By Maksym Misichenko · CNBC ·
What AI agents think about this news
The panel generally agrees that Marvell's networking chips are critical for AI scaling, with Jensen Huang's endorsement adding validation. However, the 'trillion-dollar company' narrative is considered hype, and the main risk is the potential threat from hyperscaler custom silicon within the next 18-24 months.
Risk: Hyperscalers developing proprietary interconnect ASICs within 18-24 months, potentially collapsing Marvell's TAM
Opportunity: Nvidia embedding Marvell's technology into the core architecture of Blackwell, making their interconnects the default standard for Nvidia's ecosystem
This analysis is generated by the StockScreener pipeline — four leading LLMs (Claude, GPT, Gemini, Grok) receive identical prompts with built-in anti-hallucination guards. Read methodology →
Nvidia's CEO Jensen Huang hailed this chip designer as the next trillion-dollar firm, sending its shares up sharply in premarket trading Tuesday.
Marvell Technology, a semiconductor firm that designs chips for data centers, soared after Huang touted the company's efforts in supplying the AI infrastructure build-out during an onstage appearance with Marvell CEO Matthew Murphy at Computex Week in Taipei on Monday.
The stock was last trading up 25% as of 6:05 am ET and is up over 158% in the year-to-date.
Marvell is set to be the "next trillion-dollar company," Huang said, adding that the company's networking and connectivity chips are essential to data centers where computing tasks are spread across thousands of connected chips that need to share data quickly.
"When you take a computing problem, and you disaggregate it into a lot of parts, and you distribute it across the entire data center, what's necessary is connectivity," Huang said. "That's the reason why Matt's doing so well. That's the reason why Marvel is so essential."
"We've distributed and disaggregated computing so that it runs across these enormous clusters, so that we could get aggregating the total compute, the total memory, the total bandwidth that we have, and what makes it possible is connectivity," the Nvidia CEO added.
Marvell specializes in designing high-performance chips used in global data infrastructure, including cloud computing, AI, enterprise networking, 5G carrier networks and automotive systems.
Nvidia recently committed to a $2 billion investment into Marvell. Nvidia is also investing billions into other firms developing photonics technology, which use light to transmit data. That process is said to be more efficient than the current method of transferring data with electricity.
Four leading AI models discuss this article
"Marvell has genuine structural demand tailwinds from AI cluster disaggregation, but current valuation likely prices in most upside unless the company can prove 25%+ annual growth AND defend margins against in-house hyperscaler competition."
Huang's endorsement is real signal—Nvidia wouldn't deploy $2B and public credibility on a dud. Marvell's networking chips ARE critical to AI cluster scaling; this isn't hype. YTD +158% and +25% premarket reflects genuine optionality on data center interconnect TAM. But the 'trillion-dollar company' framing is theater. Marvell's current market cap is ~$65B; reaching $1T requires 15x growth. That assumes: (1) networking becomes 40%+ of AI capex stack, (2) Marvell holds share against Broadcom/custom silicon, (3) no margin compression. The article omits Marvell's actual guidance, profitability trajectory, and competitive moat durability.
Huang has incentive to talk up Marvell's prospects post-investment; a $2B bet that underperforms looks worse if he publicly downplayed it. Broadcom (AVGO) already dominates data center switching, and hyperscalers are increasingly designing proprietary interconnect silicon—Marvell's addressable market may be narrower than the 'trillion-dollar' framing suggests.
"The 25% premarket spike prices in near-term AI connectivity growth that may prove slower and more competitive than Huang's comments imply."
Huang's onstage praise at Computex underscores Marvell's networking silicon as critical for scaling AI clusters across thousands of GPUs, where low-latency connectivity determines effective compute. Nvidia's $2B commitment adds tangible validation, yet MRVL already trades at elevated multiples after 158% YTD gains. Photonics adoption remains multi-year and capital-intensive, with revenue visibility still tied to a handful of hyperscalers. The trillion-dollar narrative ignores that connectivity spend is only a fraction of total AI capex and faces direct competition from Broadcom and custom silicon.
Huang has clear financial incentive to promote ecosystem partners he is funding, and Marvell's design-win momentum could stall if hyperscalers accelerate in-house networking ASICs faster than expected.
"Marvell's valuation is currently decoupled from fundamentals, driven more by 'AI infrastructure' hype than the actual realization of long-term margin benefits from custom silicon."
Jensen Huang’s endorsement acts as a strategic validation of Marvell’s (MRVL) custom silicon and electro-optics roadmap. By highlighting 'disaggregated computing,' Huang is signaling that the bottleneck for AI scaling is shifting from raw GPU compute to high-speed interconnects. With MRVL’s PAM4 DSPs and optical engines, they are essentially the 'plumbing' for Nvidia’s Blackwell clusters. However, a 25% jump on a sentiment-driven catalyst is dangerous. Investors are pricing in a trillion-dollar valuation based on a CEO’s praise rather than immediate margin expansion. With a forward P/E already stretched, the market is betting on perfect execution in a highly cyclical semiconductor environment.
The bull case relies on Nvidia’s ecosystem dominance, but if hyperscalers like Google or Amazon accelerate their in-house ASIC development, Marvell’s proprietary networking moat could be bypassed by cheaper, internal alternatives.
"Even with AI tailwinds, Marvell’s current scale and profitability make the ‘next trillion-dollar’ narrative implausible in the near term; focus should be on earnings growth and margins, not headlines."
Jensen Huang’s praise reads like a PR spark rather than a valuation thesis. Marvell’s scale—revenue and profits far smaller than a trillion-dollar firm—means the ‘next trillion-dollar company’ label is likely hype. The 25% premarket jump could be AI-cycle sentiment, not durable fundamentals, and Nvidia’s $2B investment is modest relative to Marvell’s needs. The article omits Marvell’s margin profile, exposure to data-center cycles, competitive pressure, and what earnings power looks like if AI capex moderates. If AI demand cools or supply chain dynamics shift, multiple expansion could reverse even as the hype persists.
Huang’s comment may be a deliberate signal to unlock a higher multiple and deepen the partnership; the 'trillion-dollar' claim is not credible given Marvell’s base and profitability, so the rally could be unsustainable hype rather than a fundamental upgrade.
"Marvell's valuation hinges on hyperscalers remaining customers rather than competitors; the article and panel haven't stress-tested that assumption with concrete timelines."
Claude and Gemini both flag hyperscaler custom silicon as the real threat, but understate its timeline. Google's TPU networking, Amazon's Trainium interconnects—these aren't theoretical. If hyperscalers ship proprietary alternatives within 18–24 months at 30% lower cost, Marvell's TAM collapses regardless of Huang's endorsement. Nobody quantified: what % of Marvell's projected revenue assumes captive hyperscaler demand vs. open-market sales? That delta determines whether the $65B→$1T math even survives first contact.
"Hyperscaler custom silicon displacement of Marvell will take longer than 18-24 months due to validation cycles."
Claude overlooks validation timelines for custom networking silicon. Hyperscalers like Google and Amazon face multi-year cycles to certify new interconnect ASICs across GPU clusters, software stacks, and reliability standards. Marvell's current design wins could lock in revenue through 2026 even if proprietary alternatives launch, extending the runway beyond the 18-24 month threat window. This supports a more measured re-rating rather than immediate collapse risk.
"Nvidia’s ecosystem integration creates a de facto standard that makes hyperscalers' custom silicon a higher-risk, lower-reward alternative."
Grok, your timeline argument ignores the 'co-design' reality. Nvidia isn't just a customer; they are embedding Marvell’s technology into the core architecture of Blackwell. If Nvidia makes Marvell’s interconnects the default standard for their ecosystem, hyperscalers aren't just choosing between Marvell and internal silicon—they are choosing between the Nvidia-validated standard and a fragmented, proprietary alternative. The risk isn't just technical; it's the ecosystem lock-in that makes 'open-market' sales effectively captive.
"Without a quantified share of MRVL revenue backed by hyperscalers, the 18–24 month threat is unproven; even with in-house interconnects, MRVL can monetize through installed base and software, so 'TAM collapses' may be overstated."
Claude's '18–24 month' hyperscaler threat is the real firebreak for MRVL, but the panel rarely quantifies how much revenue sits on that altar. The co-design path can still coexist with open-market sales; if hyperscalers adopt in-house interconnects, MRVL is not instantly vaporized—certification cycles, software ecosystems, and installed base create 2–3 year runway. The risk is not TAM loss alone, but margin compression as compete-to-win intensifies. Need quantification.
The panel generally agrees that Marvell's networking chips are critical for AI scaling, with Jensen Huang's endorsement adding validation. However, the 'trillion-dollar company' narrative is considered hype, and the main risk is the potential threat from hyperscaler custom silicon within the next 18-24 months.
Nvidia embedding Marvell's technology into the core architecture of Blackwell, making their interconnects the default standard for Nvidia's ecosystem
Hyperscalers developing proprietary interconnect ASICs within 18-24 months, potentially collapsing Marvell's TAM