Haha fair pushback. NVIDIA’s NVLink SerDes is world-class for their proprietary scale-up interconnect. But the article is focused on licensable, standards-based SerDes IP that goes into third-party chips. When a hyperscaler designs a custom ASIC, they’re licensing PAM4 SerDes from Marvell or Broadcom, not NVIDIA. Different markets. You’re right that NVIDIA’s internal SerDes capability is elite though, and that’s part of why their vertical integration is so hard to compete with.
The packaging and interconnect layer is where the real forward visibility sits right now. Ben's point about Marvell holding ~70% optical DSP share while simultaneously straddling both sides of the copper-to-optical transition is the exact dynamic that makes basket construction in this theme non-trivial. The Broadcom earnings comment about copper running to 2028 via SerDes, while Jensen simultaneously commits $4B to photonic supply chain, doesn't contradict, it confirms that the transition is a phased, layered migration where different parts of the value chain move at different speeds. The Celestial AI acquisition feels less like strategic clarity and more like a $5.5B admission that Marvell's organic photonics story was thin, but the 1.6T Ara DSP franchise on 3nm is hard to replicate. Does the breadth play (full-distance coverage) ultimately outcompete Broadcom's depth play (vertical ownership from switch to optical engine), or does being everywhere mean defending nothing?
That last question is the right one. Breadth vs depth is the central tension and I don’t think the market has resolved it. My lean is that depth wins at the platform level (which is why I don’t hold MRVL), but breadth has value as a portfolio construction tool for investors who want connectivity exposure without single-product concentration. The Ara DSP franchise is the one piece where Marvell has real depth. Everything else is assembled.
If switching Marvell's SerDes requires a full chip redesign and creates sticky custom ASIC relationships, does Celestial AI's Photonic Fabric (if successful) disaggregate the interconnect from the compute die entirely, with optical chiplets replacing electrical SerDes links and eliminating that switching cost?
Really sharp question. Yes, in theory Celestial’s Photonic Fabric could decouple the interconnect from the compute die and reduce SerDes switching costs. But that’s a 2028+ story and it would also undermine Marvell’s own SerDes moat in the process. The irony is that if Celestial succeeds spectacularly, it weakens the stickiness argument for Marvell’s custom ASIC business. Murphy is betting the interconnect bundle holds together anyway. We’ll see
I would dispute that marvell has better serdes than nvda.
Haha fair pushback. NVIDIA’s NVLink SerDes is world-class for their proprietary scale-up interconnect. But the article is focused on licensable, standards-based SerDes IP that goes into third-party chips. When a hyperscaler designs a custom ASIC, they’re licensing PAM4 SerDes from Marvell or Broadcom, not NVIDIA. Different markets. You’re right that NVIDIA’s internal SerDes capability is elite though, and that’s part of why their vertical integration is so hard to compete with.
I agree with your clarification on serdes ip licensing or used in asic design. Squint and NVLink Fusion fits a similar role.
The packaging and interconnect layer is where the real forward visibility sits right now. Ben's point about Marvell holding ~70% optical DSP share while simultaneously straddling both sides of the copper-to-optical transition is the exact dynamic that makes basket construction in this theme non-trivial. The Broadcom earnings comment about copper running to 2028 via SerDes, while Jensen simultaneously commits $4B to photonic supply chain, doesn't contradict, it confirms that the transition is a phased, layered migration where different parts of the value chain move at different speeds. The Celestial AI acquisition feels less like strategic clarity and more like a $5.5B admission that Marvell's organic photonics story was thin, but the 1.6T Ara DSP franchise on 3nm is hard to replicate. Does the breadth play (full-distance coverage) ultimately outcompete Broadcom's depth play (vertical ownership from switch to optical engine), or does being everywhere mean defending nothing?
That last question is the right one. Breadth vs depth is the central tension and I don’t think the market has resolved it. My lean is that depth wins at the platform level (which is why I don’t hold MRVL), but breadth has value as a portfolio construction tool for investors who want connectivity exposure without single-product concentration. The Ara DSP franchise is the one piece where Marvell has real depth. Everything else is assembled.
Nice read! Are POET / ALMU / LWLG on your radar?
Going to miss the free articles. Gonna have to ride 1 week trials when I can :).
$MRVL is the breadth play with upside - everything people want in a stock.
Why does Matt Murphy get audibly frustrated on earnings calls tho…
Custom asic business is lumpy and most of them don’t make it
I’m talking breadth of scale out and promise of optical scale up with celestial. ASIC is just a bonus.
If switching Marvell's SerDes requires a full chip redesign and creates sticky custom ASIC relationships, does Celestial AI's Photonic Fabric (if successful) disaggregate the interconnect from the compute die entirely, with optical chiplets replacing electrical SerDes links and eliminating that switching cost?
Really sharp question. Yes, in theory Celestial’s Photonic Fabric could decouple the interconnect from the compute die and reduce SerDes switching costs. But that’s a 2028+ story and it would also undermine Marvell’s own SerDes moat in the process. The irony is that if Celestial succeeds spectacularly, it weakens the stickiness argument for Marvell’s custom ASIC business. Murphy is betting the interconnect bundle holds together anyway. We’ll see