Great work Ben. I really appreciate your detailed analysis and your rationale on where you believe the memory market is heading. My only question is why do you believe/think RMBS has not yet benefited more from the massive memory demand. As you write, it doesn’t really matter which architecture wins as RMBS makes the picks and shovels. I would have thought that it would have a significantly higher market cap. In your opinion, is it undervalued because the market doesn’t yet fully comprehend what you are writing about. Clearly, all the memory companies have risen exponentially. I would have thought that RMBS would have as well. Something to keep an eye out for.
How is Rambus solving this ? ->”While NVIDIA’s Grace CPU (ARM-based) powers the NVL72 rack-scale systems, its fixed LPDDR5x memory creates friction for enterprises with existing x86 infrastructure”
Outstanding! Thanks for the clear explanation. RMBS is now in the subset of stocks I will be monitoring. If I’m not mistaken, they have earnings coming out next week. Will be sure to listen for any clues there and see if a starter position should be initiated or if I should take a wait and see approach. Thanks very much Ben, as a 59 year old male who didn’t grow up with social media and on-line forums, it still shocks me (pleasantly and appreciably) that experts like yourself share your wealth of information to regular people like myself. Thanks again for your excellent work.
Thanks Charles! You’re asking exactly the right question.
The short answer: RMBS is tied to x86 enterprise, not HBM. Their TAM is the DDR5 companion chip ecosystem (~$2B total: RCDs, PMICs, SPD hubs) — not the $100B HBM supercycle that’s driving SK Hynix and Micron’s valuations.
HBM is the narrative; DDR5 server memory is where RMBS actually makes money. That’s a slower-growth, more predictable market. Product revenue is up 40%+ YoY — solid compounding, but not the violent re-rating you see in pure HBM plays.
The “picks and shovels” thesis is correct — RMBS wins regardless of which memory vendor (Micron, SK Hynix, Samsung) or architecture (RDIMM vs MRDIMM) prevails. But the market prices that optionality differently than direct HBM exposure.
Worth watching as DDR5 continues penetrating enterprise. If MRDIMM ramps with next-gen Intel/AMD platforms, RMBS could re-rate. For now, it’s a quality compounder in a market chasing hypergrowth.
Thanks for the response. I’m sorry for this next question/clarification as I’m not very technical, but is MRDIMM a subset of DDR5, or are they different beasts. I’m trying to understand if DDR5 is one possible catalyst while MRDIMM is another or are they one and the same. Hopefully that makes sense.
Hey Ben, I was reading RMBS’s last few quarterly calls and then rereading your article and I was wondering if RMBS will only be able to re-rate or grow exponentially if the next generation Intel/AMD platforms take hold more significantly than they are presently OR are they also developing new products to fit into NVDA’s Rubin or new product architecture. In other words, will MRDIMM have the potential to be more architecturally agnostic or is it fully tied to the success/failure of AMD/Intel architecture.
Great follow-up. The good news: RMBS is actually more architecturally agnostic than it might seem.
Two separate businesses here:
1. DDR5 chips (RCDs, PMICs) = tied to x86 server platforms (Intel/AMD). This is where most product revenue comes from today.
2. Silicon IP licensing (HBM4, CXL, PCIe7 controllers) = goes into GPUs, custom ASICs, AI accelerators. Rambus has 100+ HBM design wins and launched the industry’s first HBM4 controller IP. NVIDIA Rubin uses HBM4 in 2026.
So yes, the chip business needs Intel/AMD server platforms to grow. But the IP business benefits from NVIDIA, AMD GPUs, and hyperscaler custom silicon.
MRDIMM specifically is a JEDEC standard, so it’s not locked to one vendor. Both Intel and AMD next-gen platforms will support it.
Bottom line: RMBS has exposure to both worlds. The x86 dependency is real for near-term product revenue, but they’re not missing the GPU/AI accelerator wave on the IP side.
on top of all that, google tpus are usually paired with x86 cpus servers too. There is a massive x86 cpu shortage narrative brewing which will only lead to rmbs product revenue growth as more capacity is built by intel/tsm(amd) to meet this cpu demand. While arm cpus are indeed taking some share from x86, the overall market is growing so much that the x86 unit volume is exploding.
That is great news. I actually thought that they were not going to participate in the NVDA architecture. I have to research this more, but my initial read looks quite positive. Are the margins for their IP revenue as strong as where they are guiding MRDIMM. If so, this looks quite positive. Thanks again for all of your help.
Great work Ben. I really appreciate your detailed analysis and your rationale on where you believe the memory market is heading. My only question is why do you believe/think RMBS has not yet benefited more from the massive memory demand. As you write, it doesn’t really matter which architecture wins as RMBS makes the picks and shovels. I would have thought that it would have a significantly higher market cap. In your opinion, is it undervalued because the market doesn’t yet fully comprehend what you are writing about. Clearly, all the memory companies have risen exponentially. I would have thought that RMBS would have as well. Something to keep an eye out for.
Great article Ben!
How is Rambus solving this ? ->”While NVIDIA’s Grace CPU (ARM-based) powers the NVL72 rack-scale systems, its fixed LPDDR5x memory creates friction for enterprises with existing x86 infrastructure”
Outstanding! Thanks for the clear explanation. RMBS is now in the subset of stocks I will be monitoring. If I’m not mistaken, they have earnings coming out next week. Will be sure to listen for any clues there and see if a starter position should be initiated or if I should take a wait and see approach. Thanks very much Ben, as a 59 year old male who didn’t grow up with social media and on-line forums, it still shocks me (pleasantly and appreciably) that experts like yourself share your wealth of information to regular people like myself. Thanks again for your excellent work.
All good. Always be learning. They report Monday
Thanks we are just a little early
Thanks Charles! You’re asking exactly the right question.
The short answer: RMBS is tied to x86 enterprise, not HBM. Their TAM is the DDR5 companion chip ecosystem (~$2B total: RCDs, PMICs, SPD hubs) — not the $100B HBM supercycle that’s driving SK Hynix and Micron’s valuations.
HBM is the narrative; DDR5 server memory is where RMBS actually makes money. That’s a slower-growth, more predictable market. Product revenue is up 40%+ YoY — solid compounding, but not the violent re-rating you see in pure HBM plays.
The “picks and shovels” thesis is correct — RMBS wins regardless of which memory vendor (Micron, SK Hynix, Samsung) or architecture (RDIMM vs MRDIMM) prevails. But the market prices that optionality differently than direct HBM exposure.
Worth watching as DDR5 continues penetrating enterprise. If MRDIMM ramps with next-gen Intel/AMD platforms, RMBS could re-rate. For now, it’s a quality compounder in a market chasing hypergrowth.
Thanks for the response. I’m sorry for this next question/clarification as I’m not very technical, but is MRDIMM a subset of DDR5, or are they different beasts. I’m trying to understand if DDR5 is one possible catalyst while MRDIMM is another or are they one and the same. Hopefully that makes sense.
Good question! MRDIMM is just the next-gen version of DDR5 for servers. Same memory standard, higher performance tier.
Think of it like: RDIMM is today’s server memory, MRDIMM is the upgraded version coming in 2026-27 with 2x the bandwidth.
Both are good for RMBS, and MRDIMM actually needs more Rambus chips per module, so it’s additive to the thesis.
Hey Ben, I was reading RMBS’s last few quarterly calls and then rereading your article and I was wondering if RMBS will only be able to re-rate or grow exponentially if the next generation Intel/AMD platforms take hold more significantly than they are presently OR are they also developing new products to fit into NVDA’s Rubin or new product architecture. In other words, will MRDIMM have the potential to be more architecturally agnostic or is it fully tied to the success/failure of AMD/Intel architecture.
Great follow-up. The good news: RMBS is actually more architecturally agnostic than it might seem.
Two separate businesses here:
1. DDR5 chips (RCDs, PMICs) = tied to x86 server platforms (Intel/AMD). This is where most product revenue comes from today.
2. Silicon IP licensing (HBM4, CXL, PCIe7 controllers) = goes into GPUs, custom ASICs, AI accelerators. Rambus has 100+ HBM design wins and launched the industry’s first HBM4 controller IP. NVIDIA Rubin uses HBM4 in 2026.
So yes, the chip business needs Intel/AMD server platforms to grow. But the IP business benefits from NVIDIA, AMD GPUs, and hyperscaler custom silicon.
MRDIMM specifically is a JEDEC standard, so it’s not locked to one vendor. Both Intel and AMD next-gen platforms will support it.
Bottom line: RMBS has exposure to both worlds. The x86 dependency is real for near-term product revenue, but they’re not missing the GPU/AI accelerator wave on the IP side.
on top of all that, google tpus are usually paired with x86 cpus servers too. There is a massive x86 cpu shortage narrative brewing which will only lead to rmbs product revenue growth as more capacity is built by intel/tsm(amd) to meet this cpu demand. While arm cpus are indeed taking some share from x86, the overall market is growing so much that the x86 unit volume is exploding.
That is great news. I actually thought that they were not going to participate in the NVDA architecture. I have to research this more, but my initial read looks quite positive. Are the margins for their IP revenue as strong as where they are guiding MRDIMM. If so, this looks quite positive. Thanks again for all of your help.