At Computex Taipei, Marvell outlined its new AI infrastructure strategy, focusing on optical interconnect technology to link multiple data centers across thousands of kilometers. The approach pools distributed compute, storage, and networking resources into a unified pool, allowing cloud providers to flexibly allocate hardware based on AI workload demands – improving utilization and resilience.
CEO Matt Murphy said optical interconnect is fundamental to cross‑region data center coordination and will see expanding deployments.
Marvell introduced two key hardware components: the Ara 1.6 Tb/s interconnect built on 3nm DSP , and the Teralynx T100 Ethernet switch , which delivers 102.4 Tb/s aggregate bandwidth and supports flexible port configurations – up to 512 x 200 Gb/s or 64 x 1.6 Tb/s .
ICgoodFind： Marvell’s optical interconnect and Teralynx switch target the growing need for agile, high‑bandwidth AI cluster networking across regions.
