Marvell (MRVL): The Trillion-Dollar Case Behind Huang's Computex Call
When Jensen Huang stood onstage with Matt Murphy at Computex in Taipei and called Marvell the next trillion-dollar company, the market did not treat it as a courtesy. The stock posted its largest single-day gain on record, jumping more than thirty percent the following session. Endorsements from rivals are usually cheap. This one was not, because Huang was not flattering a partner. He was describing the part of the AI buildout he understands better than anyone, and naming the company that owns it.
The claim deserves to be argued on its merits rather than its drama. Marvell would need to roughly quadruple from where it traded at the endorsement to reach a trillion-dollar valuation. That is a serious distance. It is also a smaller distance than it looks once the moat, the revenue trajectory, and the structure of the AI capital cycle are laid out together.
The bottleneck moved, and Marvell owns it
Huang’s argument was narrow and technical. Once a computing problem is disaggregated and spread across thousands of accelerators, the binding constraint stops being raw compute and becomes the ability to move data between the parts. His phrasing was blunt: what makes the cluster possible is connectivity. That is not a marketing line. It is a description of where the next wall sits in scaling AI infrastructure, and it is the wall Marvell has spent a decade learning to climb.
The moat is that the hard problems here are analog, not digital. Optical DSPs, coherent photonics, high-speed SerDes, and the physical layer of moving terabits across a data center without losing signal integrity are not commoditized by Moore’s Law the way logic is. They are won by accumulated electro-optics expertise and by being designed into customer roadmaps years ahead of deployment. Marvell shipped past five million coherent photonic integrated circuits and is already moving its leading parts toward TSMC’s 1.4-nanometer node. Competitors do not catch that lead with a checkbook.
What the recent dealmaking did was extend that moat from scale-out, the layer between racks, down into scale-up, the fabric inside the box where Nvidia’s NVLink has been the moat. The Celestial AI acquisition brought Photonic Fabric optical interconnect into the portfolio at exactly that layer. The XConn purchase added the highest-port-count PCIe and CXL switching silicon on the market, the connective tissue for memory sharing inside dense clusters. Combined with the custom XPU franchise, Marvell now sells the compute, the switching, and the optics for the scale-up domain. That is a full stack assembled precisely as the industry’s bottleneck migrated into it.
What the trillion-dollar number actually requires
Arguments about valuation collapse without a revenue path, so here is the path. Marvell guided its custom silicon business to surpass ten billion dollars in annual revenue by fiscal 2029. Its switching products are modeled to grow from roughly six hundred million dollars in fiscal 2027 toward a billion the following year, and management raised the expected growth of its data center interconnect business to seventy percent this year from a prior estimate of fifty. The first quarter of fiscal 2027 printed record revenue of 2.4 billion dollars, with data center now about three-quarters of the total, up from roughly half two years ago.
A company compounding its largest and highest-margin segment at that rate, with a custom-silicon backlog that converts multi-year program wins into visible revenue, does not need heroic multiple expansion to re-rate. It needs the AI capital cycle to hold and the design wins to ship. Both are happening in the reported numbers, not the projected ones.
The value chain has a direction
The strongest structural argument for Huang’s call is precedent. Capital has moved through the AI chip stack in a predictable order. It rewarded the compute layer first, which is how Nvidia became the most valuable company on earth. It then extended the premium to memory, which is how Micron, Samsung, and SK Hynix re-rated into the trillion-dollar club on the strength of the high-bandwidth memory supercycle. The connectivity layer that binds compute and memory together at cluster scale is the obvious next link in that chain, and Marvell is the purest public expression of it. The market has already shown it will pay trillion-dollar prices for the enablers of AI deployment once their indispensability becomes legible. Connectivity is becoming legible now.
Why Nvidia paid to be inside the tent
The two-billion-dollar Nvidia stake is the tell that turns this from a story into a position. Nvidia did not need Marvell’s cash to do anything. It took newly issued convertible preferred and tied the partnership to NVLink Fusion, the platform that lets third-party custom silicon plug directly into Nvidia’s interconnect fabric. The most rational competitor in the industry looked at the connectivity layer, decided it would rather have Marvell inside its ecosystem than fighting it from outside, and paid for the privilege. When the company that owns the compute layer pays to secure the connectivity layer, it is making the same bet Huang made onstage, only with its balance sheet.
The case against, and why it does not break the thesis
The honest objection is the base rate. When Broadcom crossed a trillion dollars, it was already within reach before AI earnings pushed it over. Marvell is starting from a far lower base, and a fourfold move is the kind of forecast that ages badly when read literally rather than directionally. The stock has already more than doubled this year, which means a meaningful slice of the optimism is priced, and a hawkish Fed or a single soft hyperscaler capex guide could compress the multiple fast. There is also the circular-financing critique that hangs over every Nvidia equity bet: capital flowing from the compute layer into companies that buy compute, a pattern that rhymes uncomfortably with vendor financing.
None of that refutes the thesis. It bounds it. The bear case is about timing and price, not about whether Marvell owns the connectivity layer or whether that layer is becoming the constraint. Those two facts are not in dispute, and they are the load-bearing parts of the argument. A trillion-dollar Marvell may take longer than the headline implies and may demand a better entry than today’s. But the direction Huang pointed is the direction the physics, the revenue, and the smartest competitor in the field are all already moving. The call is aggressive on timing and right on substance, and substance is what eventually sets the price.