The NVIDIA HGX B300 NVL16 Will Disrupt the PCIe Retimer, PCIe switch, and AI NIC Markets

NVIDIA is making a strong move here against Broadcom, Astera Labs, and Marvell. It may, however, benefit cabling companies

On the STH main site, we explored how the NVIDIA HGX B300 NVL16 differs from previous generations of the HGX platform. Moving NVLink switches to the center of the GPUs happened with the HGX B200. The other changes, however, will have a significant impact on other silicon providers as NVIDIA’s design direction is boxing out others from integrating into the platform. In this Substack, we are going to get into what is getting displaced, and some of the system architectural impacts of the displacement.

The NVIDIA HGX B300 NVL16 Will Disrupt the PCIe Retimer, PCIe switch, and AI NIC Markets

Starting off, let us quickly level set on the major changes. The center area, we already discussed has the NVLink switches, but underneath the heatsinks, we no longer have a single GPU, as has been the case for the entire NVIDIA HGX SXM family’s lifecycle.