As AI models move towards the era of massive "Intelligent Agents," the bottleneck of data center compute power is gradually shifting towards "connectivity," triggering a full-blown revolution in underlying infrastructure crossing from copper cables to optical fibers.
On the second day of Computex Taipei, Matt Murphy, Chairman and CEO of Marvell, a leader in AI custom chips, optical communication, and data center interconnection, delivered a keynote speech.
NVIDIA CEO Jensen Huang made a surprise appearance as a special guest, and the two leaders at the forefront of AI compute power and network interconnection stood together, bringing the deep strategic partnership between the two companies to the fore. This joint appearance quickly became the highlight of the current exhibition.
After taking his position, Jensen Huang set the tone for the entire event with one sentence: "Ladies and gentlemen, the next trillion dollar company"—referring to none other than Marvell.
The audience burst into applause. According to the Wall Street News article, this was backed by NVIDIA's announcement months ago of a $2 billion strategic investment in Marvell, highlighting the deep integration of the two companies in the AI data center infrastructure field.
With the release of the previous quarter's financial report, the market is now highly focused on Marvell's benefits in the AI supercomputing cycle.
In response, Murphy presented a highly anticipated answer: ten years ago, Marvell's data center business revenue accounted for less than 10%, but in the last quarter, this ratio had exceeded 75% and was accelerating at an annual pace of about 40%.
Based on the latest financial guidance, Wall Street generally expects its revenue to reach a staggering $16.4 billion next year.
Behind this surge in performance, Huang Renxun and Murphy revealed in their discussion the most crucial investment theme in AI infrastructure—after the bottlenecks of computing power and memory were successively broken through, "connectivity" will define the ultimate performance of the system. The two CEOs' core consensus is:
The next decisive battlefield of AI infrastructure is not computing power, not memory, but connectivity. Marvell is at the heart of this revolution.
It is worth noting that Marvell's stock price surged more than 16% in after-hours trading.
The End of Computing Power is Connectivity: AI Enters the "Useful Stage," Igniting Demand for Infrastructure Interconnection
Why has connectivity become so important today?
In his speech, Murphy explained why "connectivity" has become the most crucial constraint today through a clear logical chain:
The bottlenecks of AI infrastructure have appeared successively and been overcome in turn—computing power (led by NVIDIA, becoming the world's first company with a market value of $5 trillion) → memory (the memory sector has recently seen the emergence of three new trillion-dollar market cap companies) → connectivity (currently happening).
"The world's top hyperscale cloud service providers are redesigning their overall network architecture, realizing that the expansion of AI infrastructure has become the primary connectivity challenge," Murphy said. "This is not my personal opinion, but feedback we received from our largest customers."
In the discussion, Huang Renxun provided the most straightforward business logic:
"Useful AI has arrived, it can now be profitable, and Tokens can be profitable too.
When Token production becomes profitable, everyone wants to produce more Tokens, which is why the demand for Marvell is so high, and why our demand is so high."
Huang Renxun pointed out that AI is now moving towards an "Agent" model, which requires breaking down tasks and deploying them in distributed fashion across huge computing clusters. "When you break down a computing problem into multiple parts and distribute it across the entire data center, the most crucial thing is connectivity."
Huang Renxun did not hesitate to praise his partners, even stating on stage: “Ladies and gentlemen, this is the next trillion-dollar company (Marvell).”
Murphy expressed that relying on a single processor is no longer sufficient to meet AI workloads, and in the future, millions of processors will need to work together.
“The scale of computation is fundamentally a connectivity challenge. The entire industry has already addressed the computing power bottleneck, is currently addressing the memory bottleneck, and the next limitation that will push infrastructure to its limits is connectivity.”
“Use copper where you can, use optics where you must”
One of the most market-referenced segments of the conversation between Murphy and Huang Renxun was their assessment of the transition timeline from copper cables to optical fibers.
Huang Renxun's strategy framework was straightforward: “You use copper wherever you can, you use optics wherever you must.”
He explained that copper cables have physical limits in terms of bandwidth and transmission distance. Before breaking through this boundary, copper cables are a simple, low-cost, practical choice; once the critical point is crossed, optical fibers take over to meet the expansion needs between racks, between data centers, and across data centers.
His key conclusion was:
“In the next 5 to 10 years, we will still use a large amount of copper cables, while also deploying a massive number of optical devices. These data centers are already part of the infrastructure today.”
This “use of copper and optics in parallel, each guarding its boundary” judgment means for the market that: whether in the copper cable or optical fiber fields, Marvell is in a position to continue benefiting—the few companies in the industry capable of providing complete solutions in both directions.
Behind the copper-to-optical transition timeline lies an unavoidable physical law. Murphy explained: the transmission distance of copper cables is inversely proportional to the bandwidth, with distance halving for every doubling of bandwidth.
The current fastest commercially available system has a single-channel speed of 200 Gbps, corresponding to a copper cable length of about 2.5 meters, while the rack height is about 2 meters—considering internal wiring, 2.5 meters has reached the limit.
「When we upgrade to 400 Gbps, copper cables will no longer be able to connect the entire rack. The Copper Wall is shifting, and it has already begun.」 Each time the Copper Wall moves to the right, the number of connections increases by at least an order of magnitude, directly triggering an explosion in optical communication demand.
To address this physical limit, Marvell is doubling down on CPO (Co-Packaged Optics) technology, solving the density and power consumption challenges by directly integrating optical fibers into the package, adjacent to the compute chip.
On the day of the conference, Marvell officially launched a new 100T Ethernet switch designed for AI data centers with industry-leading low power consumption, and showcased a 51.2T switch based on CPO, completely eliminating copper trace routing at the board level.
「This is not some future concept; it is already being implemented right now.」 Murphy stated that once optical interconnectivity completely breaks the distance barrier, future data centers will no longer have the physical rigid boundaries of compute and memory, allowing infrastructure to dynamically scale in response to AI model requirements.
NV Link Fusion Builds an Heterogeneous Ecosystem: Marvell Aims to be the 「Switzerland」 of the AI Era
To meet the extremely complex network architecture requirements, NVIDIA had previously made a strategic investment of $2 billion in Marvell, and the collaboration between the two is expanding across multiple dimensions such as optical communication, silicon photonics, and NV Link Fusion.
The emergence of NV Link Fusion aims to address the customization pain points of Cloud Service Providers (CSPs). Jensen Huang explained that while cloud providers design their own custom chips (ASICs), they still want to integrate NVIDIA's system architecture.
「You don't have to buy everything from us; just buy a portion. By integrating NVIDIA's technology platform with Marvell's technology solutions, fundamentally, a decoupled, distributed, and heterogeneous data center can be built.」
In such an ecosystem, Marvell has found its irreplaceable position.
Murphy emphasized Marvell's neutral and critical position:
「We collaborate deeply with compute companies and storage companies. In many ways, we are like the industry's 'Switzerland,' maintaining partnerships with all companies.」
Welcome to join the official BlockBeats community:
Telegram Subscription Group: https://t.me/theblockbeats
Telegram Discussion Group: https://t.me/BlockBeats_App
Official Twitter Account: https://twitter.com/BlockBeatsAsia
