 We're at the CES 2020 and hi. Hi Nicholas. So who are you? Jim McGregor. I'm a principal analyst and founder with Terrius Research. So it's barely the first day and you're already doing a lot of meetings, right? Oh, I've actually already had meetings over the past two days and the show doesn't even start till Tuesday. So it's it's always an exciting time and exciting Show because there's so much to see and so much to do. What kind of research are you doing? We pretty much look at technology so we track everything from some of the basic technology that goes into semiconductor manufacturing, semiconductor development all the way through and sensors all the way through cloud-based systems, servers, and you name it. So what's the latest? There's a lot of stuff that can be said about the latest semiconductor technologies, right? Yeah. No, there is I mean pretty much a big focus that we're going to see this year and going forward is really AI. How are people using AI? How are they going to integrate AI? Whether it's on device, whether it's in the cloud? I mean we've spent five years actually more than five years now focused on IoT connecting devices, using sensors to collect data and everything else. Now it's really about creating intelligence out of that data and the the ones to watch are really the companies that can do that whether it's an intelligent washer, an intelligent phone, or an intelligent car. Has it got to do with the chip designers having more space on the die? They don't know what to do with it, so they put AI in there. It's like one way to use up space. Actually, no, not at all, matter of fact. You'll see in a lot of cases with the kind of the slowing of Moore's law where it's been a challenge to scale up and put more transistors on a die that a lot of companies have had to go to multi-dye solutions, what's called chiplets in a single package, or they've had to go to multiple devices discrete devices just to increase performance. So it's actually just the opposite of what you said. It's not that they have extra space. It's the fact that they're running out of space. Good examples. LG actually in their latest Q70 smartphone used a separate discrete AI chip from Gear Falcon just because they got better performance out of it. The latest Qualcomm solution for premium smartphones. They used the A65, used it to discrete modem just because if they integrated the modem they couldn't get the same level of performance out of it unless they go to another process node that's not available yet. So right now we're actually at a point where we're having to do packaging techniques and multi-chip modules and even discrete products just because we're running out of space. AI is really in symphony. A lot of companies are just using it for one function or two functions like digital photography and stuff like that. We're just really getting to the point where we can start using AI for a whole plethora of functions, you know, for natural language processing, image detection, image correction, you know, all these different things. So that's going to actually take even more silicon as we go forward. Is there like huge fragmentation in the AI space right now? Oh, yeah. Everybody's doing their different kind of AI. How do AI developers keep track? Well, the good thing about AI is the fact that all the software and tools and everything they're being developed or a lot of it is being developed in open source. So that's really helping the development of the community, but in terms of silicon we estimate there's over 300 entities developing silicon today for AI and that's anything from semiconductor startups, existing semiconductor companies, government labs, universities, even OEMs, you know, big companies like Google and Microsoft and Facebook, you know, developing their own AI solutions, even car companies like Tesla. So there is a huge fragmentation in the silicon, but the silicon's really kind of separate from the software and the software, the development of the software tools is really what's enabling this. And there's an abstraction from the silicon layer, so it doesn't have to be developed for certain pieces of silicon. So it's like cafe or tensor or something like that. Exactly. Cafe, tensor flow, those are two of the biggest ones out there. All being designed, you know, you can use them on any platform. So there's not like a need for somebody to step in and say, hey, let's make an open platform. Everybody just go in the same direction. Like ARM is trying to do an AI thing, right? Where ARM is trying to license AI in a certain way, but then Qualcomm, Huawei, everybody's doing a different way. Yeah, no, there doesn't have to be a specific way because no two workloads are the same. I mean, whether you're doing natural language processing or image detection, those are completely different types of models. So you might want a different AI solution for each and quite honestly, each one of those functions is multiple AI workloads within it. So it doesn't have to be the same. And quite honestly, it probably won't be the same. The fragmentation isn't bad. The challenge is really leveraging the technology and keeping up with technology because the software and everything is advancing so quickly. Is there like just a lot of hype and not much to see or is it like a lot of stuff happening? Actually, it's really improving products. It's happening so fast. It's really hard to keep up. I don't think we're seeing it and I haven't seen any of the products yet, but this is the, you know, we're really just getting into the show, but you're seeing, I would say, probably most of the major companies that are here, especially the big ones, are looking at how to use AI and when to use AI. So we're really on the cusp of how all of that intelligence gets fused into the products. Over the next three to five years, I think if you don't have AI using AI, whether it's local, cloud-based, or native, you're pretty much not going to be competitive in the market. Do you research the ARM and the X86 and how about the RISC-5. Is RISC-5 just a way for semiconductor companies to tell ARM to not increase the license fee and then like have an alternative? Actually, not at all. It's an alternative business model and an alternative instruction set architecture, but you have to remember, like in servers, there's never been a single architecture for servers because no two workloads are the same. So there's certain things that are going to run better on MIPS, certain things are going to run better on power, certain things don't run better on X86 and on, you know, ARM. Same thing even with embedded applications. You've got microcontrollers of different form factors, you've got different CPU cores, different GPU cores. You know, if anything, RISC-5 is just another one of those options, especially as we start expanding, you know, some of the different devices. I mean, think about it, we're now putting silicon in tennis shoes and just about everything else. So the number of applications and the opportunity are almost unlimited. So I've been using ARM-powered Chromebooks for it since they exist, but I feel like I'm a small minority of people using ARM-powered laptops, but maybe now things are happening with this Snapdragon, Windows 10, or maybe not. Maybe it's just... I'm using one. You're using one, too, right? We're not like a very big... It's not so big, maybe, yet. Well, I think part of that has to do with the fact that Microsoft really wasn't backing it. I think now that Microsoft has the Surface Pro X out there, you'll start seeing more interest in the ARM-based solutions, and they're getting better. I mean, we're now on the, I think, the third generation of ARM-based PCs, always connected PCs, and they're getting better. And even Intel kind of recognizes that. You know, they now have a push to put cellular connectivity into PCs and to increase the battery life beyond a day. You're right. I've been using one for a year, and I absolutely love it. And quite honestly, just having multi-day battery life and that cellular connectivity has really changed the way I use a PC. I use it more today than I did before on previous PCs, because it's easier for me to open up and type than to try to do things on my smartphone. Is it an 850 Snapdragon you have, maybe? Not yet. Actually, I do. We've got multiple versions. We've tested out everything from the A35 on up. But you're not complaining about performance, right? Performance is okay. Actually, the performance is okay. What do you need to do? One of the problems that I think they had early on, and this may be the case with the Surface Pro X, I'm not sure yet, is the fact that they've spent so much time making it look and act just like a PC, like on the first ASUS that came out, it had so much bloatware on it that the performance numbers were horrible. I actually went through and rebuilt all the software on it and it works fine. That's one of the things. I think if I had a message for the always connected PC guys, the ARM-based PC guys, it stopped making it look just like every other PC. You've got an opportunity to really make it different, or differentiate from the other platforms. And you should be doing that. It should be the true mobile experience. Give users something different. Don't give them the average PC. The one issue I've been having with mine is no Chrome ARM version of the Chrome. I had to use a Firefox, which was ARM optimized, and now Microsoft is doing Edge based on Chromium. But it's just not even a real release. You have to go and get some kind of... It's not even... The browser is the most important app. Chrome has been emulated. That is probably one of the biggest opportunities is not having Chrome native on the platform. Hopefully that gets corrected this year. Just one little other topic. I've been doing videos about the ARM part servers for a decade. And finally, something like the Amazon A1 instance. That's a big deal for ARM servers. Now it seems to be happening with lots of companies working on this. More and more. Actually, the whole ARM in servers has gone through starts and stops, especially as companies like Samsung, Qualcomm and a whole bunch of others jumped into the market and then jumped back out. But it's funny because actually ARM servers have been shipping in volume for many years, especially from Huawei. Huawei has been shipping more ARM servers than anybody else on the planet. But they don't publicize it. And they don't really publicize the devices that they have available. And they're quite phenomenal. But I think as we start seeing more from Marvell, who bought Cavium, and we start seeing more from other vendors, we've got a friend to think of the other one. Yeah, some new ones too. Yeah, there are new ones coming out. It's one of those cases where it's taking a lot longer than it probably should have. But ARM in servers, and quite honestly, I think eventually we'll even see RISC-5 in servers. So it's an opportunity. And they'll likely replace the old MIPS parts and the old PowerPC parts first. But it's kind of a natural trend because once again, no two workloads are the same. And sometimes you need, it's better to have a RISC-based processor than it is to have an X86, certain applications.