Table of Contents
Marvell CEO Matthew Murphy said AI workloads are driving more data movement and making efficient connectivity a fundamental constraint on AI infrastructure
In sum – what to know
Connectivity becomes critical – Marvell CEO Matthew Murphy said AI workloads are driving more data movement and making efficient connectivity a fundamental constraint on AI infrastructure.
Optics moves inward – Rising bandwidth requirements are pushing optical connectivity from longer-distance links toward racks and increasingly closer to compute.
XPU attach expands – Marvell expects its custom business to more than triple from FY ’28 to FY ’29, reaching more than $12 billion as XPU attach expands.
Connectivity becomes an AI infrastructure constraint
Marvell expects connectivity to become a defining layer of AI infrastructure as increasingly complex workloads drive more data movement across systems and push optical technologies closer to compute.
Speaking at Marvell’s 2026 Investor Day, the company’s chief executive officer and chairman Matthew Murphy said Marvell believes the industry is moving into a phase in which connectivity will play a role comparable to the earlier expansion of compute and memory.
According to Murphy, those workloads are creating increasingly diverse traffic patterns across AI infrastructure and raising requirements for connectivity.
“Those workloads are driving more and more data movement and diverse traffic patterns across AI infrastructure,” he said. “That’s creating greater demands on connectivity. Efficient data movement is becoming a fundamental constraint on AI infrastructure.”
Murphy said Marvell is seeing the same shift among major customers, including the world’s largest hyperscalers, which he said are rethinking network architectures as AI infrastructure scales.
Marvell’s investment thesis is that connectivity is now emerging as the next major opportunity in AI infrastructure after compute and memory. “As systems scale, more and more data needs to move across every distance, and Marvell is fundamentally a connectivity-first company,” he said.
That positioning gives Marvell exposure across different AI architectures, Murphy said, because the company works with compute and memory providers as well as cloud operators.
“As AI infrastructure continues to scale, we have exposure across the ecosystem, regardless of which compute or memory architecture is deployed,” he said.
Marvell generated $8.2 billion in revenue in fiscal 2026, with 75% coming from the data center segment. Murphy said about 70% of total revenue was associated with connectivity, compared with roughly 30% tied to compute.
Optical links move closer to compute
The company’s interconnect business is also positioned to benefit as connectivity requirements extend across increasingly shorter distances. “At longer distances like scale-across and scale-out, the industry has already transitioned into optics,” Murphy said. “Marvell is a leader in those markets, and that’s what is driving our interconnect revenue today.”
Inside the rack, however, connections remain largely based on copper. Murphy said increasing bandwidth requirements are changing that equation.
“As bandwidth continues to increase, the distance copper can reach gets shorter and shorter,” he said. “That means connections inside the rack are starting to begin the transition to optics.”
Murphy said that transition creates a new opportunity in scale-up interconnect, including technologies such as near-package optics and co-packaged optics.
“And longer term, the same physics continue to apply,” he said. “As bandwidth keeps increasing, the transitions to optics moves closer and closer to the compute.”
Marvell expects this progression eventually to extend optics into the package, creating what Murphy described as another significant opportunity in scale-in connectivity.
The company is also expanding its switching business into scale-up applications, where accelerators require high-bandwidth, low-latency connectivity.
Custom silicon and XPU attach expand
The other major growth area is custom silicon, particularly what Marvell calls XPU attach. Murphy said XPU deployments are increasingly pulling through functions such as networking, storage, memory, security, and inference acceleration.
Murphy said Marvell’s custom business grew from essentially zero over the past five years to about $1.5 billion last year. The company previously expected custom revenue to more than double in fiscal year 2028 and put it on a path to more than $10 billion in fiscal year 2029.
Marvell has now raised that expectation. “We now expect custom revenue in fiscal year 2029 to more than triple from fiscal year 2028, reaching more than $12 billion and putting us well above our prior $10 billion target,” Murphy said.
Marvell’s market and revenue outlook
Looking at the broader market, Marvell expects total data center capital expenditure to reach roughly $3 trillion by 2030, with accelerated compute representing roughly $1.6 trillion of total addressable market (TAM) opportunity.
Murphy said Marvell sees approximately $385 billion of data-center TAM addressable by the company by 2030, plus approximately $15 billion in communications and other markets, for a total addressable market of approximately $400 billion.
Within that total, Marvell expects interconnect to reach approximately $65 billion by 2030, with switching and storage reaching approximately $85 billion and custom reaching approximately $235 billion.
Murphy said the company now expects approximately $20 billion in total revenue in fiscal year 2028, representing about 67% year-over-year growth, with data-center revenue expected to grow approximately 80%.
