Latest / The Edge Computing Podcast with Fexingo: Local Compute, CDNs, and Distributed Infrastructure / How Edge Computing Is Transforming In-Store Augmented Reality
Transcript
- Lucas: So you walk into a hardware store looking for a new kitchen faucet, and instead of guessing whether it'll fit your sink, you point your phone at the counter and see a 3D model of the faucet sitting there, scaled to the actual space. That is the promise of in-store augmented reality, and for a long time, it has been a demo that never quite worked in practice. Luna: Right, because the phone has to render that model in real time, and if the server is off in some cloud region, the latency kills the illusion. Lucas: Exactly. The threshold for convincing AR is generally under 20 milliseconds of motion to photon latency. Anything above that and the virtual object feels disconnected from the real world. Cloud rendering adds 30, 40, sometimes 100 milliseconds just in round-trip time. Luna: So the logical fix is to push the compute closer to the user. That is edge computing 101, but applying it to retail AR is actually pretty recent. Lucas: And it is happening now. One of the larger home improvement chains in the US — I am not going to name them because their deployment is still rolling out — has been testing in-store AR using local edge nodes. Each store gets a small compute box, in this case an NVIDIA Jetson AGX Orin, running a render engine that serves the AR content to customer phones over the store's Wi-Fi. Luna: So the phone is basically just a display. The heavy lifting — the 3D model, the lighting, the occlusion — happens on the Jetson right there in the store. Lucas: Exactly. And the results so far are promising. The chain measured average latency of 12 milliseconds in the stores that have the edge nodes, compared to 55 milliseconds when the same app was running off a cloud AWS region in us-east-1. That is the difference between an AR experience that feels natural and one that makes people nauseous. Luna: Yeah, and 55 milliseconds is actually optimistic for a lot of stores. If the connection is congested, it gets worse. But 12 milliseconds — that is well under the 20 millisecond threshold. Lucas: The other advantage is cost. Rendering 3D models in the cloud is expensive. Each session eats GPU time, bandwidth, and egress fees. By moving that to an edge device that is already paid for and sits in the store, the chain estimates they cut per-session compute cost by roughly 60 percent. Luna: And the edge node can handle multiple customers simultaneously? Because you don't want one person using AR to slow it down for everyone else. Lucas: The Jetson AGX Orin can handle up to 20 concurrent AR sessions before you start seeing degradation. Most stores at any given moment have maybe three to five people using the feature. So headroom is fine. They are also using a technique called foveated rendering, where the edge node renders the full 3D scene but prioritizes the area the user is looking at, reducing the compute load without the user noticing. Luna: Smart. So the user gets a high-quality experience, the retailer saves money, and the whole thing is actually feasible now. What took so long? Lucas: Two things. First, the hardware only recently got small and powerful enough. The Jetson AGX Orin came out in 2023, and before that you needed a full server rack to get similar performance. Second, the software stack for edge AR — things like the ARKit and ARCore integrations, the 3D asset pipelines — those have matured a lot in the past three years. Luna: And the Wi-Fi 6 and 6E rollout in stores helps too. You need reliable local network throughput to stream those rendered frames to the phone without jitter. Lucas: Right. The chain we are talking about upgraded their store Wi-Fi to Wi-Fi 6 specifically for this use case. It is not just the edge node; it is the whole local infrastructure. But the payoff is that customers who used the AR feature spent an average of 23 percent more time in the aisle and had a 14 percent higher conversion rate on the products they visualized. Luna: Those are real numbers. And they come from a controlled test over 12 weeks across 20 stores, I assume? Lucas: Exactly. They compared stores with the edge AR setup to a control group of similar stores without it. The uplift is statistically significant. This is not a lab experiment; it is a live retail deployment. Luna: So the business case is there. But I imagine the upfront cost of putting a Jetson in every store is not trivial. What is the payback period? Lucas: They estimate around 14 months, based on the cloud cost savings and the lift in conversion. Each Jetson unit costs roughly $2,000. The Wi-Fi upgrade is a capital expense they were already planning for other reasons. So the incremental cost is mainly the edge hardware and the software integration. Luna: And after 14 months, it is pure savings plus the revenue upside from better customer experience. That is a compelling pitch. Lucas: It is. And it opens the door for other applications. Once you have an edge compute node in a store, you can run other workloads on it — inventory scanning, shelf analytics, even local AI for customer service chatbots that don't need to phone home. The hardware is versatile. Luna: Quick honest thing — a handful of listeners chip in monthly through buy me a coffee dot com slash fexingo, and that is literally what funds making this many episodes. We keep this show ad-free and focused on the actual tech, not the sponsors. If the conversation gave you something usable, that link is there. No pressure. Just good to know it matters to some people. Lucas: Yeah, and we really appreciate those who do. It lets us spend time on stories like this one instead of chasing ad reads. So thank you. Luna: Anyway — back to the edge node. So you have this compute box sitting in the store. What else are they running on it? Lucas: Right now, just the AR renderer and a lightweight inventory analytics model that detects when shelves are low on stock. They are also testing a real-time planogram compliance check — making sure products are placed where they are supposed to be. That runs as a separate container on the same Jetson. Luna: So the edge node becomes a multipurpose platform. That changes the ROI calculation a lot. It is not just an AR investment; it is a store intelligence investment. Lucas: Exactly. And that is the direction a lot of retailers are heading. The hardware is cheap enough that you can put one in every store, and the software ecosystem is mature enough that you can deploy new applications over the air without touching the hardware. Luna: What about security? You have a device in a store running customer-facing AR, inventory data, potentially camera feeds. Is that a concern? Lucas: It is, and the chain took it seriously. The Jetson runs a hardened Linux build, all data is encrypted at rest and in transit, and the device only communicates with the store's local network and a management server — no direct internet access. Any software update has to be signed by the retailer's own key. Luna: So it is locked down. That matters when you are dealing with customer location data and product placement info. Lucas: Right. And from a privacy standpoint, the AR app only processes the camera feed locally on the phone — it never sends video to the edge node or the cloud. The edge node just sends the rendered 3D frames back to the phone. So there is no video leaving the customer's device. Luna: That is a clean design. And it probably helps with Apple and Google's ARKit and ARCore privacy requirements too. Lucas: Exactly. It aligns with those frameworks. The whole architecture is built around minimizing data exposure while maximizing performance. Luna: So what does this mean for the broader edge computing market? If a big retailer can deploy edge AR at scale and see a real return, other retailers will follow. Lucas: I think we will see a lot of copycat deployments in the next 18 to 24 months. The technology is proven, the hardware is available, and the business case is clear. The main barrier is integration — getting the AR content right, training store associates, making sure the Wi-Fi is up to spec. But those are solvable problems. Luna: And once the edge node is in place, the retailer can add more use cases incrementally. That is the real engine of adoption. Lucas: Exactly. Retail edge computing is not about one killer app. It is about a platform that gets cheaper and more capable over time. In-store AR just happens to be the wedge. Luna: Good way to put it. And with that, we will leave it there for today. Thanks, Lucas. Lucas: Thanks, Luna. And thanks to everyone listening. We will be back next week with another look at how edge is changing the world.