Marvell Technology's 1.6T PAM4 digital signal processors account for nearly 40% of the total bill of materials of a high-speed optical transceiver. That concentration — one company's silicon representing four out of every ten dollars in a critical AI data center component — is not accidental. It is the outcome of two decades of portfolio restructuring and the deliberate abandonment of lower-margin businesses in mobile, consumer storage, and commodity networking. The company generated $5.56 billion in revenue in fiscal year 2024, down from $6.5 billion the year prior, while its market capitalization sat at $72 billion. The revenue-to-market-cap ratio implies that investors are not valuing what Marvell makes today — they are pricing what it will make in fiscal 2025 and 2026, when AI-related data center revenue is projected to exceed $2 billion. Data center already accounted for over 65% of fiscal 2024 revenue, a structural shift from the company's historical dependence on enterprise networking. Marvell operates as the number two player in the hyperscale custom ASIC market behind Broadcom. The company designs custom XPUs — specialized processor accelerators — for Amazon and Google, the two largest hyperscalers investing most aggressively in custom silicon as an alternative to merchant GPUs. Each custom silicon design win requires over $1.5 billion in non-recurring engineering costs and mask set fees at TSMC's 3nm process node. That barrier is not a problem for Marvell; it is the moat. The company was founded in 1995 by Sehat Sutardja, Weili Dai, and Pantas Sutardja. CEO Matt Murphy, who arrived in 2016, has executed the portfolio transformation that shed mobile and consumer businesses and concentrated resources on data center and AI infrastructure.