Qualcomm AI 100 Ultra Card

By Patrick Moorhead - November 20, 2023

The Six Five team discusses Qualcomm AI 100 Ultra Card.

If you are interested in watching the full episode you can check it out here.

Disclaimer: The Six Five Webcast is for information and entertainment purposes only. Over the course of this webcast, we may talk about companies that are publicly traded and we may even reference that fact and their equity share price, but please do not take anything that we say as a recommendation about what you should do with your investment dollars. We are not investment advisors and we ask that you do not treat us as such.

Transcript:

Daniel Newman: Let’s talk a little bit about some new Qualcomm updates on the AI 100. Did I say AI or IA? It’s AI 100 Ultra Card, Pat, what’s going on?

Patrick Moorhead: So most of you know Qualcomm as a technology provider for smartphones, and then what we’ve learned if you’ve been paying attention is they’ve been increasing their portfolio into areas like automotive where they have a 30 billion backlog and it’s the first time last quarter that it actually made a highlight reel that the business was big enough that said that it helped the business. Also into PCs and also expanding into IoT. But what a lot of people don’t know is that Qualcomm is leveraging their very scalable AI blocks that they use in a lot of different implementations into the data center or the data center edge.

First we saw the Qualcomm AI 100 and that’s currently inside of AWS and not just for automotive customers like BMW who are going with the Qualcomm solutions for their cars, for self-driving and safety, but also are open for anybody who would want to use them. And as you would expect, the solution is very efficient in terms of what it can do, for lack of a better term, TOPS per watt. I think a lot of people are wondering like, Hey, are they going to keep this going? Well, here we go. A couple of days ago they dropped the A 100 Ultra that cranks out even more performance, trillion parameter models here, which is just shocking and excuse me, a hundred billion parameter model on 150 watt card.

The crazy part about this was, it was more than kind of dog with a note here if any of you speak French out there, but this has actually showed up with two customers. The first one is HPE, and the second one is Cerberus. So Cerberus, everybody knows HPE, very successful on the edge, very successful in high performance computing, and they’re offering an AI training as a service, which I’m still waiting details on pricing in GA. They had acquired a company called Cray who is a leader in highest performant supercomputer. We’re going to talk a little bit about that afterwards.

And then Cerberus is this wafer scale, literally the size of the chip is nearly the size of a wafer, and they don’t just sell the chip, they sell the entire system. That company has seen a lot of activity and interest from US departments of X, Y, Z, and you see the success that HPE has in those same just cloud maybe the US military. Then you combine the trust that a company like Qualcomm has on the inference side, it totally makes sense. HPE has already determined what they want to use for training. They didn’t talk about what they’re using for inference. I’m waiting more details for instance, is this the data center edge? We’re going to see this at Tesla or something like that. It’s a lot clearer cut for Cerberus who doesn’t have an inference capability. They’re more of a training. And now Cerberus can come in with an end-to-end solution leveraging Qualcomm, and I’m super interested to see about what the future holds of this business unit.

Daniel Newman: Yeah, Pat, this is really interesting, for lack of a better word, this is not the part of the market that people think about Qualcomm, but it’s a really useful piece of hardware based on my first assessment. Now, again, I’m reading over what Justin Hotard is saying at HPE, I’m looking at the Cerberus commentary. You’re talking about a very low power consumption, but powerful accelerator for these AI workloads. And it really looks like they’re kind of wedging their way into the cloud. They’re wedging their way onto the prem data centers. They’re wedging their way into the hyperscale cloud potentially here to be offering … to putting their, the way ARM has squeaked its way into the PC and how lower power has found its way.

Could this lower power trend in what Qualcomm has … it has quite a pedigree in this particular space, be the beginning of a new business unit for the company. And it sounds like that’s the direction it’s going. Now, it’s early days here, but from generation to generation, you’re seeing some really good improvements. You’re talking about it looks like something, it’s a pretty significant order of magnitude over that original Edge 100 that they had put out. It’s starting to look quite compelling. They’re finding OEM partners now. You’re hearing from cloud companies that are building as well as accelerator companies. What you’re seeing here from Cerberus that are basically saying, we can partner with Qualcomm here and get the types of gains we need.

It is early for me, I got to get a briefing to be candid, to learn a little bit more about this. But Pat, this could be the next IoT business for Qualcomm. This could be the next, where’s the next big growth come from? And you got to say, it would make Qualcomm more attractive. I know there are kind of heads down all in on AI PC, but let’s face it, data center dollars and margins are just better. So if they can really find their way in here and show that there’s a demand for lower power consumption and obviously high output accelerators, this could be an interesting place. And it’s exactly where Qualcomm is known to be able to play well. So early days. But let’s keep an eye on this Pat, because you know what? From FLOPS to TOPS, it’s an AI world.

Patrick Moorhead
+ posts

Patrick founded the firm based on his real-world world technology experiences with the understanding of what he wasn’t getting from analysts and consultants. Ten years later, Patrick is ranked #1 among technology industry analysts in terms of “power” (ARInsights)  in “press citations” (Apollo Research). Moorhead is a contributor at Forbes and frequently appears on CNBC. He is a broad-based analyst covering a wide variety of topics including the cloud, enterprise SaaS, collaboration, client computing, and semiconductors. He has 30 years of experience including 15 years of executive experience at high tech companies (NCR, AT&T, Compaq, now HP, and AMD) leading strategy, product management, product marketing, and corporate marketing, including three industry board appointments.