Sean McManus, TelecomTV:
Hi, my name's Sean McManus from TelecomTV. I am at the Open Networking and Edge Summit in London, and I'm joined now by Tilly Gilbert from STL Partners. Now, STL Partners is predicting that the Edge AI revenue is going to be worth $157 billion by 2030. So Tilly, tell us first of all, what's included in your definition of edge computing there?

Tilly Gilbert, STL Partners:
So for me, edge is about everything in between the device and a hyperscale or a cloud data center. So what we're not including in there is on-device compute, the device that's on a mobile phone or on other IOT sensors, or devices. And we're also not including any compute power processing that's going on in the cloud or in other very large hyperscale locations. And we typically break it down into an on-premise edge on the enterprise sites. Network edge, which is typically hosted in technical sites that telecoms operators or tower companies own. Or regional edge, which would be your smaller data centers maybe serving your tier two or your tier three markets or cities.

Sean McManus, TelecomTV:
Great. Tell us what it is that's driving enterprises to increasingly adopt edge compute.

Tilly Gilbert, STL Partners:
So two main things as we see it, and both of them are generally about where the performance of the cloud doesn't quite meet what the enterprise is looking for. So that's generally where edge comes into play. Cloud is often one of the first ports of call that you would make because it's so mature, it's widely available across all markets. It's a well-known model that many use today. So two main reasons where cloud might not work for a particular enterprise or a particular type of application.
One is where the application requires data, which is really significantly heavy. So that's often video footage, although it could be other types of data. And there they might find that both the networking costs, so the cost of transporting that data from where it's been generated to the cloud, may be pretty expensive and maybe your network can't even necessarily handle that without upgrades. And you may find that when you get that data into the cloud that you end up with a fairly significant bill. So that's one reason is you're handling data that's a little bit too heavy duty for the cloud.
The other reason that we've heard is where the data that you are looking to use as part of the application, in this case, the AI application we're typically talking about, is very proprietary. So you really care about who sees and gets access to it, all very sensitive. And in those cases you may not feel comfortable putting that data into a cloud environment. You want to have a little bit more control over where that data resides, what jurisdictions it's under, and so on and so forth.

Sean McManus, TelecomTV:
That's interesting because for years we've been hearing about how latency is one of the real reasons to go to the edge. Is that no longer the case in the era of AI?

Tilly Gilbert, STL Partners:
It's not what we are seeing right now as the big driver. And the main reason for that is because when you're thinking about the latency of an AI system, the end-to-end latency, it might take from the point at which you give it the data, to the point at which it gives you an answer or a recommendation. Actually the networking portion is a relatively small part of that overall latency. More of the time is spent with the AI system actually crunching through and giving you an answer. And edge doesn't really help to make the data crunching any faster. It just helps to make that data transfer a little bit faster. And so typically we're not really hearing latency as a big reason. This might change as we start to get more AI applications that feed straight back into closed loop automation, where it's machines that are interacting with each other and with AI systems rather than humans. But right now, latency isn't the big thing, at least as we are hearing.

Sean McManus, TelecomTV:
Great. Now you have put a value of $157 billion on the Edge AI market by 2030. Tell us where that's coming from. What does that revenue include?

Tilly Gilbert, STL Partners:
Yes, so the big thing to note here is that this is an end-to-end revenue forecast encompassing everything that's needed for an AI solution to work. So we are counting in that 157 billion number, the revenue is coming from selling the devices. Might be cameras, IOT sensors, these types of devices, the revenue coming from the networking, and from the compute infrastructure, and the platform that's needed to make that readily available. And the most significant portion is actually the end application itself. So the actual software application that runs an AI algorithm and provides some business value to the enterprise. And then the last thing that's included in there is integration and support. So some applications in particular will require some bespoke integration work or ongoing management, and those revenues are also captured as part of that number.

Sean McManus, TelecomTV:
Which sectors do you think are going to be the biggest adopters of this?

Tilly Gilbert, STL Partners:
In our forecast, the three big ones are manufacturing, transport, and retail. And those are the three that we see driving the most significant revenues, mainly because they fit in that sweet spot of having a considerable number of applications, not all, but enough AI applications that they want to adopt, that they see real business value in adopting, but where the cloud may not be the best environment for them.

Sean McManus, TelecomTV:
Could you give us an example of maybe a couple of applications that are really good fit for some of those?

Tilly Gilbert, STL Partners:
I think the best examples because they're getting some of the most traction today are examples around computer vision. And these are applicable across all three of those verticals. So in manufacturing, it might be more about monitoring your assets, whereas in retail you might be looking at the movement of people throughout your store. In transport, it might be about making sure that you've got safety across public transport networks. And computer vision use cases are being deployed partly because there's an existing mature device ecosystem. So cameras, CCTV cameras, already exist today, partly because they do derive significant business value for the enterprise. And actually they're a little bit less difficult in terms of some of the technical integration that you might need to do. You can typically run them as an overlay system without having to stop and re-architect the way that a large amount of your IT or your OT across your enterprise works. So they're the big ones that we see in the next one to two years, really driving value in those three verticals.

Sean McManus, TelecomTV:
Have you got any thoughts on the impact this will have for network operators?

Tilly Gilbert, STL Partners:
I've got lots of thoughts. So we've done a recent really interesting study looking at just the networking piece of this. So is AI likely to increase or change enterprise networking requirements? And actually we're somewhat moderate about the real step change in demand today for changes in the network, although I think it will come in the future. But for me anyway, the more interesting portion is, can telecoms operators access a little bit of about 157 billion by providing some of the compute infrastructure in some of those technical sites that perhaps no longer house as much networking equipment, and have some of the characteristics that are needed to be good mini AI data sensors? They will have power, they will have very good connectivity. And so we are doing lots of work with operators to try and help them to understand what the business case might look like for leveraging some of these sites and turning them into something that the enterprises could access and run their workloads in.

Sean McManus, TelecomTV:
That's fantastic. Thank you very much, Tilly.

Tilly Gilbert, STL Partners:
Thank you.