How Edge Computing Enables Real-Time Mixed Reality Collaboration

Lucas: You know, for years the promise of mixed reality has been held back by one stubborn problem: latency. Luna: The kind that makes you feel sick after a few minutes, right? Lucas: Exactly. When your brain sees a holographic object that lags behind your head movement — even by 30 or 40 milliseconds — you get that queasy disconnect. It's the single biggest barrier to making mixed reality a real productivity tool. Luna: And the typical fix has been to throw more compute at the headset itself, but that adds weight, battery drain, and cost. Lucas: Right. So a global architecture firm I've been following took a different approach. Instead of stuffing a supercomputer into the headset, they put edge compute nodes in each office where their teams collaborate. Luna: So they're doing the rendering locally, not in the cloud? Lucas: Yes. They have a dedicated edge rack in their New York office, another in London, and one in Singapore. Each node runs a high-end GPU that renders the shared building model in real time. The headset only does display and tracking — the heavy lifting happens twenty feet away. Luna: And they're collaborating across time zones on the same holographic model? Lucas: Exactly. An architect in New York can tweak a wall, and the change appears in the London model with under 20 milliseconds of sync delay. The local rendering is under 10 milliseconds. Compare that to cloud-based rendering where the round trip is 80 to 120 milliseconds even on a good day. Luna: And that cloud latency is what makes people nauseous. So edge computing is really the enabler here. Lucas: It is. And they're not just using any edge — they're using 5G network slicing to prioritize the MR traffic. The edge nodes are connected via a private 5G slice that guarantees a certain bandwidth and latency. Luna: So the network itself is carved out just for this application. That's such a specific use of slicing. It's not just 'faster internet' — it's dedicated infrastructure. Lucas: Right. And the result is that their design review meetings, which used to happen over Zoom looking at static renders, now happen with a shared holographic model that everyone can walk around and annotate. They report a 40 percent reduction in design iteration time. Luna: That's a huge productivity gain. But I'm curious about the cost. Edge hardware isn't cheap. Lucas: It's not. Each node costs roughly fifty thousand dollars to set up — GPU, CPU, storage, networking. And then you have the 5G slicing contract. But they calculated that the savings from fewer design errors and faster approvals paid for the first node in under six months. Luna: So the ROI is there if you have enough scale. But for a smaller firm, that's a big upfront investment. Lucas: That's the trade-off. But cloud providers are now offering edge as a service — you can rent a pre-configured node in a local data center rather than buying your own. That brings the entry cost way down. Luna: Huh. So the model is shifting. And this all ties back to something we've talked about before: the killer app for edge computing isn't just one thing — it's latency-sensitive use cases that cloud can't handle. Lucas: Mixed reality collaboration is shaping up to be one of those killer apps. And it's not just architecture — I've seen this used in automotive design, medical training, and even live event planning. Luna: The common thread is that you need real-time interaction with a shared 3D space, and that's just not possible with cloud alone. Lucas: Before we go deeper into the technical architecture — and I have some specifics on how they handle state synchronization — I want to mention something that keeps this show going. Luna: Yeah, and it's relevant because we're talking about infrastructure that makes something possible, and our own infrastructure relies on listeners who see value in ad-free content. Lucas: Exactly. We deliberately don't run ads on these episodes — no sponsor reads, no mid-roll interruptions. If you find the show useful and want to support that choice, the link is buy me a coffee dot com slash fexingo. Luna: Even just a one-time contribution helps us keep the servers humming and the episodes ad-free. All right — so back to the synchronization piece. Lucas: So the key technical challenge is keeping the model consistent across three offices in real time. They use a technique called 'operational transformation' — basically, each edge node processes local edits immediately and sends the deltas to the other nodes. Luna: Like Google Docs for 3D models? Lucas: Exactly the same concept. But because the network is a private 5G slice, the deltas arrive within a few milliseconds. Conflicts are resolved using a last writer wins rule, which works for most design changes. Luna: So if two architects edit the same wall at the same time? Lucas: The last change takes effect, and a notification pops up in the headset. It's not perfect for all scenarios, but they found that in practice, conflicts are rare because people naturally focus on different parts of the model. Luna: And the edge nodes are also caching the entire model locally, so if the network goes down, you can still work with a local copy and sync later. Lucas: That's exactly what they do. Each node stores a full copy of the current model version. If the 5G slice drops, the local rendering continues uninterrupted. When the network comes back, the nodes reconcile the changes. Luna: That resilience is a huge selling point. Architects don't want to lose work because of a network blip. Lucas: And it speaks to a broader principle: edge computing doesn't just reduce latency — it provides autonomy. The local node can function independently if the connection to the rest of the world falters. Luna: So in a way, it's like having a mini data center in your office just for this one app. Lucas: Exactly. And that's the direction we're seeing more and more enterprises go — not just for MR, but for any application where real-time interaction is critical. Luna: What about the headset themselves? Are they using a specific brand? Lucas: They standardized on the Microsoft HoloLens 2, but they told me the next generation is going to be device-agnostic using a standard protocol called OpenXR. That way they can switch to whatever headset has the best specs at the time. Luna: Makes sense. So the edge node handles the rendering, the headset is just a display — that decoupling gives them flexibility. Lucas: Exactly. And as more headsets come to market with higher resolution and wider field of view, they can upgrade without touching the backend. Luna: I want to circle back to something you said earlier about the 40 percent reduction in design iteration time. That's a really concrete metric. Do you know how they measured it? Lucas: They tracked the time from initial design review to sign-off on a major building component — like a facade or a floor plan. Before MR, that process involved multiple rounds of static renders, comments via email, and follow-up meetings. With the holographic reviews, they cut the number of meetings by half and reduced the time per meeting by about 30 percent. Luna: Because everyone can see the same thing simultaneously and point at it. Lucas: Right. And the annotations are spatial — you can leave a note attached to a specific beam or wall that appears in the model for the next reviewer. Luna: That's a far cry from a red circle on a PDF. Lucas: It is. And it's made possible entirely by that edge compute node sitting in the corner of the office. Without it, the latency would be too high for a natural collaborative experience. Luna: So the edge node is the unsung hero here. It's not the headset, it's not the 5G slice — it's the local compute that makes it all feel real. Lucas: And that's a important lesson for anyone looking at mixed reality for their business: don't just buy headsets and expect the cloud to handle it. Invest in the edge infrastructure first. Luna: What about security? The model data is presumably proprietary — having it on an edge node on-premises versus in the cloud must be appealing. Lucas: That's another advantage. The edge nodes are in locked server rooms with access control. Data never leaves the building unless it's being synced to other offices over the encrypted 5G slice. For a firm that handles sensitive building designs for airports or data centers, that's a big deal. Luna: So edge computing gives them both latency and security benefits. Lucas: Exactly. And as the cost of edge hardware continues to drop — GPU prices have come down about 15 percent year-over-year — we'll see this model become more common. Luna: I could see this spreading to other industries that rely on 3D models: automotive, aerospace, even film production. Lucas: I've heard of a car company using a similar setup for clay modeling reviews — designers in Detroit and Munich can look at the same holographic car surface simultaneously. Luna: And the edge node is local in each studio. Lucas: Exactly. That's the pattern: put compute where the collaborators are, connect them with a low-latency network, and let the magic happen. Luna: So if someone is listening and thinking about trying this, what's the first step? Lucas: Start with one use case and one pair of offices. Rent an edge node from a cloud provider — they'll help you configure the GPU and the network slice. Then run a pilot for a month with a small team. Measure the time savings and user satisfaction before scaling. Luna: I like that. It's a low-risk way to test the waters. Lucas: And if you do, let us know how it goes. We'd love to feature a listener success story in a future episode. Luna: That's a good note to end on. Thanks, Lucas. Lucas: Thanks, Luna. Talk to you next time.