Transcription
Hey everyone. Welcome back to Chip Stock Investor. It's Kasey & Nick. Hey, all.
This update is part of a larger one that we did with our semiconductor insider members on a CSI Live event. You can check out that membership over on our website, chipstockinvestor.com.
We will start the discussion off with AMD. And this is where we're all focused on, accelerate computing and AI infrastructure and cloud and enterprise. Both big drivers behind everything these days. Yes. Including the AMD pump in the last few months. Oh man. This one really irks me. It's still just, it drives me nuts. Okay, everybody wants to talk about the MI 350 and the upcoming MI 400. And it's good stuff. AMD is getting some good growth off of these chips. I mean, this is a really good picture. It looks incredible. I look at those chiplets. It's been two years now, I think, where we've been talking about this or we're pretty close to it. And I guess we'll just, we'll talk about it again. We'll just keep emphasizing this point.
You are gonna see a lot of charts today from our friends over at fiscal.ai. They have a new enterprise plan that lets you dig even deeper into your analysis of these companies. They have a lot of great new features built on their new data platform, including the availability to click and source where the information came from. For example, a company's annual reports, but even if enterprise isn't what you're looking for, fantastic tools for getting visuals on complex financials and breaking down complex business relationships. Like what we're doing today with AMD, Astera Labs and Arista Networks. You can get 15% off any paid plan, including that enterprise plan with our special link here on the screen, or you can find that in the video description. 15% off fiscal.ai/csi.
Here's AMD's Q2 2025, and we called this out last year to a lot of on the receiving end of what we said on YouTube specifically, a bunch of retail investors that are probably involved with the recent pump, got really ticked off when we said the most important segment is going to be the client in gaming revenue. And lo and behold, here we are in Q2 2025. What's AMD's largest revenue segment again after data center declined sequentially for two quarters in a row. Oh, it's client and gaming. Client and gaming. So what does this mean? AMD's taking market share from Intel. Same story that's been going on for many years at this point. In addition to that, not only is client end gaming the largest revenue segment, it is also paying the bills. Data center of course, dipped into the negative on an operating loss basis, negative 155 million. That a lot of that is because of the write down, uh, from the loss of sales to China, 800 million charge. That data center segment is below corporate average for profitability. So even if you were to add that back, data center would still be a smaller line item for operating income than client and gaming right now. Total operating margin of 11.5% in Q2. And yes, client and gaming, PC also the PlayStation and Microsoft Xbox business heating back up again, are the most important businesses right now for AMD. We can't just sit around talking about chiplets and data center AI and why AMD is gonna become the next $1 trillion market cap because of AI inference. It is not gonna happen like that.
This is another area we need to talk more about. The FPGA business, which competes with Lattice. AMD spent a lot of money to acquire Xilinx back in early 2022. And we're starting to see some more things said about this. But if you believe AI inference is gonna be as big as, everyone hopes it'll be, you've gotta be really happy that AMD has Xilinx. So they released the new Ultra Scale, Spartan Ultra Scale FPGAs, and they announced a new collaboration with Bosch in Europe for Robotaxis, that's good stuff. AI inference is going to happen at the edge, more of it over time will be get pulled out of the data center and get sent to the edge. And so this is a good thing for AMD. This needs to be talked about more in our opinion, a lot more than the data center business because this is actually a very, very profitable business. And when it does scale back up for AMD, this will also be very, very relevant. And they're also gonna be getting some cash from the sale of ZT systems. 3 billion in cash and stock. We'll talk a little bit more about that here in just a moment. And all of their outlook, of course, does not include any revenue from any shipments to China of the AMD Instinct, MI 308.
So let's talk about their cash. Okay, we'll start with the cash flow first and then finish off with the balance sheet. So they did have a pretty big jump in free cash flow, in Q2 2025. Total free cash flow over 1.7 billion. That does include 549 million from discontinued operations. That's probably the China business, maybe also the ZT manufacturing business that's gonna get sold. And then at the bottom we highlighted Q4 2024 quarter net stock buybacks. And then the first half of 2025, there in pink, we have highlighted net stock buybacks. Uh, this is another actually really good area that should be highlighted more often for AMD. So they kind of wound down some of their buybacks in Q4 there in yellow, that corresponds with the last slide, 171 million net repurchases in Q4 uh, when the stock price was flying high. And then see where they ratcheted that back up during the first half of the year, especially during the big sell off in March and April, 1.14 billion net stock repurchases. Uh, expect this number to decrease again in Q3 with the stock price back up near all time highs. Wise use of cash. It's a wise use of cash, so they're gonna get the extra 3 billion in liquidity from the ZT systems manufacturing business. Kasey, what's AMD gonna do with that money? They could do a lot of things with that money. They could pay down some of that debt. They've got 3.2 billion in debt. They could make some acquisitions or they could do some stock buybacks. Lots of good stuff.
AMD's great, but it's not Nvidia. We're gonna circle back to this again at the end of this part of the presentation, when we talk about Arista Networks. AMD is great on its own merits. It's not the next Nvidia, but as we've been saying now for about two years, it really doesn't need to be. We just need to focus on the right things and allow AMD to do its thing.
Let's talk about another fabless chip designer along with AMD. Let's talk about Astera Labs, which, yes, they develop chips, they develop patents. And what exactly do they do with those? It's all about networking. It's one of the key parts of the computing system. So this could be networking from, in particular for Astera Labs, what got them to this point is networking from GPU to GPU, especially for Nvidia GPUs, and the big ramp up here for Blackwell in the last year, but also from storage to, uh, the logic part of the system. So within all of these different connections in a Data center server from logic chip to logic Chip or externally out to the outside world from the data center, or from storage to logic or whatever the case may be, there's a chip inside of all of these different cables and connectors. If you haven't seen it, we do have a Astera Labs deep dive, we did last, I wanna say almost a year ago now on some of the products that they develop and design. Pretty cool stuff, but it's all about networking. Think of it like the freeway system on ramps, off ramps of the data center. The problem with this market is there's a lot of competition here. A ton of competition. Some of the bigger players, of course, Broadcom and then Marvell as well. AMD that we just talked about. Nvidia also produces some of these products. So lots of competition. Alphawave semi we have there in the middle that's getting acquired by Qualcomm. And we'll talk more about that here in just a moment on why that's a smart move for Qualcomm and why that might be a problem for a company like Astera. And then we have Credo circled there on the left. That's one we took position in early June of this year. And maybe you're wondering why we took a position in Credo versus Astera Labs, and we'll talk about that. The two are very close peers or competitors. They have a lot of overlap in what they design.
Here's a look at the competitive landscape, excluding Nvidia. Broadcom, the blue bar, uh, this is just through Q1 2025. Most of these companies have yet to report. So we just left it for comparability purposes. Broadcom's the giant. AMD data center revenue does not break down the different components of that, but most of it is CPUs, some GPUs, and to a lesser extent, some networking chips. Marvell in purple and green, and then really, really tiny little bars on the far right credo and Astera Labs. But here's the Esther Labs Q2 report, which was Epic to say the least, 150% year over year revenue growth to 192 million and operating margins at almost 21%. That's really fantastic scale for a business that's still relatively small. So Astera Labs doing a good job ramping up here, probably also needs to factor for just the fact that this stuff is in high demand and they're getting an Nvidia bump. So kind of writing the coattails of Nvidia here, and we would continue to point out that this is kind of a Astera Labs is basically a way to play Nvidia Blackwell scale up as a lot of what they design directly goes into those systems, especially.
Okay, Q3 guidance. This is also quite good. Expected revenue to be a 9% quarter over quarter increase between 203 and 210 million, at the high end of guidance, that'll be 9% increase and then GAAP operating margin still staying at around 20%. So like we showed you earlier, kind of re-acceleration of the accelerated compute and AI market for data centers and the CEO and co-founder talked about two, there were other items, but we picked these two out of the earnings call. So one is NVLink fusion, so the heightened collaboration with NVIDIA on that. That's basically Nvidia getting lots of other chip companies into its ecosystem rather than becoming all out competitors as they try to get a piece of the data center and AI market. And then also this partnership with Alchip Technologies, which is a smaller designer IP company. They design ASICs for hyperscalers, kind of like what Broadcom and Marvell do. We wanna call that one out in particular here in just a moment. But first let's talk about that NVLink fusion. So the first few years of this AI data center boom was all about interconnecting lots of GPUs together. And that's why NVIDIA made that purchase of Mellanox back in 2020, uh, to be able to network all those GPUs together. However, as we begin to scale up into Blackwell, currently GB 200 and GB 300 is ramping up production through the second half of this year. We're going from lots of GPUs, to needing more networking. Not just scale up where you increase the density of compute, in the rack, but scale out. So adding more servers into the data center for things like not just backend data center and AI training, but also inference. And so Nvidia, one way that they are trying to protect the market share that they've built is this NV Link ecosystem where companies can use the NV Link tech that they got from like Melanox and Design within the Nvidia ecosystem. And some notable names here. Yeah, MediaTek, Marvell, that AIchip Technologies, Synopsys and Cadence they're gonna be using, and Astera can benefit from all of those additional customers, but it does introduce some potential future competition like Fujitsu and Qualcomm. Which is one of the big reasons that Qualcomm is purchasing Alpha Wave semi. They want the interconnect, the networking IP from Alpha Wave, to be a, a better position to compete in this market that has been, basically let's call it just enabled, or created by Nvidia. So interesting things here.
Let's briefly call out Al Chip one that we, I don't think have ever so much as mentioned on the struggle bus the last year or so. Um, based in Taiwan. I feel like it should do better because it looks like AI chip, they should have got a bump because you said it, it will probably now happen. Yeah. Someone's going to talk about this new company they just found called AI chip. AI chip. It. It fits. Yeah, it's, it's an L though. It's a lowercase l, not an i. Sure. If you say so. You see the revenue down sequentially, the last couple of quarters but still a sizable business. And one of the things they do is they design custom accelerators, XPUs, ASICs for hyperscalers. If Astera Labs were to follow a similar pattern as its large competitors, Marvell, and especially Broadcom, it might be interested in eventually acquiring a company like this, or at least acquiring the IP from a company like Al Chip. Uh, so maybe, uh, a topic we can circle back to later on if of interest, but that's Al Chip, uh, one of the collaborators with Astera Labs.
Okay, so maybe let's now slide on over to Credo, because this is the one that we actually took a position in, instead of a Astera Labs. Credo has a smaller market cap than Astera at 19 billion. Astera has about 28 billion in market cap, at least as of right now. Yeah, a few minutes ago. But Credo's on pace to report more revenue than Astera. Yeah. At least that's the guidance. Their quarter isn't quite finished yet, so I think we maybe have another month or so before we get a Credo update. But the reason we decided to reenter another position here is one of Credo's largest customers, if not the largest customer, is Amazon AWS as they are trying to scale out their AI services for their cloud infrastructure customers. So yeah, after Amazon's Q2 update in particular, it's clear AWS is spending a lot of CapEx right now. They don't break down specifically what goes towards data centers versus what goes to the e-commerce warehouse side of the business. But what we do know is Amazon is interested in picking up more of this AI business, 'cause that is what has sent companies like Microsoft Azure and Google Cloud, Oracle Cloud and others soaring the last few years they've picked up market share versus AWS. And so if they're going to aggressively build out more of this AI infrastructure, we think that could put Credo in a good position. And so we kind of treat this as like a CapEx hedge against Amazon AWS if a decent chunk of that AWS CapEx goes to someone like Credo. So that's why that one is in there, why it made the cut versus Astera Labs. We already have lots of NVIDIA's hyperscaler customers, so we opted to go with Credo instead.
Okay. Let's do Arista. Another great quarterly report from Arista. Yeah, Arista, you can see we've mega zoomed into our flow chart here. Biggest parts of the end market industry, tech, equipment, devices dealing in trillions of dollars now, data center market going from 1 trillion to 2 trillion in global installed base in the coming years. And here's Arista's revenue and operating margin. 2.2 billion in revenue operating margin, nearly 45%. And why are these two components what's necessary in a growth story? This is called operating leverage. So the first part of a business's growth story, it's best years of growth, is really obvious. And it's what every investor. On the planet has been trained to look at first revenue growth, and it is important. That's the important place to actually start. But the other component of a growth story is the increasing profit margins. And Arista has been on a fantastic run on both of these fronts for years now. Revenue growth plus operating margin expansion creates massive growth in earnings per share, or free cash flow per share. We'll come back to that at the end.
A couple of things that are not showing up in that Q2 report though. First the balance sheet over 8.8 billion in cash and short-term investments as of the end of June, but, there's a big one that's not reflected in that yet. We speculated that there'd be a big acquisition at some point, and we got kind of a big acquisition for Arista, not a huge acquisition. Paid 1 billion for VeloCloud from VMware, which is now part of Broadcom. So drop in the bucket on their cash balance sheet, and that acquisition finished on July 1st, and this report ended on June 30th, so that's not reflected in their balance sheet yet. Okay. Good stuff. Uh, Arista Networks, a little known cybersecurity play for their big networking business for data centers.
Okay. This is an interesting slide that we're going to show you now that Arista provided and, I know I have had a million articles recommended to me about AI inference, and especially going back to AMD, how AMD's in my instinct series is going to be very important to AI inference. Mm-hmm. Tell me more about what that actually means. For a deeper understanding of how this actually works, we have the double video and article in November, 2024 that we can link somewhere, or you can just find it in the Arista Networks channel. But this is the problem with the AMD argument that inference is going to send it to the moon. Notice in this chart on the, the top of this chart is the, the workload, the AI workload. On the left is dedicated training, the middle is multi-tenant and hosted, so let's just call it traditional cloud computing workloads. And then on the far right AI inference, and we have the important line highlighted here. How many GPUs, or in this case Arista, is saying XPU because they're a big consumer of Broadcom networking chips, and so they work very closely with, with a company like Broadcom making ASICs for hyperscalers. If one of these hyperscaler customers is building a data center for dedicated AI training, they need tens of thousands, if not soon, up to a million GPUs or XPUs interconnected together. That's a market that Nvidia owns like 90% plus. But you move into traditional cloud and inference and these workloads are very different. And so the networking is also very different. And in fact, you could argue so different that, AMD might not actually even be in good position to, to participate unless they partner with someone because their networking business is, is relatively small. The interconnects, the networking is very different. A lot of this multi-tenant cloud and inference is going to be front end, not backend training, and it's gonna be built on ethernet. And notice the big difference here. Hundreds to thousands of accelerators in the middle and maybe tens to hundreds of accelerators on inference. That's a very different market than what Nvidia owns, where they can sell tens of thousands, if not hundreds of thousands of accelerators to a customer. Is AMD gonna be the next Nvidia? Arista Networks just answered that question with this slide right here. The answer is no. It's a very different market and AMD is not built for the same type of explosive growth that NVIDIA has put up the last, especially the last three years. I guess let's move on, before we give away the final conclusion.
Here's Arista's Q3, 2025 outlook. Revenue growth of 24% year over year at 2.25 billion and an adjusted operating margin decrease down to 47% from just over 49% last year. The margin, yeah, it's going to possibly go down on an adjusted basis, but as you saw from the slide earlier, it can be lumpy from one quarter to the next. But the takeaway here 24% year over year growth. These were Arista's targets at the end of 2024, when they gave their report, especially six months ago, they were expecting like 17% annual growth this year, it's now 25%. Uh, the margin profile is on the rise, and they're investing heavily into new markets. Lots of new data center campuses need to be built both for AI front and backend training. And they're participating in both of those now which is fantastic news for Arista. And the more these data centers choose to use good old fashioned ethernet, the better for Arista. There's a reason why NVIDIA is expanding with that NVLink fusion and also making ethernet based products as well. And, uh, Arista is in just really good position to benefit from data center AI inference. If you're looking for an explosive growth story in data center, AI inference, Arista is probably the company that you want. I would argue possibly more than AMD. Now, if we push AI inference out to the edge AMD FPGAs are where it's at, but we might be a little bit further off from that.
So let's talk about Arista. We're showing you our reverse DCF we did in May. And where the point where we said, things were looking quite reasonable for the valuation at the time. Yeah, at this point, the stock price was around $90 and we said per share profit growth of 12% for the next five years. And was it a fair value? We can say, yes, it was. Yeah, and remember, this is a reverse DCF, it's not our estimate. It's just one scenario, one possible scenario that got the stock to fair value. And if you thought that was a low hurdle to clear, then Aristo was a buy. So let's update it. Here we are in August. Uh, we've updated the trailing 12 month earnings per share on a GAAP basis and free cash flow per share. Based on the valuation, obviously going up from 90 bucks to as of this conversation right now, $134 we bumped up the per share profit growth to 20%, kept the terminal growth rate the same. That's one scenario of many using this particular reverse DCF scenario that gets us to fair value today. So the question you have to ask is this still a fair price to pay for Arista? 20% growth over the next five years. Well, back to the slide before two components to a growth story. Yes, it's revenue growth first, but also margin expansion, and you combine both of those factors together. Arista product revenue, 33% CAGR over the last five years, services 23%. Probably gonna get a big bump from the acquisition of VeloCloud in the next year. Operating margin sizable increase over the last five years. Here's what that equates to. Earnings per share, CAGR of 43% free cash flow per share, CAGR of 41% over the last trailing five year period. So for the next five years is 20% per share growth reasonable? It could be. It could be. So who do you want for AI inference? Repeat after us Arista networks. AMD's good. It really is. The only reason we keep dogging on it is because, we just get really tired of the nonsense going on out there about why it's not the greatest thing since sliced bread, but even better, the greatest thing since Nvidia in early 2022. And it's not the same thing. Okay. We brought it full circle. I think that's good for today. Check out our website, chipstockinvestor.com for more blog articles. You can get a Semiconductor Insider membership and we have some exciting stuff coming in the next few months so we'll keep you posted. See everyone later here at Chip Stock investor.