Marvell Technology CEO Matt Murphy has identified connectivity as emerging as the next central constraint in artificial intelligence infrastructure as computing, memory and storage systems scale across chips, racks, clusters and data centres.

Speaking at The Six Five Summit 2026, Murphy stated that the industry's initial focus on accelerators, such as GPUs and XPUs, has been followed by a cycle in memory and storage. He noted that the next challenge involves connecting these resources within and across data centres.

Marvell is positioning itself for expansion in networking, optical connectivity, custom silicon and memory expansion. The company's data infrastructure segment represented less than 10% of revenue a decade ago and is expected to account for more than 80% of future revenue. Murphy reported that Marvell currently sees over $2 billion in quarterly AI-related data-centre activity.

Memory and custom silicon opportunities

Murphy highlighted several growth areas for the company:

• Memory expansion: CXL (Compute Express Link) products could become a billion-dollar to multibillion-dollar business as customers seek to use memory more efficiently.
• Custom silicon: Custom AI accelerators are being deployed for training and inference workloads, alongside merchant silicon.
• XPU attach: Marvell has secured more than a dozen designs for customised networking interfaces, memory expansion and security components surrounding AI processors.

Regarding optical networking, Murphy noted that the company's acquisition of Inphi provided silicon photonics capabilities, while the acquisition of Celestial AI expanded co-packaged optics technology. He expects near-package optics to see larger adoption before co-packaged optics.

"The AI application became kind of the killer app of data infrastructure," Murphy said. "We are still at the very, very early stages of the deployments."