eta consulting Atlanta — Reno · 2016 →

Marvell Technology

Aerospace, Defence & Technology · Founded 1995 · Santa Clara, California · Listed Nasdaq: MRVL · ~7,500 employees

Marvell Technology was founded in Santa Clara in 1995 by three engineers — the brothers Sehat and Pantas Sutardja, and Sehat’s wife Weili Dai — on the specific premise that the read-channel silicon inside a hard-disk drive, still designed at the time in bipolar and BiCMOS processes, could be rebuilt in mainstream CMOS. Seagate Technology became the founding customer and the CMOS read channel became the founding product. Fabless from the beginning, Marvell listed on Nasdaq under MRVL near the end of the dot-com bubble in June 2000 and has, over the three decades since, moved through storage, then embedded networking, then wireless and Ethernet, and — in the current data-centre era — into custom silicon for hyperscale AI and cloud infrastructure. The firm now runs at roughly seven and a half thousand employees, over ten thousand patents, and around eight billion dollars of annual revenue.

The company designs custom and standard-product semiconductors across four principal end markets: data-centre infrastructure (custom-silicon AI accelerators, high-speed connectivity chips, optical DSPs, and switching silicon for hyperscale cloud infrastructure); enterprise networking (Ethernet switches and controllers, security processors); carrier infrastructure (5G base-station silicon, transport-network devices); and automotive and industrial (in-vehicle networking, industrial-Ethernet, safety and security processors). Manufacturing is contracted out — Marvell has been fabless throughout its history — with design, verification, and system-integration work concentrated in Santa Clara and at a network of international design centres.

What distinguishes Marvell among fabless semiconductor firms of its scale is the shift the company has been making from a merchant-silicon business into a bespoke-silicon business for the largest cloud and AI customers. Custom silicon for hyperscalers — an ASIC designed specifically for a particular AI training or inference workload, in partnership with the hyperscaler whose workload it will run — is a category that did not meaningfully exist a decade ago and now underwrites a significant share of Marvell’s forward pipeline. The firm’s founding read-channel engagement with Seagate turns out, thirty years later, to have been the first example of a pattern Marvell has since built its identity around: the deep-collaboration semiconductor project rather than the merchant catalogue.

Visit Marvell marvell.com