[Q1 CY24] ARM Short Squeeze, Fabrinet Sus
$ARM, $FN, $QCOM, $NYCB
IMPORTANT:
Irrational Analysis is heavily invested in the semiconductor industry.
Please check the ‘about’ page for a list of active positions.
Positions will change over time and are regularly updated.
Opinions are authors own and do not represent past, present, and/or future employers.
All content published on this newsletter is based on public information and independent research conducted since 2011.
This newsletter is not financial advice and readers should always do their own research before investing in any security.
Welcome to a special earnings roundup.
Gotta start with the most interesting one.
ARM
ARM 0.00%↑ experienced a crazy short squeeze in after hours.
Look at the after-hours volume compared to day volume! ARM was up 40% at one point. Totally nuts.
Before covering the earnings, I want to remind readers old and new to check out my two lengthy posts on ARM which built a (clearly wrong) short thesis.
Part 1:
Part 2:
I have shorted ARM twice based on months of research. First at 63.69, covered at 57.85. More recently at 71.25, covered at 68.98. 5% of NAV each time.
These exact trades are being disclosed because I want to emphasize the following:
SHORTING IS DANGEROUS.
I dodged multiple bullets.
Shareholder Letter
Our highest-ever royalty revenue was driven by multiple factors. Firstly, we continue to benefit from higher royalty rates as the adoption of Armv9 technology increases. The royalty rates for Armv9 products are typically at least double the royalty rates for equivalent Armv8 products, and this will continue to generate royalty revenue growth as multiple end markets transition to Armv9.
Chips based on Armv9 technology now contribute around 15% of Arm’s royalty revenue, up from around 10% last quarter.
Interesting that v9 rates are double or higher than v8. Don’t think embedded will ever move to v9 but that is fine. Huge Q-on-Q v9 growth is mostly smartphone.
Secondly, Arm continues to gain market share in the growth markets of cloud servers and automotive which drive new streams of royalty growth. Lastly, the broader semiconductor market is showing signs of recovery, particularly in smartphones which returned to strong growth in Q3.
Automotive is probably not making a difference. GH200 ramp is helping but no idea how much. Most of the royalty growth is smartphone and this was priced in.
One key solution is ArmNN, an optimized AI engine which seamlessly integrates with the most widely used AI frameworks and enables developers to create AI applications for mobile and consumer-electronic devices. Since its launch in 2018, ArmNN has been deployed on more than 700 million devices enabling a seamless way to port AI applications to existing products.
This is hopium. Nobody cares about running neural networks on ARM CPU cores themselves. All the accelerators built onto the SoC run the vast majority of AI workloads.
The beat against sell-side expectations is driven by surprisingly strong licensing revenue. They got 5 new Total Access licenses signed Q-on-Q!
Three of the new Arm Total Access agreements were with companies that had previously been an Arm Flexible Access licensee and upgraded to Arm Total Access - the first time we have seen such an upgrade.
Very surprising that multiple Flex Access customers decided to upgrade. I was under the assumption the Flex licensees cared about reduced costs and would not be motivated enough to pay for v9 cores.
Earnings Call Transcript:
Harlan Sur - JPMorgan
Yes. Thank you. Good afternoon and congratulations on the strong results, guidance and, of course, execution. December quarter, as you guys mentioned, right, second consecutive quarter of strong licensing, second consecutive quarter of book-to-bill greater than 1, strong ACV. It sounds like many of your customers across all of your end markets are focusing on accelerated compute and AI and the requirements for more compute capability. And that's obviously being reflected in the strong licensing performance.
How much of the expansion on recent licensing deals has been more about adding your AI specific IP, right? Like, your Ethos NPU or taking advantage of some of your Helium and Neon vector extensions for AI workloads or Compute Subsystems adoption versus just buying up the stack on more powerful cores? And then more importantly, like do you guys see the strong licensing momentum continuing into fiscal '25?
Rene Haas - CEO
Yeah. Hi, Harlan and thank you for the kind words. I'll take the first part of your question and then let Jason comment on the second half. One of the new products that we released relative from a licensing standpoint is something we call Arm Total Access, which Jason referred to as ATA. That gives customers access to a broad set of Arm technology including our most advanced CPUs and NPUs. And one of the things that we are seeing is exactly what you described. We're seeing demand for incorporating CPUs with anything that helps with AI acceleration such as vector extensions.
Additionally, the ATAs give customers access to the NPUs, which they can also use for an off-load. What we are seeing anecdotally relative to when we engage customers is that the need for more compute, the need to be able to handle what I would call ability unknown relative to these large language models that either run on an edge device or in a hybrid way is fundamentally driving a need for more compute than they had before.
So they are looking to upgrade to give themselves flexibility on the design and also to maximize their ability to deliver the most efficient product, whether that's lots of different cores or a smaller set of devices that may or may not include an NPU. So in summary, yes, your question, I think, is accurate relative to the conclusion of AI demand is driving a need for a lot of different products. And I'll let Jason kind of comment to the longer-term trend that we see.
Analyst asked about how much of the licensing strength is AI and across which IPs. CEO gives a wishy-washy answer on NPU. ARM’s NPU IP is bad and has very little traction. This is fine given how well their licensing business is growing with CPU IP and CSS chiplets. CFO followed up that sales cycles have shortened.
Vivek Arya -BoA
Thanks for taking my question. I just wanted to clarify, Rene, is this on the v9, is the 10% to 15% related to number of customers, number of chips or revenue related to those sales? Because I think in the shareholder letter, it's qualified as v9 of 15% of royalty revenue rather than -- I guess a bigger question is, just so that we have an apples-to-apples sense of how many of your smartphone units are actually using v9 right now versus the ones that use v9 in the last quarter? Is that a better way to track v9 adoption? [Multiple Speakers]
Rene Haas -CEO
Yeah. So let me try to answer your question and maybe Jason, Ian, if I'm missing some facts, you guys can fill in. First off, the number when we say 10% and 15%, that's a percentage of our overall royalty revenue, so that's the way to think about that. When you think about the number of units that are moving from v8 to v9, I don't think we have anything specific that I can give you on this call.
BoA Analyst asked a great question. It matters a lot that the v9 growth is in share of overall royalty revenue, not number of chips. This means ASPs of the chips/mix is going up. Suggests strong Nvidia Grace revenue ramp. If only Qualcomm did not shortsightedly kill the Nuvia server chip program.
Matthew Ramsay - TD Cowen
Thank you very much, guys. Good afternoon. I wanted to go back to the Armv9 conversation on a couple of points. I noticed that this is the first time and maybe I'm just dumb and didn't see it, but I think this is the first time you had explicitly put in the shareholder letter and in writing that you were at least double royalty rates from v8 to v9. And I guess I wanted to ask you about that in a broader sense. Is that sort of across the board, across end markets and also across customers that are traditionally processor licensees and also ones that are traditionally architectural licensees and do the systems themselves?
So I guess that's the first part of the question. Is that a blanket statement across the board. And the reason I ask it if you go back to lots of conversations around the IPO time frame, there were some aggressive ramps of royalty rates across your mobile business. And I think we were all trying to figure out whether the v8 penetration to v9 would drive those kind of expansion in results, or if you would need some significant contributions from sort of system license and the like to get those results. And so any context there about the applicability and breadth of that comment on doubling royalty rates on v9 would be really helpful. Thanks, guys.
Rene Haas - CEO
Yeah. Thanks, Matt. So I'll attempt to answer it and let Jason and -- if Jason wants to chime in. Yes, you're right. This is the first time we've done it although we only have done 2 of these letters, so we don't have a huge installed base to refer to. We wanted to provide some specific clarity because we had been receiving some level of questions around the thing you just asked about relative to how to think about v8 versus v9.
The double, the V9 rate for the equivalent -- double v9 for the equivalent v8 is sort of a rough guidepost. But in some cases, it's quite a bit more. The Neoverse royalty rates have their own unique tables. The automotive royalty rates have their own unique tables. And some of the most high-performance CPUs that we ship into the client section have very, very significant lifted rates over version 8. So double is a good rule of thumb for like-for-like.
But in some cases, it's even better than that. So that's -- but we did want to sort of provide just some clarity because we thought when folks look at the numbers in the absence of that context, there would just be a question of just help me work out how you got here.
TD Cowen Analyst is right. Disclosure of v8 vs v9 rates is new info and very interesting. CEO added some color that I think is quite helpful.
Conclusion:
ARM is obviously overvalued at the time of writing. ($92.30, 2/7/2024)
I have no idea what the correct valuation is.
Managment disclosed several pieces of new/useful info, answered questions with candor, and really seem to know what they are doing.
Nvidia GH100 and Microsoft Cobalt (Newverse CSS) ramps seem like major upside to royalty growth that will more than offset problems in embedded market.
All those Qualcomm physical design engineers who defected last year should be very pleased.
Intel should be terrified.
Fabrinet
Fabrinet is sus.
Their guidance was unexpectedly weak, which is fine. If management spoke with candor and answered questions well, it would have been fine.
Instead, they repeatedly tried to push a bovine-excrement excuse that weakness is driven by Intel 100G ramp-down.
The quarter-on-quarter, the guidance we've given for Q3 contemplates a number of factors, one of which is the transfer out of the 100-gig Intel business will really begin in Q3. And the impact will really be in Q3 and in Q4. So that's one factor that's, if you like, factored into our guidance.
I think if you -- over the next couple of quarters, you'd see that 100-gig business decline and really that as that 100-gig business transfers out, but it's factored into our guidance. So we -- but I wouldn't be prepared to put a number on it right now, Tim.
Yeah. As I said, it's a combination of the Intel business transferring and also those initial AI programs, they're passing into stable production now. They're through the initial part of the growth curve. So that's why we called out that the growth's starting to moderate in Q3 for that very reason. I'm not sure what else is -- Alex. I can't really get into the specifics of an individual customer. But again, it's a combination of both of those, continued growth in AI, but then offset by business transferring out and 100-gig transceiver business.
I **HATE** this excuse. They brought it up three times. Fabrinet management is hiding something. Liquidated my entire FN 0.00%↑ position the next day to take profit. Currently investigating COHR 0.00%↑ and MRVL 0.00%↑ as optics exposure replacements.
Qualcomm
Matt Ramsay - TD Cowen
I guess for my first question, Cristiano, it was good to hear, I guess, the formal announcement of an extension with your partnership with Samsung. And you mentioned, I think, in your prepared script that it started with 2024 devices, but I assume it's longer than that. So maybe you could give us a little context as to the links and any details you can share on the new agreement. You guys obviously have your new custom CPUs coming into the Snapdragon road map and some expanded NPU product as well. So we see kind of what the split is of share in the flagship at Samsung currently, but I'm just trying to understand a bit more about what this means for on a go-forward basis.
Qualcomm CEO refused to answer this question. Share with Samsung dropped dramatically in 2024. Qualcomm only has North America for Galaxy S24 base/plus. Ultra they have globally but that is much lower volume. Exynos is back in a big way.
No, absolutely. Thank you for your question. Look, let me step back a little bit and say we're extremely pleased with our performance in auto, especially when you look at the overall market. Right now, you look at Qualcomm results with record revenues and very strong, I think, year-over-year growth. In 2023, in this year that just closed, we launched 75 models with our silicon with a significant improvement in silicon content as it relates to those immersive cockpit and in many cases, processing for safety. So we're very happy with the business.
QCOM 0.00%↑ CEO is right on this. The entire automotive semi market is going through a rough down-cycle and somehow Qualcomm is still growing nicely. It is off a very small base… but still impressive.
Stacy Rasgon - Bernstein
So given that you've got Android, Android was pretty strong in December, and it's flat into March. How are you thinking about June seasonality given all these trends and moving pieces? I know it's usually down a bit for March, but I guess in the current environment, how are you thinking about June seasonality?
Akash Palkhiwala - CFO+COO
Stacy, it's Akash. No change to the shape of the year comments that we made last time. Following second quarter, we do expect third quarter to be the lowest quarter. It's one of the quarters where we do not have any significant flagship launches. And as a result, you kind of see a decline in third quarter then growth back into the fourth quarter. And it's -- when you look at second to third quarter, we expect a trend consistent with the last 2 years.
This question by Stacy Rasgon summarizes what happened to Qualcomm off of earnings. Down ~4% then relatively flat as most logic comps traded higher off AI sentiment. The market was expecting continued growth in H2 2024 but the smartphone market recover is over, normalized, and stagnant.
Bonus: NYCB
Do you want to listen to some sell-side analysts verbally crucifying a management team? Check out NYCB’s latest call.
Seriously… go listen to the call here.
Great entertainment value.
This was beautiful artricle. Much appreciate this.
do you have a plan to cover Broadcom? They keep increasing margins while relatively flat revenue.