Marvell Triples Custom Chip Target to $12 Billion and Lays Out $70-90 Billion Revenue Path by Fiscal 2031
Investor Day at The Pierre, October 6, 2026, details XPU attach windfall and a $400 billion total addressable market
Marvell Technology used its first Investor Day in over two years to deliver a dramatically upgraded long-term outlook, centered on an explosion in demand for custom silicon attached to AI accelerators. The company raised its fiscal 2029 custom revenue target to more than $12 billion, up from the more-than-$10 billion goal set just two months ago on the August earnings call, and laid out a fiscal 2031 total revenue range of $70 billion to $90 billion, with $80 billion at the midpoint. That compares with $8.2 billion in revenue generated last fiscal year, implying roughly 10x growth over five years and a 55% to 60% compound annual growth rate.
XPU Attach Becomes the Real Story, Not Just XPUs
The single biggest upward revision came from what Marvell calls "XPU attach" — semi-custom silicon such as NICs, memory expansion controllers, storage accelerators, and a newly disclosed category of AI infrastructure management chips, all of which sit around a hyperscaler's custom AI processor rather than being the processor itself. CEO Matt Murphy said the increase in the fiscal 2029 custom target to tripling (from doubling previously) from fiscal 2028 levels "is primarily driven by the rapidly expanding XPU attach opportunity we see in front of us, including the strategic agreement that we announced in August," referring to the $120 billion, 6.5-year deal with Google covering custom products attached to the TPU ecosystem. Will Chu, who runs the custom business, introduced two entirely new attach categories to investors: AI infrastructure management silicon, where baseboard management controllers must now track more than 1,500 sensors per server generating hundreds of megabytes of data daily (a "1,000x" increase in required processing performance versus traditional servers), and inference acceleration chips, which Chu said could each represent "a $1 billion opportunity with some reaching several billions." Critically, management framed custom's long-term $30 billion fiscal 2031 target as only about 13% share of its own $235 billion projected custom TAM — a level analyst Tore Svanberg flagged as conservative given Marvell's design-win hit rate, to which Murphy responded that the company is being "pragmatic," not aggressive, particularly in how it has modeled the value of the Google warrant agreement.
Total Addressable Market Quadruples to $400 Billion
Marvell now sizes its 2030 data center and communications TAM at approximately $400 billion, a more than 4x increase from the under-$100 billion figure the company last presented. Within that, interconnect TAM is expected to grow at roughly 65% annually to $65 billion, switching and storage at 40% to $85 billion, and custom at 55% to $235 billion. The company's own revenue is modeled to grow faster than the TAM (45% CAGR) and faster than underlying data center capex, which Marvell now assumes reaches roughly $3 trillion by 2030 at a 35% CAGR from 2025, with growth decelerating from the current 70%-plus pace toward roughly 20% in the outer years. Murphy was explicit that this capex moderation is baked into estimates as a hedge: "We think we've built in our model just a lot of cushion for the AI business to kind of digest over time."
Scale-Up Optics Emerges as a New, Nearly Zero-Base Revenue Stream
Perhaps the most underappreciated shift is the speed at which scale-up networking — connectivity between accelerators within and across racks — is moving from copper to optics. Dave Lazovsky, who joined Marvell through the Celestial AI acquisition earlier this year, explained the physics driving this: at 200 gigabits per second, copper's reach is limited to about 2.5 meters, "it forces you inside the rack." With memory-hungry reasoning and KV cache-driven inference workloads pushing hyperscalers toward multi-rack scale-up domains of 512 to 576-plus XPUs, optics becomes unavoidable. Marvell has essentially zero revenue in scale-up optics today but expects "multi-hundred million" dollars starting next year with an aggressive ramp after that. Management pointed to dollar content of "thousands of dollars of scale-up optics per XPU" since optics are needed on both the XPU and switch sides of each link. Xi Wang, who runs the connectivity business, also disclosed a newer, longer-dated opportunity called "scale-in" — bringing optics inside the package itself using next-generation light sources like microLEDs and micro-VCSELs to replace electrical die-to-die interfaces. Murphy confirmed this is excluded entirely from the fiscal 2031 model: "We didn't want to ignore it because I actually think... scale-in will be another massive meaningful leg of growth to Marvell... adding an entirely new basically fourth leg."
New CFO Introduces "$30 in '30" Earnings Framework and Raised Margin Targets
New CFO Dan Durn, three months into the role after a career that included Goldman Sachs' semiconductor M&A group and more than a decade as a public company CFO, used the event to set a fresh long-term financial model: gross margin of 56% to 59%, operating margin of 44% to 46% (six points above the company's prior long-term target), a 15% tax rate, and free cash flow margin above 36%. The combination implies non-GAAP EPS exceeding $30 per share in fiscal 2031 — a figure Durn branded "$30 in '30." He noted the gross margin range is mix-dependent and effectively self-hedging: the low end would require custom silicon to significantly outperform the base case, while connectivity outperformance would push margins toward the upper end. Durn also highlighted that operating expenses are expected to grow at roughly half the rate of revenue, driving the margin expansion.
Near-Term Guidance Raised Again, Driven by Connectivity
Beyond the multi-year targets, Marvell raised its fiscal 2028 total revenue guidance to approximately $20 billion (67% year-over-year growth), up from the $18 billion outlook given in August, with data center revenue growing approximately 80%. Murphy said the incremental $2 billion versus the August guide is being driven by the same connectivity trends: accelerating scale-out optics, growing scale-up optics, and a switching portfolio "firing on all cylinders." Notably, management said the company's original fiscal 2029 data center revenue target of $18 billion — set a year ago after already being raised from $15 billion — will now be hit almost a full year early, in fiscal 2028.
Supply Chain and Customer Concentration Addressed Directly
COO Chris Koopmans used his presentation to preempt two of the most common bear concerns: supply constraints and customer concentration. He detailed Marvell's practice of giving key suppliers five-year capacity forecasts, visited in person by Koopmans and Murphy, well ahead of locking in capital for capacity reservations: "That fiscal '31 forecast of $70 billion to $90 billion that Matt put up, that might have been a surprise to some of you, not to our suppliers." On concentration, Koopmans noted the top four U.S. hyperscalers alone can deliver "the vast majority" of the company's targets, with more than 35 distinct sockets worth millions of dollars each at just those four customers, explicitly stating "we aren't counting on some major new customer win or a new home run design win" to hit the model. On the question of whether hyperscaler capex forecasts might be inflated — a classic cyclical risk raised by analyst Mark Lipacis — Murphy pushed back by noting the five-year framework is really closer to a three-year call given existing visibility into fiscal 2027 and 2028 revenue, and that the company has built capex deceleration and a range of outcomes into the model as protection.