Marvell Sells a $625 Billion 2030 Before Its Own Optics Have a Shipping Date

The charge: 2030 priced six days early

Six days before Marvell's Investor Day in New York, three banks have already written its headline. Cantor Fitzgerald raised its target to $330 from $300↗ and kept a Neutral rating, arguing the 6 October event could reframe Marvell's data-center opportunity at up to $625 billion by 2030, enough for a 40 to 45% revenue CAGR and at least $20 of calendar-2030 earnings per share ¹.

Morgan Stanley raised its target to $268 from $246 and stayed at Equal Weight, expecting the event to frame fiscal-2030 revenue above $40 billion ¹. RBC expects Marvell to lift its fiscal-2029 AI revenue estimate by at least $2 billion↗, largely on the Google deal ². The targets moved and the ratings did not, and the stock is up more than 184% this year ¹.

The exhibits: real cash, undated parts

The business underneath is real. Fiscal second-quarter revenue rose 37% to a record $2.739 billion, with data center at $2.172 billion, 79% of the total ¹. Nvidia invested $2 billion in Marvell in March inside the NVLink Fusion partnership ¹. Google's expanded agreement↗ covers custom inference accelerators attached to the TPU ecosystem and carries a warrant for up to 7% of Marvell's shares, tied to revenue milestones ³.

The supply chain is following the story. GlobalFoundries expanded a multi-year silicon-germanium agreement↗ at its Burlington, Vermont fab for near-packaged and co-packaged optics, terms undisclosed ⁴, a week after AT&S named Marvell the customer behind its Kulim substrate expansion ⁵. At ECOC 2026, Marvell demonstrated 2nm optics with no launch dates or pricing ⁶.

The defense: the engineers date it later

At the same show in Málaga, the people who would build that transition put it further out. Applied Materials' Chris Cole, speaking in a personal capacity, said the packaging slow-and-wide requires is four to six years away ⁷.

Ciena's Peter Winzer expects high-speed SerDes to stay necessary through the 400G generation ⁷. OpenAI said the simpler signaling still needs reliability proof, and Nvidia described its own VCSEL design as an exploratory concept, not a product plan ⁷. The show note↗ behind those claims sees no named deployment before the second half of 2027 ⁷.

The cross-examination: timing already cost 7%

The market has priced this gap once already. On 28 August Marvell beat estimates, raised its fiscal-2028 target to about $18 billion, pushed the Google payoff into fiscal 2029, guided gross margin down about 90 basis points on custom mix, and fell 7% ³. On 28 September it fell another 3.77% as AI hardware fatigue spread, on a day OpenAI said it had halted training of its latest models ⁸.

A fiscal 2030 above $40 billion averages out to more than $10 billion a quarter, and that frame is a bank's projection, not Marvell's guidance ¹. The company's record quarter, posted at the end of August, was $2.739 billion ¹⁹.

The verdict watch: 6 October

The bulls' math ends in 2030; the packaging shop's calendar starts there. Investor Day is set to detail the fiscal-2029 custom trajectory, ranges for the Google agreement and a refreshed long-term operating model ³. Which of the Street's numbers Marvell signs itself is the only one that counts next Tuesday.