AMD Acquires ZT Systems for AI Rack Design Capabilities – Six Five Podcast EP228

So, uh, here's the news. 9 billion in cash stock. They're gonna do, uh, a lot of it outta cash or a little, do a little bit of a debt, uh, offering, uh, as, as well.

And you might be like, who in the heck is ZT Systems? Well, their biggest customers, rumored are AWS uh, and Azure. They design, they integrate, they manufacture and deploy rack level hyperscale, uh, AI systems.

So think about the entire rack and everything that goes into that, and everything that's connected, uh, to it. They have customer support, they have installation, uh, everything. It's about 2000 people.

Uh, they're located in New Jersey cus uh, a company a $10 billion revenue that you, that you never heard of. And, you know, they've been in business for 30 years. In 94, they started off with PCs.

And in 2024, the company says it ships hundreds and thousands of servers annually. So what does this mean for, for a MD So competitively, the game has changed, right? Uh, and Nvidia really drove this with, uh, you know, starting at the chip and then going to the platform and then doing not only the GPU, but also the CPU and the networking.

And hey, let's, uh, let's just do the, the whole, uh, system and basically A-A-D-G-X, uh, rack. And they, uh, accelerated that cadence to not once every three years, but once, once a year. So there's a lot, a lot of innovation and very hard to keep this, uh, wheel of innovation, uh, going.

5 billion backlog without having infrastructure capability. 5 billion, uh, to attacking an addressable market? Not my numbers, not your numbers, but AMD's numbers of a $400 billion GPU and Accelerator TAM in in 2027.

You've gotta find a way, you've gotta make a big move to, uh, parabolically, uh, achieve that revenue growth. And this is exactly, uh, what they did here. Uh, this adds on to the $1 billion, uh, of investment that a MD has made in software with companies like Silo Nod, AI and MIP Ology.

Um, and, you know, I believe that this is gonna come down to execution. And I think, uh, AMD's, Lisa Sue, who I got the pleasure of talking with last night, uh, is an absolute monster at, at acquisition. Um, and also, you know, we're looking at, um, you know, just numbers here, it, it's a lot less people and a lot smaller than their last acquisition, which was Xilinx.

So I'm not saying this is a layup and not, not saying there's, uh, zero risk, but, uh, it makes, uh, total sense. So there's about a thousand employees in, uh, in the design and about a thousand employees in manufacturing and deployment. Um, uh, Lisa is going to spin out and sell the manufacturing arm.

It would be totally dilutive. I mean, uh, Supermicro, um, you know, their, their, their margins, uh, are in the, uh, single digits. Yeah.

Uh, which would not be good for, uh, a MD who operates, you know, 40%, uh, uh, margins, uh, on there. So Dan, uh, I left you a little bit of oxygen, I think a lot of oxygen in the room. I blew through that in about three minutes.

What is your take If I could, uh, point everybody out there that's watching to Pat's, uh, extensive takes Forbes. I think you posted a LinkedIn article. You definitely use that inside access to give what I would say the most comprehensive rundown of anything I read this morning.

I'm not blowing smoke. I'm not blowing smoke. Uh, I wrote the second best analysis, um, because I put some, some work into that as well.

Everybody that's, uh, not watching that I just winked really loudly. But, um, pat, look, here's what's going on, right? Um, the number of 400 billion's, about twice what our number is in terms of the size of that particular space.

But there are reasons to believe that we we're grossly underestimating the speed by which AI proliferates. And that has a lot to do with these sort of various, uh, bubble perspectives that are out in the marketplace. This perception of a bubble, if it is sort of a slow at first, then all at once sort of implementation at the end.

And then these, these cloud scale companies, these SaaS and ISV providers are gonna be building out AI and delivering it at scale. It's going to happen a lot faster than most of these CAGR numbers reflect. And that was where Lisa could end up, uh, being very accurate with her big ambitions in terms of 400 billion.

Um, having said that, I mean, the company's making the investments here to be able to put the designs together. You know, you didn't really use the word pat you did in your article, not here, but about being open. But look, there are these sort of two schools of thoughts right now when it comes to design.

There's kind of this all in end-to-end, DGX Super Cloud, everything built in one place, all Nvidia, top to bottom, anything that's not this sort of all masked as a single skew, right? And you just consume it. And then there's the other side of the, the house, which is, you know, companies like a MD, um, partnering with companies like Broadcom, um, building out systems with open architectures, using different, you know, standard ethernet network, you know, using advanced, uh, GPUs, various different CPU head nodes, and being able to put these things together and have then develop upon them.

And so that's also where this sort of investment in softwares come into play. You have this big Cuda invest, uh, big Cuda lock-in that everyone talks about, but we've seen a MD making big investments in software. We know Intel has referenced a future with one API, which we'll have to wait until they get to their dis uh, their discreet data center, GPU.

Um, but having said that, this whole thing is sort of evolving to a software battle. So being able to build the most significant designs, being able to go open versus closed. Um, you know, and then the other thing that I think is really important to mention is what the hyperscale cloud providers want.

And I think there is a push and pull here. Of course, they want what everyone's going to quickly consume, they wanna sell through. And I think Nvidia has been a big success for them.

Having said that, on the other side, I think a lot of them want more choice and openness. They want to be able to bring some competitive price in, they need more capacity. Uh, they wanna be able to build their own network backends, uh, their own data center.

Uh, you know, they wanna be able to build uniqueness into their data centers with how they handle networking, for instance. Um, we know AWS has gone that route in some capacities and they wanna be able to address that. And so I think a MD should get some credit here for sort of leaving that door open.

You know, they're not going down the manufacturing, they're not gonna go kind of close end, end to end on the box, but they are going to provide the design that enables that open capability, pat. So, you know, I'm not as, I haven't spent as much time in this as you have, but it seems like an encouraging way to drive that four and a half billion, you know, north, the four and a half billion of committed annual pipeline north. Um, of course attaching more epic, not just, it's not just about instinct, it's about the whole stack.

And then of course, software. And if the true, if it's true about this sort of expansion into, you know, PyTorch and Jacks and higher abstraction layers where co programming can be done, and of course they're adding more capacity, then rock 'em can land. And that's been a, that's been kind of one of those friction points is how successful and how quick can the company move with Roku.

Yeah, good stuff, Dan. And yeah, I, uh, failed to bring up, uh, that, but I did in my article on the software side, you know, they made three tuck in acquisitions. And even though Roku's been around forever, it was really, uh, focused on HPC, which is not hyperscaler data center, generative ai.

Yep. Um, one final thought, Dan, it, it appears to me that, that, uh, the NVIDIA and a MD combo is putting, you know, with this acquisition, they're putting even more distance between than and their competitors. I am really, again, interested, uh, interested, uh, to see, uh, Intel and, and what, what they're doing, uh, with, you know, next generation, uh, accelerators.

And in 2025, uh, data center, uh, GPUs, uh, in Intel has some extensive, uh, capabilities, uh, particularly on the hardware side. They get, they get rack scale architecture. In fact, years ago they, they actually coined, uh, coined that term.

Uh, but it'll be interesting to see how Intel can turn that into gold, uh, as well.