Marvell announced partnership with NVIDIA to add NVLink Fusion technology to custom XPU platform, validating multi-vendor custom silicon strategy and creating new IO interface dependency for hyperscaler designs
“Earlier this month, we announced a partnership with Nvidia, adding their NVLink Fusion technology to our expanding custom platform. Marvel Custom Silicon with NVLink Fusion provides our customers with an accelerated path to custom scaleup solutions, offering greater flexibility and choice in developing next generation AI infrastructure. This announcement further validates the proliferation of custom XPUs as a strong complement to merchant solutions... we could for example work with the customer do the the XPU design with them and then through IO chiplets uh take eye interface between the XPU and the um NVLink chiplet and then that interfaces to the to the scaleup network.”
Why it matters
NVIDIA's decision to license NVLink for third-party custom XPUs (via Marvell) signals that hyperscalers' desire for custom silicon is compelling enough that NVIDIA is adapting its proprietary interconnect strategy. This is a significant shift from a purely merchant model and validates the durability of custom silicon as a multi-year trend.
Investment implication
This validates that custom silicon vendors and integrators (Marvell, potential ASIC partners) will capture disproportionate value as hyperscalers invest in XPU differentiation. NVIDIA's willingness to partner rather than compete head-on suggests it sees custom silicon as complementary, not cannibalistic. Investors should monitor further licensing or chiplet partnerships as evidence of IO standardization pressures.