Nvidia CEO Jensen Huang used Marvell's Computex stage to make a very public call: the custom silicon specialist could be the next chip company to reach a $1 trillion valuation. Investors did not need much convincing.
When Jensen Huang says a company could be worth a trillion dollars, Wall Street listens. It did exactly that on Tuesday, June 2, when Huang joined Marvell Technology CEO Matt Murphy at Computex 2026 in Taipei and described Marvell as the likely "next trillion-dollar company." Marvell shares jumped more than 25% in premarket trading, a move that pushed the company's market value toward the $200 billion mark and showed how quickly investors are repricing the less obvious winners of the AI infrastructure buildout.
The timing mattered. This was not a casual compliment from a friendly partner. Nvidia disclosed a $2 billion investment in Marvell on March 31 as part of a broader agreement around NVLink Fusion, the rack-scale interconnect technology that lets third-party AI accelerators work more closely with Nvidia CPUs, networking gear, and data center systems. Huang appearing in person on Marvell's stage turned that technical alliance into a market signal.
According to Data Center Dynamics, Murphy used the Computex keynote to frame connectivity as the next major bottleneck in AI data centers, with Huang's appearance underlining the point. That is the real story behind the stock move. Nvidia is still the company everyone watches for GPUs, but the next phase of AI spending is spreading into custom chips, optical links, switching silicon, and the physical network fabric needed to make huge clusters behave like one machine.
Marvell sits in that layer. It is not trying to replace Nvidia in the GPU market. Its stronger position is as a designer of custom and semi-custom silicon for hyperscalers that want accelerators built around their own workloads. Amazon has worked with Marvell around the Trainium family. Microsoft has been linked to Marvell through its Maia program. Google has used Marvell in its Axion custom CPU work, and reports over the past year have tied the two companies to further custom AI chip discussions.
That customer list explains why Huang's endorsement landed so forcefully. The largest cloud companies want more control over AI infrastructure costs, power consumption, and supply. They also do not want to walk away from Nvidia's ecosystem, because Nvidia's software, interconnects, and data center architecture are already deeply embedded in how AI systems are built. Marvell gives them a way to pursue custom silicon without necessarily leaving Nvidia's orbit.
The custom silicon angle Huang is betting on
NVLink Fusion is the bridge between those two priorities. Historically, a cloud provider choosing custom chips faced a harder tradeoff: build more of the stack itself, or stay inside Nvidia's tightly integrated GPU systems. NVLink Fusion softens that choice by allowing custom accelerators to connect into Nvidia's broader rack-scale architecture. In that setup, Marvell becomes less of a rival to Nvidia and more of a preferred design partner for customers that still need Nvidia's networking and platform advantages.
That is a strategically powerful place to stand. Hyperscalers have every reason to diversify away from full dependence on Nvidia GPUs, especially when AI demand is rising faster than supply chains can comfortably absorb. But they also have little interest in rebuilding every layer of the system from scratch. A Marvell-designed accelerator that can sit alongside Nvidia infrastructure gives them leverage without forcing a full break from the platform that already works.
Marvell's own numbers have started to support that story. The company recently told investors that its custom silicon opportunity could reach more than $10 billion in annual revenue by fiscal 2029, driven by existing cloud customer programs and the broader move toward purpose-built AI infrastructure. That is still a long way from justifying a trillion-dollar valuation on its own, but it gives investors something more concrete than a stage remark to underwrite.
The other half of the thesis is optical connectivity. Marvell has been ramping 1.6 terabit optical digital signal processors and related networking silicon, products that become more important as AI clusters grow from thousands of chips to tens of thousands. At that scale, the question is not simply how fast each processor runs. It is how efficiently data moves between processors, memory, switches, and racks without turning latency and power draw into the limiting factor.
That is why Murphy's "data center without distance" idea is more than conference language. If optical interconnects can make physically separated compute resources behave more like one coherent system, cloud providers gain flexibility in how they build, expand, and operate AI capacity. It also makes Marvell's role harder to categorize. The company is not just selling components into the AI boom. It is trying to become part of the architecture that determines how the boom scales.
The market will now look for evidence that Huang's public blessing changes customer behavior. ASIC design wins usually take years to convert into revenue, so the next quarterly call will matter less for near-term earnings and more for signs that Amazon, Microsoft, Google, or other cloud buyers are expanding programs tied to NVLink Fusion-compatible systems. The stock reaction was immediate. The real test is whether Marvell can turn that endorsement into a deeper backlog before the trillion-dollar claim becomes just another line from Computex.
Also read: Alaska joins the sound money movement by stripping all state taxes from gold and silver • Blackstone closes its largest Asia private equity fund ever at $13.1 billion as institutional capital concentrates at the top • SpaceX is carving out 5% of its IPO for insiders and waiving lock-up rules for them while everyone else waits
