The AI infrastructure trade may be getting bigger again. That was one of the clearest messages coming out of Marvell Technology, Inc's
"We expect the emergence of Agentic AI to further supercharge demand," CEO Matt Murphy said.
The company believes the next phase of AI won't simply involve larger models - but AI systems increasingly interacting with other AI systems across sprawling clusters and workflows.
And according to Marvell, that dramatically changes infrastructure requirements.
Marvell Says AI Networking Demand Is Exploding
Murphy repeatedly emphasized that networking is becoming increasingly critical as AI workloads evolve beyond traditional one-shot inference models.
"Networking has become significantly more important," he said.
The company argued newer reasoning models, mixture-of-experts architectures and agentic AI systems generate significantly larger amounts of traffic across AI clusters.
That could benefit Marvell's networking, optical interconnect and custom silicon businesses.
In particular, management highlighted growing demand for scale-out and scale-up networking products, along with high-bandwidth optical connectivity solutions used inside AI data centers.
Marvell also raised multiple long-term growth targets during the call, citing stronger-than-expected AI infrastructure demand.
Marvell Thinks AI Needs More Than Just GPUs
One of the more interesting takeaways from the call was Marvell's argument that agentic AI could significantly expand demand beyond GPUs alone.
"Agentic AI is expected to drive a significant increase in the number of CPUs deployed in AI infrastructure," Murphy said.
That matters because much of Wall Street's AI infrastructure narrative has focused almost entirely on GPUs and compute capacity.
Marvell is instead arguing that networking complexity, memory requirements and data movement increasingly become bottlenecks as AI systems grow larger and more autonomous.
The company believes those trends could create a massive long-term opportunity for interconnect, switching and optical networking technologies.
And while Marvell shares traded lower following earnings, management's comments reinforced a broader idea increasingly spreading across the semiconductor industry:
The AI infrastructure buildout may still be in its early innings.
