Marvell CEO calls connectivity the next AI bottleneck, as Nvidia's CEO proclaims Marvell the "next trillion dollar company" - Data Center Dynamics

At Computex, Marvell CEO Matt Murphy highlighted that while AI has driven demand for compute and memory, the next bottleneck lies in data center connectivity. He noted that the industry has sequential... Read original →

The AI race has shifted from raw compute to system-level interconnect efficiency. Marvell’s focus on high-speed SerDes, optical I/O, and co-packaged optics (CPO) is now critical to overcoming GPU cluster communication bottlenecks—forcing upstream advanced packaging (e.g., CoWoS) and downstream liquid cooling to evolve in lockstep. Geopolitical pressures are accelerating U.S.-EU efforts to reshore supply chains, yet Marvell’s design centers in Israel and Taiwan, China may face heightened export controls, increasing compliance overhead. Broadcom, with its Tomahawk switches, remains dominant, compelling Marvell to deepen NVIDIA ecosystem integration; meanwhile, Intel and Ampere could leverage open standards like UCIe as a counterplay. Over the next 18 months, 800G/1.6T optical modules will surge, driving demand for silicon photonics and DSPs. Connectivity is no longer ancillary—it defines the scalability ceiling of AI infrastructure, redirecting capital from 'compute arms races' to 'bandwidth foundations.'