: So the CDN market is consolidating around a few players, while the edge compute market is still fragmented. Lucas: That's a fair characterization. And the next few years will determine which CDNs successfully make the transition. I'm watching Fastly closely — they've bet everything on Compute@Edge. Their revenue from compute is growing, but their legacy CDN business is declining. If they can cross the chasm, they could be a major player. If not, they might get acquired. Luna: And Cloudflare is the 800-pound gorilla. They have the scale, the developer mindshare, and the financial resources to invest. Lucas: They do. But they're also taking on the cloud providers directly, which is a dangerous game. AWS and Azure have much deeper pockets. It's going to be a fascinating battle. Luna: I think we should also mention the security angle. Running compute at the edge introduces new attack surfaces. A vulnerability in a CDN's runtime could give an attacker access to thousands of customer workloads. Lucas: That's a very real concern. In 2024, there was a disclosure about a sandbox escape in Cloudflare Workers. Cloudflare patched it quickly, but it highlighted the risk. If you're running untrusted code on shared edge infrastructure, you need robust isolation. Cloudflare uses V8 isolates, which are relatively secure, but no system is perfect. Luna: And the flip side is that edge compute can also improve security — you can run firewalls and threat detection at the edge, blocking attacks before they reach the origin server. Lucas: That's the promise of the 'secure access service edge' or SASE model. Cloudflare's Cloudflare One product combines edge compute with zero-trust networking. So the same edge nodes that cache content also run security policies. That's a natural extension of the platform. Luna: We've covered a lot of ground. Let's try to wrap up with a forward-looking thought. What's the one trend you're watching that could change the edge compute landscape in the next two years? Lucas: I'm watching the deployment of edge nodes in underserved regions — Africa, parts of Asia, Latin America. Right now, most edge nodes are in North America, Europe, and a few Asian metros. But companies like Cloudflare and Edgio are building points of presence in places with high mobile penetration but poor cloud connectivity. If they succeed, they could unlock a wave of new applications — mobile payments, agricultural AI, telemedicine — that rely on low-latency compute without needing a full cloud region nearby. Luna: That's a great point. The edge isn't just about speed — it's about access. Lucas: Exactly. And that's the story that doesn't get told enough. The CDN business model is evolving, but the ultimate impact is about bringing compute to where people are. Luna: And on that note, we should mention that this conversation is brought to you by our listeners. 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. Lucas: Yeah, even a couple of dollars a month helps us keep these episodes free for everyone. Thanks to those who already chip in. Luna: Alright, let's get back to it. So you mentioned OpenEdge earlier — do you think there's a role for standards in edge compute, or will it remain fragmented? Lucas: I think we'll see some de facto standards emerge, like the WebAssembly runtime that Cloudflare and Fastly both support. But full interoperability? That's a long way off. The edge is still the wild west. Luna: And that's what makes it exciting. Thanks, Lucas. Lucas: Thanks, Luna. And thanks to everyone listening.