Most of the semiconductor complex opened Monday in the red, dragged down by Anthropic CEO Dario Amodei’s weekend essay calling for a deliberate slowdown in frontier AI development. OpenAI’s Sam Altman subsequently voiced support for the idea. Marvell Technology shares fell about 7.6% in premarket trading as semiconductor stocks declined following calls from technology executives for a slower pace of advanced artificial intelligence development. Within that broader rout, MRVL was among the most actively watched names premarket.
That makes what Bank of America published last Friday all the more worth understanding.
The Analyst Who Thinks Marvell Is Underselling Itself
BofA’s Vivek Arya hosted CEO Matt Murphy and CFO Dan Durn at an investor lunch in New York last Wednesday and came away convinced that Marvell is still underestimating how big its AI opportunity could get. His math says the market for custom AI chips, plus the components wired to them, could reach about $300 billion by 2030, roughly five times the roughly $55.4 billion Marvell sized for 2028 back in June 2025.
Strip out high-bandwidth memory, which Marvell does not sell, and about $240 billion is left for Marvell to chase. A 5% to 7% share would translate into $12 billion to $17 billion in sales. Marvell’s current 2028 outlook sits below $10 billion, a target Arya called “very conservative.” He reiterated his Buy rating and $365 price target, about 54% above where the stock traded Friday.
The Numbers Behind the Rally
MRVL has nearly tripled this year, up about 180% since the start of January.
Net revenue for the three months ended August 1 reached $2.74 billion, up 37% year over year, driven primarily by a 46% increase in data center sales on strong AI-related demand. Earlier in the year, Q1 FY2027 had already set the tone: record first-quarter fiscal 2027 revenue of $2.418 billion, up 28% from a year earlier, accompanied by a raised full-year outlook.
Marvell is increasingly emerging as a direct beneficiary of a structural shift in AI infrastructure: the transition from standardized accelerators toward custom silicon designed for hyperscaler workloads. Custom silicon already generates approximately $1.5 billion in annual revenue, close to 18% of FY26 sales, and management expects the business to more than double in FY28, supported by programs already secured with the four largest hyperscalers.
The October Date That Matters
Marvell will host its Investor Day the morning of Tuesday, October 6, 2026, in New York City, with presentations from Chairman and CEO Matt Murphy and the senior leadership team.
That event is now the next real catalyst. The question it has to answer is whether management will formally expand its addressable market framing to something closer to Arya’s $300 billion picture, or hold to targets the Street increasingly views as sandbagged.
Amazon, Marvell’s lead customer, should keep growing its orders every year per management, while Microsoft is set to begin ramping next year. A credible order pipeline spanning both, detailed publicly on October 6, could reframe the stock for investors still anchored to the original 2028 revenue target.
What Could Go Wrong
The Amodei-driven selloff this morning is a reminder of how sentiment-sensitive any AI infrastructure name remains. Marvell’s development opportunities are closely connected to AI infrastructure spending, meaning a slower development cycle could ultimately mean less need for networking and connectivity devices in AI data centers. Customer concentration is the other structural risk: Amazon alone accounts for a significant share of revenue, and any shift in that customer’s chip strategy carries outsized consequences.
Some bears have already flagged high customer concentration, premium valuation, and slowing profit growth as reasons to hold rather than chase. Those concerns do not disappear because Arya disagrees with them.
The Bigger Picture
The custom-chip market is reshaping how AI compute gets built. Google, Amazon, and Microsoft are all designing silicon tailored to their own workloads rather than buying off-the-shelf accelerators from Nvidia. Marvell sits at the intersection of that shift, supplying not just custom logic but the optical interconnects and switching fabric that hold the whole data center together.
October 6 is when management has to put hard numbers around a $300 billion forecast. A dip driven by an AI leader’s blog post may be exactly the wrong moment to lose sight of that.

