Marvell Surges 32% to a $260B Cap — Can Jensen Huang's "Next Trillion-Dollar Company" Call Actually Hold Up?

On June 2, the standout name in US equities wasn't NVIDIA or Broadcom — it was a chip company most retail investors barely knew: Marvell Technology. MRVL closed up ~32.6% at $290.79, the largest single-day move in its public market history, just edging out the record set in May 2023 when the company raised its AI business guidance. The same day, the S&P 500 hit a record 7,609.78, risk appetite was running hot across the board — but Marvell's candle was lit by a single sentence.

That sentence came from Jensen Huang. At a public appearance at Computex in Taipei on Monday, Huang took the stage alongside Marvell CEO Matt Murphy and said Marvell has the potential to become the next trillion-dollar company. For a firm that had just broken into the $250B market-cap tier, that framing put Marvell in the same narrative frame as NVIDIA and Broadcom. The market responded immediately — and broadly. Broadcom (AVGO) tagged along for a ~5% gain on the day, as the entire AI networking category got repriced.

What Happened That Day

The timeline: video from the Computex stage circulated Monday morning, with Huang singling out Marvell for a public endorsement. That endorsement came with harder context — NVIDIA had already announced in March that it was investing $2 billion in Marvell and entering a wide-ranging collaboration covering NVLink Fusion, custom XPUs, scale-up networking, optical interconnects, and silicon photonics across multiple AI infrastructure tracks.

In other words, Huang wasn't being polite. He was publicly surfacing a financial commitment and a partnership that were already in place. That combination — a CEO endorsement backed by real capital already deployed — hits differently on an AI infrastructure thesis. Marvell gapped up premarket, spiked in the first hour, and ground higher into the close without any meaningful pullback. Volume was a multiple of the prior month's daily average, with heavy institutional footprints. This was not a retail sentiment day.

The size of the move itself deserves attention. A 32.6% single-session gain for a company north of $200B market cap is genuinely rare. For context, NVIDIA's largest-ever single-day gain was around 24%. Marvell's June 2 move registers not as a routine beat-and-raise reaction but as a full narrative repricing.

Why the Market Bought It: Networking Is AI's Real Bottleneck

To understand why one sentence could move a stock 32%, you need to understand what Marvell actually sells. It doesn't make training GPUs or inference chips. It makes the layer of infrastructure that lets all those chips talk to each other — networking and connectivity silicon: switch chips, DSPs inside optical modules, custom ASIC interconnect solutions, and the NVLink Fusion, custom XPU, and scale-up networking products Huang mentioned by name.

Why does that matter? A modern AI training cluster may involve tens of thousands of GPUs working a single job in parallel, passing terabytes of data between chips every second. No matter how fast the GPU is, if the highway between chips is congested, actual cluster throughput hits a wall. The industry has reached broad consensus: networking has gone from a supporting role to the genuine bottleneck in AI data center buildouts. Huang has made the point repeatedly over the past two years — the limit on scaling compute isn't the individual chip; it's cross-chip data movement.

Marvell isn't the biggest player in this space (Broadcom is), but it occupies a distinctive position: it builds highly customized solutions for the major cloud providers. The bull thesis is that the more hyperscalers push to build their own AI infrastructure and reduce dependence on NVIDIA's proprietary NVLink stack, the more strategic value accrues to suppliers of custom silicon, scale-up networking, optical DSPs, and silicon photonics. Marvell has been explicit about this in recent earnings calls — it has set a long-term target of more than $10 billion in custom chip revenue by FY2029.

The $2 billion NVIDIA investment lands in this context as a direct bet on Marvell's roadmap. NVIDIA obviously has NVLink and InfiniBand, but next-generation photonic interconnects carry real technology risk and the path hasn't converged. Tying the external ecosystem more tightly is a rational hedge. Marvell taking that capital is, in effect, NVIDIA vouching for its technical direction. That's why institutional commentary after the close read the 32% move not as a sentiment spike but as an official roadmap endorsement.

The Trillion-Dollar Math: From $260B to $1T

The excitement is real — but the question investors actually want answered is whether the math works. How far is Marvell from a trillion-dollar valuation?

Start with the numbers. Post-rally, Marvell's market cap sits around $250–260B. A trillion dollars means roughly another 4x from here. There are two conventional paths. First, revenue catches up: at a 20x price-to-sales multiple — already a premium reserved for AI category leaders — Marvell would need annual revenue around $500 billion. That's a long way from where it stands: FY2026 full-year revenue came in at $8.195B, up 42% YoY; Q1 FY2027 revenue was $2.418B, annualizing to roughly $9.7B. Second, the market assigns an extreme growth premium: at 60x+ earnings, net income would need to scale to $15B+.

Neither path is easy. $500B in annual revenue would put Marvell among the largest semiconductor platforms in existence — well beyond what any single AI networking niche delivers on its own. Getting there requires simultaneously scaling custom silicon, AI networking, and optical interconnects into a dominant combined platform while capturing a very visible share of net-new AI data center spend. That's not impossible — hyperscaler AI capex is already being projected at $100B+ per firm, and networking's share of that wallet is expanding from the low single digits toward double digits — but this is a 5-to-8-year story, not an 18-month trade.

So how should investors read the "trillion-dollar" framing? Not as a valuation anchor. More as a category designation: Huang placed Marvell in the same narrative tier as NVIDIA and Broadcom. The stock has already priced in a large bite of that story in a single session. Sustaining the move requires real order flow and guidance to back it up. If the next two or three quarters don't show continued acceleration in data center and custom silicon revenue, $260B becomes resistance, not support.

The risks are worth naming. First, valuation is already stretched — any guidance miss will be amplified. Second, if NVIDIA advances its own NVLink roadmap by another generation or begins pulling the external ecosystem in-house, that's direct pressure on Marvell's addressable market. Third, hyperscaler custom ASIC timelines shift; Marvell's custom business isn't a signed-and-stable revenue stream. The 32% single-day move has pre-consumed a lot of the good news — the risk/reward of chasing from here has deteriorated materially.

Three Signals to Watch

After the noise clears, a handful of specific data points will determine whether June 2 was a starting line or a peak:

• Next earnings: data center revenue and custom silicon guidance. Watch the sequential growth rate in the data center segment, custom silicon's share of total revenue, and whether the company maintains or raises its long-term FY2029 custom chip target of $10B+. A miss on any of these compresses the narrative fast.
• Hyperscaler custom ASIC program cadence. Cloud providers' in-house chip programs are the core long-term growth variable for Marvell. Next-generation design wins, tape-out timelines, and whether the customer base broadens beyond current anchor relationships directly determine whether the share thesis extends.
• NVIDIA's own networking roadmap. The next-gen NVLink spec, how NVIDIA resolves the InfiniBand-vs-Ethernet question at scale, and where it draws the boundary of NVLink Fusion's external openness will set the ceiling on Marvell's opportunity. If NVIDIA continues treating networking as an open-ecosystem layer, Marvell benefits. If it begins internalizing that stack, the valuation logic needs to be rewritten.

This content is for informational purposes only and does not constitute investment advice, trading advice, or any guarantee of returns.