• The NVLink Fusion announcement a year ago introduced a rack-scale platform designed to let customers develop heterogeneous semi-custom AI infrastructure
• Its partnership with Marvell will ensure tight integration of Marvell-built custom semi-silicon XPUs to air directly on the same high-speed fabric as NVIDIA’s CPUs, GPUs, and DPUs
• Essentially, this opens up NVIDIA’s proprietary interconnect to partner-designed chips, providing a ‘standardized’ ramp to NVIDIA’s AI factories

• Networking is the next frontier for scaling the AI infrastructure. SiPh and Optical interconnects interoperability and capability will play a crucial role
• The partnership with Marvell will ensure seamless integration of Marvell’s optical connectivity directly with NVIDIA’s NVLink and Spectrum-X switching platforms. A great win for both parties
• The partnership places a heavy emphasis on silicon photonics and advanced optical DSPs

• Bringing AI from RAN to core is the current big infrastructure investment for telcos and network infrastructure partners
• At GTC, we saw multiple telcos from Comcast, AT&T, PT, Indosat, Tbk, Spectrum, T-Mobile, Akamai Technologies, Nokia, and others are partnering with NVIDIA for AI Grid to bring powerful GPU-accelerated distributed inference infrastructure
• Marvell Technology has been a leading player in building silicon for the world's network infrastructure players in baseband processing and RAN silicon. With this partnership, they aim to combine this with NVIDIA’s AI software stack to create high-performance, power-efficient carrier-grade infrastructure

This puts strong pressure on other leaders, including AMD and ROADCOM.

For more insights, view our recent report: Global AI Server Compute ASIC CPU Attach Shipment by Chip Model, Hyperscalar, Design Partner & Foundry Forecast, Mar 2026