Six Banks, One $90 Billion Bet: The Silicon Scarcity Play Behind Marvell’s 238% Rally

(SeaPRwire) –   By: Reginald Vance

Silicon is the scarcest resource in AI infrastructure. Marvell just became the most expensive expression of that scarcity. The stock closed Tuesday at $287.01, up six percent on a market cap of $251.69 billion. It has climbed 238 percent year-to-date. A wave of analyst price target increases hit in a single week. TD Cowen went from Hold to Buy and lifted its price target to $350 from $245. Jefferies pushed to $450. Evercore ISI jumped to $433. Raymond James held at $295 with a Strong Buy rating. The consensus sits at Strong Buy with 26 Buy calls and four Holds. The average target is $313.84, implying about nine percent upside from current levels. The $70 billion to $90 billion revenue projection through FY31 is not a forecast. It is a bet on silicon scarcity holding its price. When a hardware vendor climbs 238 percent in eleven months, something fundamental has shifted. Sean O’Loughlin at TD Cowen admitted the firm had been wrong about share loss in Marvell’s connectivity business. He noted growth drivers had shifted toward connectivity. Margin risk tied to custom XPU programs had eased. A year ago TD Cowen downgraded the stock. Now the same firm sees over $30 in earnings per share within three years. The gap between yesterday’s pessimism and today’s conviction is the price of being wrong in this market. Semiconductor stocks run on cycles. The market is repricing the bottleneck.

The Investor Day held October 6 laid out the supply architecture in detail. By FY28, revenue is projected near $20 billion with roughly $18 billion coming from the data center segment. Custom chip revenue is expected to top $12 billion by FY29. Jefferies analyst Blayne Curtis estimates interconnect revenue could reach $37.5 billion at the midpoint of FY31 guidance. Custom revenue is projected to near $30 billion by that same year. Evercore ISI cited Marvell’s intellectual property portfolio and flexible business model as competitive advantages. The company spans connectivity, optical products, and custom silicon design. This includes discrete optical transceivers, custom ASIC programs, and interconnect IP. TD Cowen projects earnings per share could exceed $30 within three years. These are not incremental improvements. They represent a wholesale repositioning of Marvell from infrastructure commodity vendor to critical silicon bottleneck holder. The connectivity business was once viewed as declining. TD Cowen now says it has become the center of the growth story. Curtis sees connectivity as the core of Marvell’s entire outlook. The revenue architecture shifts from discrete products to integrated interconnect platforms. That changes the competitive calculus for every hyperscaler building data centers. When one vendor controls the silicon interconnect layer, the negotiation power flows upstream.

Cash flow efficiency is where this story either pays off or collapses. The data center segment is projected to drive the vast majority of revenue by FY28. Cloud capital expenditure on AI is the variable that could break the model. Supply availability for custom silicon remains a constraint. When cloud providers pull back spending or foundry yields slip, that revenue range stops being achievable. Raymond James analyst Simon Leopold called the FY31 outlook extraordinary but flagged execution and supply availability as critical variables. He also noted cloud spending on AI as a factor to watch. The consolidation endgame is clear. A handful of hyperscaler customers will own the roadmap. Marvell’s survival depends on staying attached to that roadmap through connectivity and custom silicon. The premarket slip to $282.29 on Wednesday showed this market moves on margin. The gap between the $313.84 average target and Jefferies’ $450 estimate tells you how much conviction this thesis carries. TD Cowen’s $350 target and Jefferies’ $450 estimate straddle that uncertainty range. If hyperscalers redirect capital away from custom silicon, the entire projection framework unravels. The nine percent implied upside is generous. It assumes nothing breaks between now and FY31. No foundry slowdown. No customer defection. In semiconductor, nothing is ever guaranteed. The stock is expensive because the thesis is real. The risk is that the thesis has already priced itself into the valuation.

Author bio: Reginald Vance is a venture partner specializing in semiconductor valuation and advanced materials, with over fifteen years tracking foundry supply chains, custom silicon programs, and AI infrastructure capital allocation cycles.