In 2016, Matt Murphy inherited a semiconductor company with a very different future from the one it occupies today.
Marvell was closely associated with storage and consumer-focused chips. But Murphy saw an opportunity to reposition the business around a more consequential part of the technology stack: the infrastructure required to move, connect and process enormous volumes of data. Under his leadership, Marvell has shifted decisively towards custom silicon, networking and optical connectivity for cloud and AI systems.
That bet has become increasingly significant as AI has pushed computing beyond the limits of individual processors and servers. The challenge is no longer simply building faster chips. It is connecting accelerators, memory and servers efficiently enough to scale entire AI systems.
Marvell has spent years building relationships with the world’s largest cloud companies and now supplies custom silicon across the major US hyperscalers. In 2026, that ecosystem widened further through a $2 billion strategic investment from NVIDIA and an expanded custom-silicon relationship with Google.
It is a striking evolution for a company that once centred its business on storage. Matt Murphy’s achievement has been to identify where the next infrastructure bottleneck would emerge, then build the technology, partnerships and organisation to meet it.
His career gives him an unusually broad view of that process. Before Marvell, Murphy spent 22 years at Maxim Integrated, moving across engineering, sales, marketing and general management. He later chaired the Semiconductor Industry Association and the Global Semiconductor Alliance, giving him a perspective that extends beyond one company or technology cycle.
Today, Murphy brings together decades of semiconductor experience with a close understanding of how the world’s largest technology companies are building AI infrastructure. His central argument is becoming increasingly difficult for the industry to ignore: as AI systems grow, the ability to move data may matter almost as much as the ability to compute it.
