Marvell Technology wants to turn a side business into a juggernaut. At its October 6, 2026, Investor Day, the chipmaker said it is targeting approximately $30 billion in annual custom silicon revenue by fiscal year 2031.
For reference, that business brought in about $1.5 billion in FY2026. The goal implies roughly 20 times growth in five fiscal years.
The numbers behind the pitch
The custom silicon target sits inside a larger company-wide forecast. Marvell is guiding to total revenue between $70 billion and $90 billion for FY2031.
The $30 billion custom silicon figure represents the midpoint contribution within that range. Marvell’s total revenue for FY2026 came in at $8.2 billion.
The custom silicon bet breaks into two main pieces. The first is custom XPUs, which are AI accelerators built to order for specific hyperscale cloud customers.
The second piece is what Marvell calls “XPU attach.” This covers the supporting silicon that surrounds an accelerator, including networking and storage chips.
Marvell estimates the XPU attach opportunity could reach $15 billion by FY2031. If that lands, it would leave attach revenue roughly balanced with standalone XPU sales.
Who is buying
Marvell is not starting from zero on the customer front. The company said it is already shipping custom silicon to all four major US hyperscalers and cited more than 35 sockets at those top four customers.
The headline relationship is with Google. On July 29, 2026, Marvell secured a commercial agreement covering custom products for Google’s TPU ecosystem, including accelerators and memory controllers.
The Google deal could be worth up to $120 billion through FY2033, contingent on certain conditions being met.
Beyond custom chips, Marvell pointed to a wider prize in interconnect and optics. The total addressable market for that business is estimated at around $400 billion by calendar 2030.
How the market took it
Wall Street’s initial reaction was warm. The announcements were followed by price target upgrades and a jump in Marvell’s stock.
Cautionary voices flagged the company’s already high valuation as a reason for restraint, and emphasized that the projections lean heavily on Marvell successfully executing programs it has already won, rather than mainly on landing brand new business.
Why this matters beyond Marvell
By targeting attach revenue on par with XPU revenue, Marvell is positioning itself as a supplier for the whole system — accelerators alongside the networking, storage, and memory controllers that turn a pile of chips into a functioning AI cluster.
Concentration is the obvious risk. Selling heavily to four hyperscalers means a shift in spending at even one of them could dent the forecast. The Google agreement’s conditional structure adds another layer of uncertainty.
The width of the guidance also deserves attention. A $70 billion to $90 billion range leaves $20 billion of daylight, a reminder that even Marvell sees meaningful uncertainty in how the AI buildout unfolds.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.