A Computex keynote built around Marvell’s role in AI infrastructure, with Jensen Huang joining to reinforce the message that AI’s next bottleneck is connectivity. The speakers argued that copper is still useful near-term, but optics will increasingly take over across racks, packages, and data-center fabrics as bandwidth and scale rise.
Watch on YouTube ›Get the market thesis, key claims, assets, contradictions, and follow-up questions from any financial video — then unlock a version personalized to your portfolio, watchlist, and favorite speakers.
This transcript is essentially a product-and-strategy keynote from Marvell, framed by a joint appearance with Nvidia CEO Jensen Huang. The core thesis is simple and repeated often: AI infrastructure has moved through compute and memory bottlenecks, and the next major constraint is connectivity. Marvell presents itself as the company best positioned for that shift because it spans the full connectivity stack, from long-haul coherent optics between data centers, to PAM4 optics inside data centers, to copper-based scale-up links inside racks, to die-to-die interconnect and packaging inside the chip package. The most concrete part of the presentation is the layer-by-layer breakdown of AI infrastructure by distance. The speaker argues that each distance is a different engineering problem and therefore requires different technology. …
Near term, the stock/setup is driven by the new 100T switch, CPO demo optics, and continued buzz from the Nvidia partnership. The tactical risk is that the story is already crowded, so any delay in productization or soft guidance could cause a pullback.
Over the next few quarters, the base case is that investors keep rewarding Marvell if connectivity demand, 1.6T ramps, and CPO progress continue to confirm the thesis. The key question is whether the company converts the narrative into durable shipment and margin expansion rather than just event-driven enthusiasm.
Structurally, the transcript argues that AI infrastructure is entering a regime where optical interconnect becomes core infrastructure, not an accessory. If that transition plays out, Marvell’s long-run value is as a platform supplier for data movement across the entire AI stack.
AI infrastructure’s next major bottleneck is connectivity, after compute and memory.
This is the main thesis repeated throughout the keynote and reinforced by Jensen Huang.
Copper remains useful, but bandwidth and distance limits will force more optical connectivity over time.
This is the core technical bridge from the present to the future state.
Marvell has the broadest connectivity portfolio, spanning kilometers to millimeters.
The speaker repeatedly frames Marvell as the only one-stop shop across the stack.
Between your platform and Nvidia's, with custom networking and compute needs, does it feel like the time is now to enable customer flexibility together?
Jensen agrees, saying if you buy nothing but Nvidia it's okay, but if you must design your own ASIC, they're still happy having Nvidia inside that data center. He says customers have the benefit of a general-purpose high-efficiency system (Vera Rubin) along with the ability to specialize, which is why Nvidia is in all clouds and it's wonderful to see Marvell expand into all these different clouds.
Do you agree that the timing is right now for Marvell and Nvidia to work together on NVLink Fusion and custom networking given the era of agents?
Jensen says if you buy nothing but Nvidia, that's okay, but if customers must design their own ASIC, Nvidia is still happy to be inside that data center. He notes that between Marvell and Nvidia, customers get the benefit of a general-purpose high-efficiency system plus the ability to extend and specialize, citing that Nvidia is in AWS and Marvell is expanding into all clouds.
How do you see the transition from copper to optics playing out, especially as you go inside the rack?
Jensen says copper should be used as much and as long as possible, but it has limits with bandwidth and distance. The right strategy is to scale up with copper as long as you can, then scale up further with optics, scale out with optics, and scale across with optics. Over the next 5-10 years we'll use a ton of copper AND tons of optics. The bottom line is AI is now profitable, token production is profitable, so demand is high across both.
Unlock the full claims, asset map, scores, related transcripts, follow-up questions, and AI chat — shaped around your portfolio, watchlist, favorite speakers, and risks.
