Why Marvell Expects Supporting Chips to Have Richer Margins Than Custom Accelerators

During Marvell's Investor Day Q&A, analyst Harlan Sur connected two questions: if attach products contain more Marvell IP, can they reach the market faster and carry richer margins? CEO Matt Murphy said the company expects the XPU-attach side to be more gross-margin rich, citing its IP, differentiation and volume. That is management's product-economics explanation, not a published comparison of actual product-level margins.

XPU refers to a custom processor such as an accelerator; XPU attach describes chips working around it. A cloud customer may need a network controller tailored to its data paths, or chips extending memory and connecting storage. Marvell can adapt a technology core from an existing volume-shipped product to the customer's architecture. In many such semi-custom programs, it says it works with the customer on architecture and leads front-end design around its existing IP.

Reuse creates an opportunity to sell an existing technology base across customers and product generations. Murphy said existing cores can be tailored quickly, whereas an XPU requiring a new process and package may take longer. The comparison is conditional: it does not establish that every accelerator starts over, or that volume-shipped IP has already been fully amortized. An attach function may persist across accelerator generations, but continued customer adoption still matters.

Gross margin measures what remains from each dollar of revenue after cost of sales. Gross-profit dollars also depend on price and volume, while operating profit incorporates expenses such as R&D. A richer expected attach margin therefore does not by itself establish more profit per chip or program. Buyers may seek a share of the economic benefit of reuse, and competition or adaptation work can affect what Marvell retains. These are possible constraints, not disclosed outcomes of specific negotiations.

At the same event, CFO Dan Durn described custom silicon as carrying relatively lower gross margins, with customer-funded non-recurring engineering and scale supporting its operating-margin profile. That discussion compares custom business with the company's other product mix. Murphy was comparing attach products with full-custom accelerators. The reference points differ. The CFO's aggregate category cannot simply be reassigned to full-custom accelerators alone, and neither statement supplies actual margins for the two subcategories.

The consolidated statements offer a separate observation. FY2027 Q1 revenue of $2.4178 billion and gross profit of $1.2608 billion produce a margin of approximately 52.1%; Q2 revenue of $2.7393 billion and gross profit of $1.4556 billion produce about 53.1%. These figures cover the whole company. Without a product revenue and gross-profit split, the movement cannot be attributed to attach. Long-term non-GAAP targets cannot replace same-period, same-basis product actuals either.

IP reuse, design adoption, production and recurring revenue are distinct developments. More discriminating evidence would include named attach-product ramps, attach's contribution to custom revenue, and profit changes explainable on a consistent basis. A rise in consolidated gross margin alone cannot separate the effects of mix, pricing and accounting adjustments.

Marvell's proposed opportunity is specific: sell more supporting functions built on an existing technology base alongside large accelerators, and seek a richer margin through differentiation and scale. That explains the product choice while leaving gross-margin percentage, gross-profit dollars and operating profit as separate measures to track. Product-level or ramp disclosures could move today's management explanation toward a more direct operating test.