Jensen Huang’s $70 Billion Endorsement Didn’t Eliminate Marvell’s Risks

Can a single comment add more than $70 billion of market value to a company overnight?

That is effectively what happened to Marvell Technology (MRVL) after Nvidia (NVDA) CEO Jensen Huang mentioned the company at Computex in Taipei as a potential future trillion-dollar chip company. Shares surged roughly 32% in a single session and climbed another 12% in hours of trading, pushing the stock to around $325 per share.

Huang has earned enormous credibility with investors. Few executives have been as right about the direction of AI. Marvell, too, has grown its business at a blistering pace. See How Marvell Is Becoming Broadcom’s Most Credible Rival. But endorsements do not eliminate business risk.

Marvell now trades at more than 80x forward FY’27 earnings and roughly 53x estimated FY’28 earnings, up from about 22x earnings just about a year ago. The market is pricing Marvell for near-flawless execution.

The question investors should ask is simple: did the risks decline as much as the valuation changed? Here are four reasons why the answer may be no.

Four Risks Investors Shouldn’t Ignore

The Concentrated Hyperscaler: Marvell’s custom ASIC business depends heavily on a small number of customers, primarily Amazon, Microsoft, and Google. Data center revenue accounts for about 74% of total sales.
The biggest risk is customers doing more of the work themselves. Amazon already designs its Trainium chips through its own Annapurna Labs division, and Marvell’s filings specifically warn that customers could reduce their reliance on the company through vertical integration. If just one major customer cuts spending, delays a program, or brings more chip design in-house, Marvell’s growth outlook could be materially affected.

The Over-Ordering Risk: When demand is surging, large customers often order more chips and components than they immediately need. This can make demand look stronger than it really is before customers work through inventory and slow new orders. That’s important because Marvell’s valuation assumes today’s AI infrastructure spending remains exceptionally strong for years to come. Marvell’s own filings warn of elevated inventory and potential order deferrals. If hyperscalers spend 2026 and 2027 working through excess inventory, growth could slow sharply. The risk is amplified by emerging constraints elsewhere in the AI build-out, including power availability, grid connections, and data center construction timelines, which could slow infrastructure spending even if demand for AI remains strong.

Leaner Structural Margins Compared to Peers: Unlike Broadcom (AVGO), which benefits from VMware’s high-margin software business, Marvell is primarily a hardware company. Broadcom generates roughly 60% non-GAAP operating margins, while Nvidia’s margins came in at about 65% in the most recent quarter, versus mid-30% adjusted operating margins for Marvell. That’s an important difference because a dollar of revenue simply produces less profit at Marvell. The ramp of increasingly complex 3nm and 2nm custom chips can also pressure margins in the near term. At over 80x forward earnings, investors are assuming that strong revenue growth will eventually translate into much higher profits, a leap that remains far from guaranteed.

Severe Supply Chain Bottlenecks: Marvell is entirely fabless, meaning it designs chips but relies on TSMC for advanced manufacturing and CoWoS (Chip-on-Wafer-on-Substrate) packaging, one of the industry’s key constraints. Any yield issue, manufacturing hiccup, or geopolitical tension in the Taiwan Strait could immediately limit Marvell’s ability to fulfill demand. The company also competes for scarce leading-edge capacity alongside giants such as NVIDIA, Apple, and AMD. If supply remains tight, larger customers may command greater priority, potentially constraining Marvell’s growth even when end demand is strong.

Challenge Is No Longer the Business. It’s the Valuation

Is Marvell too risky? Not necessarily. Demand for custom AI silicon could remain strong for years. The challenge is valuation. At roughly 22x forward earnings, investors had room to absorb delays, inventory corrections, margin pressure, or customer concentration risks. At more than 80x, those risks still exist, but the market is giving them very little weight.

Chasing single-name moves is its own kind of risk. The Trefis High Quality (HQ) Portfolio takes the other side of that bet: 30 quality names, sized and re-balanced with discipline, and a track record of outpacing the S&P 500, S&P Mid-cap, and Russell 2000.