Ashish Chand
Analyst · Truist Securities
Yes. Well, I think this is a very exciting topic. So I'm going to start -- bear with me, I'm going to start with a little bit of basic information and then build it up. So, as we think about AI for the last 3 to 4 years, the first 2/3 of that journey has been more around Chatbots really. And then over the last, let's say, 1 year or so, we are now seeing the whole phenomenon around agents. But a lot of those agents still exist in the digital world, right, inside a data center. Now those agents are emerging into the physical world, and they need a fair amount of orchestration. And those agents could take the form of robots, humanoids, different kinds of equipment, AGVs, et cetera, et cetera. So really, the idea that in workplaces, whether they are manufacturing workplaces or other workplaces, you might have employees that are human and employees that are actually agents working together. You might even have agents and agents working together, right? So that's the kind of future workplace scenario. So, the announcement we -- you, I think, are referring to was actually made as a combination of NVIDIA, Accenture and Belden. And there was a different announcement, I think, about the gray space, which is also relevant, but let me focus on the first one. So, we announced the successful completion of a pilot and we are on the cusp of commercializing this with a very large automotive customer in the U.S., but this was essentially a virtual safety fence application. And it leveraged a few things from each of us. So from Belden at the core, it was the time-sensitive networking portfolio. And I just want to differentiate time-sensitive networking, which is very prevalent in the high-end mission-critical spaces like industrial manufacturing or process is different to the conventional best effort networking, which is more relevant in enterprise spaces, right? So, we used our time-sensitive networks. We use Belden Horizon as the orchestration platform. Accenture built an application on top of Horizon that took that data into the NVIDIA Omniverse and used their libraries to build this entire autonomous system for virtual safety. And I think there were some interesting highlights. So first of all, we did not have any data going to the cloud for safety. Everything was on the site on the edge so that it was very low latency. Now the data stream itself was raw video from a camera versus thousands of sensors on the floor, right? So, it was a camera feed. In fact, it was 3 cameras. So, there was kind of triangulation and spatial depth created in that process. And just from the feed of 3 cameras, this autonomous system was able to analyze and review that data and really as a human being, act as a traffic cop for safety. And I think the third thing this did was it basically removed any ambiguity that network is actually the fourth critical technology to make this digital transformation successful. The other 3 being AI, data engineering and cloud. And although in this case, we didn't send any data to the cloud, obviously, over time, models have to be trained on the cloud. And so data will go to the cloud, but it will be selective data. Now in terms of scaling, there are different studies available. There's one that says -- that's pretty prominent that says that in about -- by 2030, so let's say, in 5 years, the number of physical devices that need network connections will reach close to 1 trillion IoT connections. So 4 to 5x of what we have today. And all of these will need some kind of edge compute capability because all the data will not go to the cloud. And I can easily see another aspect here that, that data will be multimodal. It will be vision, sound, vibration, temperature, pressure, et cetera. Remember, now we have agents in the physical world who are dealing with these kinds of data streams, right, different kind of variety and volume of data versus simple digital data in the data center. And the applications that we are currently exploring or actually piloting include quality inspection, passenger safety, asset location and these go across a few different vertical markets. So sorry, we'll give you a long-ish answer because I want to start with the fundamentals, but the core finding for us here is that without time-sensitive networking and without an orchestration platform like Belden Horizon, it is very difficult to make that edge and IT/OT convergence convert into Physical AI. And I think that's what we've successfully proven here. We are being obviously modest in terms of where all this can go, but really, there is no limit.