NVIDIA introduced NVLink Fusion, a new silicon platform designed to empower partners to build semi-custom AI infrastructure by tightly coupling their chips with NVIDIA GPUs and networking systems. Revealed at COMPUTEX 2025, NVLink Fusion enables a wide range of silicon and data center vendors—including MediaTek, Marvell, Alchip Technologies, Astera Labs, Synopsys, and Cadence—to develop AI-optimized silicon using NVIDIA’s advanced interconnect fabric. Fujitsu and Qualcomm are among the first major CPU vendors to announce plans to integrate their custom CPUs with NVIDIA GPUs using NVLink Fusion, aiming to create sovereign, scalable, and power-efficient AI infrastructure.
NVLink Fusion allows for scale-up architectures with up to 1.8 TB/s bandwidth per GPU—14 times faster than PCIe Gen5—and a scale-out path via NVIDIA’s Spectrum-X Ethernet and Quantum-X800 InfiniBand platforms. The solution is especially tailored for model training and agentic AI inference, with design flexibility that extends to custom ASICs and CPUs. Partner companies emphasized their contributions ranging from ASIC design, chiplet IP, and interconnects to full rack-level deployments. AI data centers built with NVLink Fusion will be orchestrated using NVIDIA Mission Control, a software stack designed to automate infrastructure deployment, validation, and workload management.
Fujitsu’s new Arm-based 2nm CPU, MONAKA, and Qualcomm’s custom CPUs will be the first to directly connect to NVIDIA GPUs through NVLink Fusion. NVIDIA aims to simplify the buildout of massive AI factories by opening its full-stack AI infrastructure—including SuperNICs, rack-scale systems, and orchestration software—for partner integration. This move positions NVIDIA as not just a chipmaker, but a platform enabler for an ecosystem of AI infrastructure developers.
- NVIDIA unveils NVLink Fusion to support semi-custom AI infrastructure with partner silicon.
- MediaTek, Marvell, Alchip, Astera Labs, Synopsys, and Cadence to offer design services and IP for custom silicon.
- Fujitsu and Qualcomm to couple their CPUs with NVIDIA GPUs in rack-scale deployments.
- NVLink Fusion enables 1.8 TB/s GPU bandwidth—14x PCIe Gen5—and supports Spectrum-X Ethernet and Quantum-X800 InfiniBand.
- NVIDIA Mission Control software automates orchestration of AI factories.
“A tectonic shift is underway: for the first time in decades, data centers must be fundamentally rearchitected — AI is being fused into every computing platform,” said Jensen Huang, founder and CEO of NVIDIA. “NVLink Fusion opens NVIDIA’s AI platform and rich ecosystem for partners to build specialized AI infrastructures.”