Conference Transcript - RMBS 2026-06-02

Operator

We don't. Okay. Okay, great. So I would like to introduce Rambus, a leading supplier of memory solutions and IP, and notably a key play on x86 CPUs into AI data center. We're pleased to have with us today Sumit Gagneja, Senior Vice President, Chief Financial Officer, and Matt Jones, Vice President of Strategic Marketing. And with that, let's get started. Gentlemen, good morning, and I I think you have a few slides for us.

Sumit Gagneja, CFO

We do, thanks. Good morning. So, Sumit Gagneja. Just before we jump in, quick reminder. Today we may be making forward-looking statements which are subject to risk and uncertainty. So, I refer you to our filings on the SEC on that. So, with that, let's jump in. So, we're Ambos. We are a fabulous semiconductor company and IP company. we have three distinct product business lines if you may one is our product with the semiconductor product business which is anchored around our DDR5 RCD chips as well as companion chips like he makes power management chips that part of the business has been growing 25% CAGR over the five years and last year it grew 40% and it's back in the middle of data center compute build out and the infrastructure build out that's where our products play the second line of business is more around the silicon IP business which is primarily focused on interfaces for HPM the high bandwidth memory PCI Express and security IP that business is around 125 million dollars last year and growing 10 to 15 percent in that range and our third part of our business is our parent royalty business which is roughly 200 to 220 million dollars per year and that's a hundred percent larger business so if you put it all together we have a very strong national model which generates revenue growth from all the three aspects of our product lines or business lines as well as we have high gross margin at the total company level if you put it all together 78 to 80 percent 82 percent margin for the full company and our operating margins are around 45 percent so we generate a lot of cash which helps us to invest back into the business both organically and organically yeah so i'll take it from here

Matt Jones, Other

so as sumit mentioned we'll sort of build this back up uh you know from the bottom of the three ways we go to market uh at the bottom uh foundational patents the rambus was founded you know 30 plus years ago on on key technology for interfacing memory to compute in in high-speed systems we continue to grow that patent portfolio that's been in a core part of our business and foundation in both revenue and cash generation to build the rest of the business on over time from there we moved into silicon ip this is where we build blocks of ip that we sell to asic and more recently hyperscale and cloud providers for integration into things like ai accelerators we'll talk about the positioning of where that fits here in just a moment and then finally on top of that our most recent business and the one that gets a great deal of attention in terms of its scale and importance never as much as today as we're talking about the growing need for compute in AI and even more so as we get into agentic AI is our memory interface chip business. And these are the chips that sit on modularized memory in the data center. Just a brief sort of view of where the products go from an IP and a silicon chip product. Tristan's going to be asking probably a number of questions here, but to help visualize this, the silicon IP, again, these would be the elements that would go into something like an accelerator chip, a GPU, an XPU today. These are high-speed interfaces for memory like HBM, GDDR, LPDDR, and then the interface side of it would be PCI Express, CXL for connectivity. Importantly, for these accelerators and for the data that moves between the accelerator and the CPU is the security of that data. So both securing the data at rest on the accelerator chip, as well as the data in motion with things like IPsec and MACsec is a security IP that's an important part of our IP business, as well as those high-speed interfaces. On the left-hand side, attached to the CPUs, are the DIMM modules. These are registered DIMMs or RDIMMs in today's DDR5 generation and growing into things like SOCAM2 for LPDDR, which you see where our chips fit onto those physical modules that plug next to the CPUs. And then finally in terms of the expanding chip roadmap, we entered the space in the DDR4 generation and have grown our chip portfolio across both DDR5 with the RCD chip, which has been the main contributor to our revenue growth, but the PMIC, SPD hub and temperature sensor, the companion chips we often talk about. But you see a richer opportunity for us for things like SOCAM 2 into the client space where some of the technologies from the data center are now water falling into the highest performance compute AI PCs and the continued build out there. We obviously look forward with a roadmap and some of the investment that we'll talk about here this morning into DDR6 and other memory technologies as we see AI continue to grow and drive performance needs.

Operator

so with that we'll turn it to Tristan okay thank you for my first question I wanted to talk a little bit about the evolution of the company so you had a big wave of you know upgrading from ddr5 to ddr4 to ddr5 that led to a significant content increase and share gain for the company and now we're seeing x86 cpus which used to be a mature market growing well in the double digit driven by ai applications if you could talk about you know how you benefit from those trends obviously given your ardent content and and what is the outlook now that ai and specifically influencing is driving x86 really as a central part of of ai yeah that's a great question tristan so certainly ai in general

Matt Jones, Other

has been a catalyst for the adoption of DDR5 high-performance memory in general in the data center, driving both our product content and our silicon IP business as we see more and more accelerators being built by the hyperscalers. One of the important things that we do see is we move from training to inference, and in today's vernacular, really talking about agentic AI and later physical AI as we get into more AI applications. That inferencing exercise and the orchestration, the tools that get used in agentic AI, those are driving the CPU ratio that many people are talking about today from the favor of GPUs to more in line with CPUs to GPUs. The products where we have played historically in DDR5 is really coupled well with both x86 and ARM-based CPUs. So more CPUs means more opportunities for our chip business. And with the recent introductions of a SOCAM 2 chipset solution, we see the diversity of both DDR and LPDDR as we see that CPU resurgence swing. It's something that we've been talking about for some time within the ecosystem, knowing that we would see, you know, the momentum shift back towards CPUs as we got into this space. So we're very pleased to see it playing out that way, Tristan.

Operator

And just continuing on, you know, on those responses, do you think that that trend will be lasting? I know there is not any real CAGR yet, multi-year for CPUs, but how do you see that as obviously inferencing continues to gain traction? you have agentic ai we know all the data sits on x86 cpus in data center so nav with the sochem 2 you have a direct participation on the gpu itself but if you know what what are your view in terms of the sustainability of those trends yeah i think

Matt Jones, Other

that the durability um is is important to understand because i think one of the things that you know with agentic ai being the you know the most talked about element of ai today and specifically around you know the potential resurgence or you know growth pace of uh of cpus is i think we're in the very early innings of this uh happening you look at some of the solutions that have been announced uh you know especially in the arm community uh arm zone agi for example will begin shipping at the end of this year so some of these solutions that are being talked about is driving us in this direction are coming and so i think it will be something that does sustain what i try to share with people is if we remember back in 2023 when we all started talking about ai it was you know putting a gpu card into an existing server and you've got ai now we're at a place where there are racks of servers and interconnect and software it's a system. Agentic AI is even more so going to be a system-based solution where you have many, many servers working together. So I think it's going to take some time to build out, but it's a very exciting setup for CPUs. And you've hit it on the head twist, and it's very difficult right now for anyone to say, this is how big we think it's going to be or what the kegger is going to be as we're in those early innings. But it is, you know, right on the trend here. And I think we've all gotten more sophisticated in talking about this which is why we're having a conversation about what's going to happen as opposed to what happened as we did kind of when AI took off with training in the first place and just to

Sumit Gagneja, CFO

add to that a little bit right in terms of like you talk for the CPU volume growth right as compute needs drives more number of course within the CPU that's that's growing and to feed that the number of channels are also growing the number of channels per CPU so that all adds up to sort of like how we play

Operator

into that market from growth perspective so I think you know as demand is very strong driven by AI across the board there's been clearly a lot of commentary on the last earnings season about supply constraints at every level whether it's substrates or obviously do I'm how do you see those trends you know either benefiting you or maybe restricting the, you know, the demand, including for x86 CPUs, and, you know, what have you done to get around and navigate, you know, this very tight end environment?

Sumit Gagneja, CFO

Yeah. So, like, I think we have mentioned, and some other companies have also mentioned, tightness in the supply, especially around the, for us, it's more on the back end with OSAT, assembly and test. and what we're doing internally is kind of looking at a supply chain and making it more robust right so we're looking at dual sourcing longer-term relationships and all that and like we mentioned we have you know strong balance sheet and generate a lot of cash so we mentioned that in our Q1 earnings call as well we were strategically building inventory to make sure like we were able to serve our customers from from their demand perspective okay and and from a capacity

Operator

standpoint, any particular steps that you're taking in addition to building some inventories, anything longer term, you know, any supply long-term agreements that you have to kind of build into 27 as it looks like the demand probably is not slowing down next year as

Sumit Gagneja, CFO

Yeah. So from our perspective, like I said, we're diversifying the supply chain. And as we're looking at that, we are looking at also longer-term agreement, right? Long-term relationship, I should say, with our suppliers on that.

Matt Jones, Other

Yeah, I think, you know, we're already well into 2027, you know, discussions, which is ahead of, you know, the planning cycle. It would have been a year ago. So I think that is something that's a trend as well there, Tristan.

Operator

Okay. And then on the hyperscaler side, obviously, you know, DOM has drastically increased as a percentage of material. How do you see, you know, the hyperscalers reacting to this trend? is it fair to say that they get all the D1 that they need and really the supply bottleneck for D1 will be outside of hyperscalers and reminding the audience that you're primarily hyperscaler-based? Or is there any other variable that we should be considering, whether it's any shift in CPU mix or D1 content as the hyperscalers try to mitigate those higher costs, with what we're seeing from a supply standpoint?

Matt Jones, Other

Yeah, Tristan, that's a great question. There's a little bit to unpack there. So thanks, first and foremost, for reiterating the fact that we are, as a company, with the revenue growth that we've seen focused on the data center and the build-out of data centers and AI specifically. So I think what's been written about the most or talked about the most from other companies and throughout the industry is the pressure that we've seen from rising memory pricing and DRAM specifically has been more on the client and mobile side. So potentially not only some pricing out happening there, but some specific comments from vendors talking about taking supply and putting it towards the data center, whether it's memories or CPUs. So I think there's been a very deliberate effect on the buy side. I don't see a behavioral change with the hyperscalers. The build out has been talked about as once in a lifetime, too important to miss type of discussion. So while on the memory side of things, they're saying it's a set your price market, we've seen that largely those prices are being paid at this point. So I haven't seen a great deal of choking off. Now, CPU vendors, you know, memory does have, you know, an end to the runway, so to speak. But as of now, we're seeing continued strength there. And it's, you know, the reaction you asked maybe is probably just, you know, swallowing hard and moving forward by the data center build out folks. In terms of the other things that could impact this going forward, Tristan, I think we're very well positioned as a company for new players coming into the market. The rise of ARM CPUs, the shift in balance within the x86 players, all of those things are things that we've positioned ourselves well for. we talk about being a you know mid 40% market share with the main chip in DDR5 so that means we have a presence across the industry and are well positioned for

Operator

that and and as you mentioned you know the market share so a few years ago you know DDR4 you had about 20% share then you had 40 now you're increasingly talking about you know mid 40s what what's leading those share gains what's the fundamental reason that's clearly company specific that is driving you know higher market share and and you know how sustainable is that trend going to be as we get into new products and also ultimately DDR6 yeah I think two

Matt Jones, Other

things that that you know give rise to the position we have and certainly give a backbone to why we think we'll be able to maintain it their focus and execution so you know we have been 100 focused on our end customers in the data center and our chip business as we went from through the ddr4 cycle and through the pandemic as we saw some shortages you know we built strong relationships with our customers and taking every wafer we got and applying it towards the data center, while some of our competitors may have had hard choices to make at that time. But we've also executed very well, delivering quality products and being ahead of the curve with early investment. So the market share uplift from DDR4, where we entered the market, to DDR5 was all about early investment, working with our customers with those first chips and and then continuing to work with them into each sub-generation or speed uplift of DDR5, continuing to maintain our leadership position there. We've continued to invest both in the specification setting, things like DDR6 within the JETEC standards body, as well as that product investment that we talk about being 23 to 25% of revenue in R&D to continue to drive to maintain that leadership. We worked hard to get to the front. We don't expect to stop the effort from being there. So we look forward to coming generations.

Operator

Great, and then you mentioned just a few minutes ago, the ARM, NVIDIA has talked recently about a 200 billion TAM for their CPUs, including Vera, with lots of traction. you know, not just, you know, Vera, Robin, but standalone, how do you see the demand for ARM-based CPUs in AI, and are they all going to be based on LPDDR5 or, you know, versus 606 CPUs, and, you know, how do you see that increasing mix of ARM architecture benefiting you? Yeah, no, it's a

Matt Jones, Other

great question so we're really excited about the the rise of of ARM solutions into the market the diversity you know the sort of the new flank of competition I think has been exciting there that you know smaller cores you know different kind of architectures that they bring has been exciting for us you know first and foremost when you talk about the the memory utilization to date other than one vendor, NVIDIA, with their Vira solution that uses LPDDR and SOCAM2. The rest of the field has continued to use DDR5 RDIMs. In fact, the ARM community led the adoption of DDR5 and has continued to be a wide adopter of that. With respect to LPDDR, we've been very excited. Recently, we announced a chipset for the SOCAM 2 module form factor that gives us a very strategic and early play also into that part of the ecosystem. So as we see, the continued rise of CPUs were well positioned with the growth in DDR5. I talked about some of the thought leadership just a minute ago as we look to the transition to DDR6, but also we recognize that LPDDR is going to have a strong place potentially in the data center and being a part of that is something we're very excited about. But I think we're seeing the ARM numbers continue to rise, and I think that's going to be a trend that we see for some time. Not only ARM, but also alternative processor architectures, just a diversity of the workloads driving very specific needs in compute.

Operator

And when you said not just ARM, are you referring to RISC-V?

Matt Jones, Other

RISC-V becomes part of the discussion after some time. We've had a long life of X86. Now we see ARM. Is it RISC-V or is it something else? It's just the natural evolution of technology in the space, Tristan.

Operator

Any chance you could give us maybe a very quick outlook on RISC-V in terms of opportunities, again, for your product? and geographically is it going to be more china-centric or you know any given that you are living and breathing directly in that cpu market we're interested in your views on

Matt Jones, Other

that emerging opportunity with i think the the key thing now is that where we are with so cam 2 and lp ddr uh rdim with ddr they're jedex standard products so the we we strong participation there I think, you know, whether it's RISC-V or, you know, an architecture to be determined out of the workshop someplace else over time, I think we've seen as an industry the importance of not only the supply chain, but the technology cohesion within JEDDIC-defined standards being important there. So, you know, our goal is to have chipsets for all modularized memory. So CAM-2 was the most recent addition. So we'd be excited about the emergence of that. I think you mentioned the regionality of it. That's what we're seeing in the early days of RISC-V is really a China-based solution for various commercial and technology reasons, but certainly an exciting development to come.

Operator

And then I think there was a view just a year, year and a half ago that you were not participating directly into AI, GPU. obviously that's changed with SOCAM too could you give maybe some you know background as to what happened initially there was SOCAM and then now with that second generation you're coming out of JEDEC and you have a solution that addresses issues that were specific to SOCAM and you know what is the potential opportunity as you now directly participate on that GPU trend you know you've talked already about having a next generation of product you know what's your expected market share and what exactly are you solving with that

Matt Jones, Other

incremental content? Yeah that's a great question because I think SOCAM 2 is really a fantastic example of the importance of the value that we bring to the memory space and also the importance of the standard body in that space. So SoCam, or S-O-C-A-M-M, was the proprietary solution that was a joint effort between some memory suppliers and a very large GPU concern out there. Unfortunately, the goal had been to take the LPDDR that was historically needed to be soldered onto the motherboard and modularize it because you were seeing memory failures in training runs that were very expensive in either swapping out the hardware, so the GPU-CPU subsystem as a whole, or and also rerunning the training run. So it became a very specific point of failure. The original SOCAM was to take that memory from the motherboard and modularize it. Well, that's difficult. It's more difficult than it sounds. The signal integrity and the power integrity concerns, the chips that we bring to the market, you know, those are problems that are difficult to solve. It's putting memory even through a simple connector changes the entire, you know, signal perspective, the clocking, you know, the power delivery, etc. So with today in the LPDDR5X generation of SOCAM2, As I mentioned, we've announced a chipset that includes voltage regulation solutions and an SPD hub. But as Luke, our CEO, talked about in our most recent earnings call, as we look forward to this, it's a very strategic position for us to have built with this chipset because as we move into LPDDR6 and future generations, we would expect our content opportunities to increase as they have as well in DDR. So things more like P-Mix away from voltage regulators, RCD-like chips for signal integrity for more complexity. You know, really, if we're going to make LPDDR work in the data center, we have to remember it was a memory technology that was built for mobile phones. It has fantastic, you know, characteristics around power consumption, but has some capacity and some reliability constraints that need to be, you know, brought up to bring this forward, and we're very excited about helping the ecosystem maybe deepen the penetration. So your question before of, you know, how does that play out? We're hopeful that, you know, we become completely agnostic, as we have been, to LP versus DDR, just as we are to x86 and ARM going forward.

Operator

Exciting times. Any questions from the audience? We have less than two minutes left. One that probably we should touch on before concluding is the MRDM opportunity. Clearly, that's a whole new TAM layer that's coming up that you've quantified as a multi-hundred million opportunity. How do you see that unfolding in terms of the timing? What does it solve in terms of a problem given the current focus on performance? And how well positioned are you in that emerging opportunity?

Matt Jones, Other

Yeah, so MR-DIMM from a technology standpoint uses the existing memory infrastructure, the memory slots, today's DRAM chips, and with a new chipset from Rambus, gives us the ability to take those slots and double the capacity and the bandwidth on each module. So a very exciting technology to solve some of the bottlenecks that we see in the data center and with some AI workloads specifically. You know, we've talked about that being, as you said, a multi-hundred-million-dollar opportunity, you know, but it's really going to be driven by, you know, the penetration of the usage and the workloads there. In terms of timing, it's very platform-dependent as we move forward. So AMD's Venice platform will be important, and then Intel's Diamond Rapids. We were very pleased to see AMD talk about the production of the Venice CPU, which should lead to Venice-based servers by the end of this year. So we'll start to see some of the utilization and pickup of MR-DIMM. Look forward to Intel being part of that ecosystem as well. I think they've made some comments about being a little bit later than that end-of-year entry point. But those will be the platforms that we look for for that penetration and how successful MR-DEM will be at growing from here, Tristan.

Operator

Thank you very much for presenting. And there's going to be a breakout in a few minutes at the Mizanin level, Aster 3 to 1B. Thanks again.

Sumit Gagneja, CFO

Thank you.