Topic: Custom XPU revenue target for fiscal 2029 and new program progress
Key points:
Fiscal 2029 custom silicon revenue target confirmed at $10+ billion, based on a $55 billion TAM and 20% share (≈$11 billion).
Existing programs, new ramp, and XPU-attached programs all sized up significantly since wins; line of sight to targets remains.
New custom program is on track, hitting all milestones; it represents about one-third of total custom business growth expected next year.
Mgmt stance: Bullish — management explicitly confirmed the $10+ billion target and highlighted strong trajectory from fiscal 2027 to 2029 and beyond.
Q2 — Harlan Sur
Topic: SRAM-based XPU offload engines and design differentiation
Key points:
Marvell has best-in-class SRAM design capability and IP, a legacy from the Avera acquisition and prior IBM/GlobalFoundries roots.
SRAM technology has evolved from networking into AI products, especially for inferencing workloads.
SRAM is one piece of a broader strategy including advanced packaging, high-speed IO, and manufacturing expertise.
Mgmt stance: Bullish — management emphasized SRAM as a key differentiator and a reason for winning XPU-attach designs, with strong trend visibility.
Q3 — Analyst (unidentified)
Topic: Breadth of custom silicon customer base and potential expansion into compute TAM
Key points:
Marvell has custom engagements across all U.S. hyperscalers (some XPU, some XPU+attach, some only attach).
Current fiscal 2029 targets are based on designs already won and locked (as of last summer); newer wins are “insurance policy” and not required.
High-speed IO, SerDes, and die-to-die performance have driven a pronounced increase in activity over the last 6 months.
Mgmt stance: Bullish — management stated the business is inflecting, design activity is “through the roof,” and the roadmap has strengthened since last summer.
Q4 — Michael Arcuri (on behalf of Aaron)
Topic: New XPU/custom socket wins
Key points:
No additional details provided on the type of chip or accelerator.
The opportunity pipeline (50+ types outlined last June) continues to grow in both sockets and dollars; each program is larger than a year ago.
Revenue line is moving in the right direction, validating capabilities.
Mgmt stance: Neutral — management declined to give specifics but confirmed continued progress and pipeline growth.
Q5 — Blayne (analyst)
Topic: CXL opportunity alongside accelerators
Key points:
CXL business is “very real”; Marvell engaged a few years back, initially for traditional x86 server dynamics, now vectored into AI.
XPU-attach line item (including CXL) is expected to exceed $1 billion in revenue in the next couple of years.
Memory cycle concerns are driving additional CXL adoption; trend line continues to size up due to higher CapEx penetration and inferencing growth.