With the hyperscalers developing their own ASIC custom silicon for datacentres, as an alternative to buying Nvidia GPUs, Nvidia aims to ensure that Marvell’s custom chips can remain compatible with Nvidia’s datacentre hardware.
Marvell is currently working with AWS, Google, Meta and and Microsoft in developing custom AI XPUs and CPUs
The Marvell-Nvidia partnership revolves around NVLink Fusion which enables developers to design semi-custom AI silicon using the NVLink interconnect to connect to Nvidia datacentre systems.
The companies will also work to develop Nvidia’s Aerial AI-RAN for 5G/6G networks including the development of optical interconnect and silicon photonics technology.
“Together with Marvell, we are enabling customers to leverage Nvidia’s AI infrastructure ecosystem and scale to build specialised AI compute,” says Nvidia CEO Jensen Huang.
“By connecting Marvell’s leadership in high-performance analogue, optical DSP, silicon photonics and custom silicon to Nvidia’s expanding AI ecosystem through NVLink Fusion, we are enabling customers to build scalable, efficient AI infrastructure,” says Marvell CEO Matt Murphy.
