NVIDIA has made a significant strategic shift in its approach to AI infrastructure by unveiling NVLink Fusion, a new silicon technology that opens its proprietary high-speed interconnect to third-party hardware manufacturers.
Announced by CEO Jensen Huang at Computex 2025 in Taipei, NVLink Fusion allows customers and partners to use non-NVIDIA CPUs and GPUs in tandem with NVIDIA's products. "NVLink Fusion is so that you can build semi-custom AI infrastructure, not just semi-custom chips," Huang explained during his keynote presentation.
The technology has already attracted an impressive roster of early adopters. MediaTek, Marvell, Alchip Technologies, Astera Labs, Synopsys, and Cadence are among the first partners to integrate NVLink Fusion into their custom silicon designs. Additionally, Fujitsu and Qualcomm Technologies plan to build custom CPUs that can be coupled with NVIDIA GPUs to create high-performance AI factories.
NVLink Fusion represents a technical breakthrough in AI infrastructure development. The fifth-generation NVLink platform delivers 1.8 TB/s of bidirectional bandwidth per GPU—14 times faster than PCIe Gen5—enabling seamless high-speed communication for complex large models. This performance boost is critical for AI workloads, as every 2x increase in NVLink bandwidth can lead to a 1.3-1.4x improvement in rack-level AI performance.
For cloud providers, NVLink Fusion offers a path to scale out AI factories to millions of GPUs using any ASIC, NVIDIA's rack-scale systems, and NVIDIA's end-to-end networking platform, which delivers up to 800Gb/s of throughput. The entire ecosystem is managed by NVIDIA Mission Control, a unified operations platform that automates the complex management of AI data centers.
This strategic move consolidates NVIDIA's position at the center of AI development by embracing rather than fighting the trend toward custom silicon. By opening its proprietary interconnect technology to partners, NVIDIA is positioning itself as the essential fabric connecting the AI infrastructure of the future—even when those systems aren't built entirely with NVIDIA chips.