TMTB: Citi conference takeaways - MRVL DDOG

MRVL

Main presenters: Matt Murphy (Chairman & CEO) and Willem Meintjes (EVP & CFO).

Summary: Marvell clarified that its tougher stance on customer confidentiality should not be read as a change in outlook; the company is tracking ahead of its $1B+ custom AI target this year and still planning for growth. Management reaffirmed a $55B+ custom TAM ($40B XPU / $15B attach) with 18+ sockets (and more closed since AI Day), while near‑term momentum in optics continues as lanes move 200G→400G. Operating leverage is pushing OM toward the ~38–40% long‑term model despite 58–59% GM pressure from custom ramps.

---

Key Takeaways

• Guidance & tone: CEO addressed investor concerns: Marvell “rotated harder” to protect customer trade secrets, which was not intended to signal a business change. Fiscal ’27 outlook will be given when visibility is appropriate; for now, the company remains ahead of its $1B+ custom revenue goal this year and continues to plan for growth.

• AI custom pipeline/TAM: At AI Day, Marvell sized custom TAM at ~$55B (≈$40B XPU + $15B attach) and now counts 18+ sockets; the team has won additional programs since, and is tracking ~50 more sockets worth $75B+ lifetime revenue (some already closed). Long‑term, Marvell targets ~20% share of both overall and custom TAM.

• Market structure: Custom silicon won’t be winner‑take‑all; multiple engagement models will coexist, though the majority of units should come from full‑turnkey providers. Management noted custom spend could exceed the entire x86 CPU market by 2028.

• Margins & investment: Custom ramps hold GM to ~58–59%, but OM has scaled to ~36.2%, approaching the 38–40% long‑term target. NRE is recognized as contra‑OpEx; R&D has been “turbocharged,” with >80% of total R&D focused on data center/AI.

• Optics/DSP momentum: Near‑term double‑digit growth expected; quarterly noise reflects the module manufacturing layer in the supply chain. Technology cadence is advancing from 100G→200G→400G per lane, enabling 800G, 1.6T, 3.2T roadmaps.

• Networking scale‑out & scale‑up: Post‑Innovium (2021), Marvell is positioned for 51.2T switching transitions (from 12.8T) and is leveraging SerDes + high‑performance switching for scale‑up solutions that pair with its IO/XPU roadmaps.

• Beyond the top 4 hyperscalers: The company is engaging sovereign and “second‑tier” cloud providers; optics cuts across all, and these customers are increasingly pursuing bespoke custom silicon for differentiated workloads.

• Capital allocation: With steadier FCF and recent asset‑sale proceeds, Marvell plans to step up capital returns while keeping flexibility for bolt‑on M&A to accelerate the portfolio.

• Photonics & pluggables: Marvell ships high‑volume SiPho for DCI today and has an integrated light‑engine path; inside‑DC silicon‑photonics/CPO mass adoption remains years out (timing uncertain). Near term, it’s “all pluggables,” with LPO starting to see adoption, and Marvell offering a one‑stop range across pluggables, AOCs, and AECs.

---

Q&A Highlights

• What changed on custom/XPU? Nothing fundamental—Marvell tightened confidentiality due to program sensitivity; it remains ahead of its $1B+ custom target this year and will update when ’27 visibility firms.

• How to ~20% share? Combination of a bottoms‑up view on 18+ wins, a 50‑socket pipeline (with some already closed), and the company’s process/package/IP and supply‑chain strengths.

• Is custom winner‑take‑all? No—multiple models will exist, but full‑service turnkey providers should command most shipments; custom spend could top the x86 CPU market by 2028.

• Profit model through ramps: Expect GM pressure from custom, but OM leverage is building (guided ~36.2% vs. 38–40% LT). NRE offsets OpEx; R&D remains heavily tilted to AI/cloud.

• Optics outlook: Some timing noise, but demand signals remain strong; moving to 200G/400G per lane underpins 800G–3.2T transitions.

• Capital returns vs. M&A: With healthier FCF and sale proceeds, management will increase returns but keep dry powder for bolt‑ons that accelerate growth.

• Non‑hyperscaler AI CapEx: Active engagements with sovereign and emerging cloud players; as their spend rises, needs get more bespoke, expanding Marvell’s custom silicon opportunity.

• CPO vs. pluggables: CPO inside the DC is longer‑dated; the next several years remain pluggable‑led, with LPO adoption starting and Marvell participating across LPO/AOC/AEC.

---

DDOG

Main presenter: David Obstler (CFO)