Nvidia CEO Jensen Huang appeared at Marvell’s featured event during Computex 2026 alongside Marvell CEO Matt Murphy. On stage, Huang stated:

"Marvell will be the next trillion-dollar company."

That was it. There were no earnings figures, no new product launches, and no major announcements. Yet from the opening bell, Marvell’s share price surged throughout the day. The stock’s 32.52% gain marked the largest single-day increase in Marvell’s history, narrowly surpassing the previous record set in May 2023.

On March 31, 2026, Nvidia and Marvell announced a strategic partnership. Through Nvidia’s NVLink Fusion platform, Marvell will be integrated into Nvidia’s AI Factory and AI-RAN ecosystems. NVLink Fusion is a semi-custom AI infrastructure architecture developed by Nvidia. Previously available only to Nvidia’s own hardware ecosystem, the platform is now being opened to third-party chip designers. Under this framework, Marvell will provide custom XPUs and NVLink Fusion-compatible networking solutions, while Nvidia will contribute supporting technologies including Vera CPUs, ConnectX networking cards, BlueField DPUs, NVLink interconnects, and Spectrum-X switches. As part of the partnership, Nvidia will make a strategic investment of US$2 billion in Marvell.

Marvell’s selection was not accidental. As Jensen Huang explained:

"When you break a computing task into many pieces and distribute it across an entire data center, what you really need is connectivity. That’s why Marvell is so good."

This statement highlights a structural shift currently taking place in AI infrastructure development. Over the past several years, market attention has focused primarily on computing chips — which GPUs are faster and more power efficient. But when training a frontier AI model requires tens of thousands of accelerators working together, the bottleneck shifts to how those chips exchange massive amounts of data with ultra-low latency. Training state-of-the-art models now requires tens of thousands of accelerators operating simultaneously, meaning the bottleneck has moved from raw computing power to communication efficiency between chips. Marvell occupies a critical position in this layer of the infrastructure stack.

Marvell’s Real Situation

Jensen Huang’s endorsement is undoubtedly powerful, but once investors step back and look at the numbers, several realities must be acknowledged. Marvell’s current market capitalization remains a considerable distance from the US$1 trillion target. Even after the rally pushed its market value above US$250 billion, the company would still need to grow roughly fourfold to reach US$1 trillion. Marvell has previously projected that its custom-chip business could exceed US$10 billion in annual revenue by fiscal 2029. However, moving from a US$10 billion revenue target to a trillion-dollar valuation requires a series of assumptions regarding profit margins, competitive positioning, and valuation multiples.

In addition, Marvell faces formidable competition. Nvidia is simultaneously investing in other companies developing silicon photonics technologies. These technologies transmit data using light rather than electrical signals and are widely viewed as more efficient. Marvell is not the only company pursuing this opportunity; it simply happens to be the one publicly highlighted by Jensen Huang.

There is another dimension to this story that deserves separate discussion: to what extent should a CEO’s public endorsement legitimately move another company’s stock price?

Jensen Huang is no longer merely the head of a semiconductor company. His influence within the AI industry has reached a point where his opinions increasingly function as market signals. With Nvidia now valued at more than US$5 trillion, Huang clearly understands what it takes for a company to achieve a trillion-dollar valuation.

But this also raises an important question:

If a few words from a single individual can create US$60 billion in market value in a single trading day, has the market’s pricing mechanism become too dependent on who is speaking, rather than what is being said?