NVIDIA introduced NVLink Fusion™, a new silicon technology designed to help companies build custom AI systems using NVIDIA’s advanced NVLink computing fabric. This innovation allows major industry players to create powerful AI infrastructure tailored for demanding workloads like AI model training and agentic AI inference.

Top tech companies such as MediaTek, Marvell, Alchip Technologies, Astera Labs, Synopsys, and Cadence are among the first partners adopting NVLink Fusion. This technology helps integrate custom chips with NVIDIA GPUs to create high-performance AI data centers, known as AI factories. Fujitsu and Qualcomm also use NVLink Fusion to combine their CPUs with NVIDIA GPUs, boosting AI capabilities in rack-scale systems.

Jensen Huang, NVIDIA’s CEO, said, A major transformation is happening data centers need to be redesigned as AI becomes part of every computing platform. NVLink Fusion opens up NVIDIA’s AI ecosystem, enabling partners to build specialized AI infrastructure.

NVLink Fusion supports cloud providers in scaling AI factories to millions of GPUs using various ASICs and NVIDIA’s full networking platform, which offers up to 800Gb/s throughput. This includes NVIDIA ConnectX-8 SuperNICs, Spectrum-X Ethernet, and Quantum-X800 InfiniBand switches with co-packaged optics coming soon.

Industry Leaders Back NVLink Fusion

• MediaTek is expanding its collaboration with NVIDIA to build scalable and flexible AI infrastructure beyond automotive, addressing cloud-scale AI needs.
• Marvell sees NVLink Fusion as a foundation for advanced AI systems that handle trillion-parameter AI models with high performance and agility.
• Alchip Technologies supports NVLink Fusion by providing advanced chip design and manufacturing to enable efficient AI model training and deployment.
• Astera Labs focuses on fast, low-latency interconnect solutions critical for maximizing AI server performance.
• Synopsys supplies AI chip design tools and interface IP that support NVIDIA NVLink Fusion’s scalable ecosystem.
• Cadence delivers comprehensive IP portfolios that complement NVIDIA’s technology to build powerful and energy-efficient AI data centers.

Fujitsu’s next-gen FUJITSU-MONAKA CPU, built on 2-nanometer tech, and Qualcomm’s advanced custom CPUs both integrate seamlessly with NVIDIA GPUs using NVLink Fusion, driving efficient and high-performance AI computing.

NVIDIA’s NVLink Platform Leads the Industry

The fifth-generation NVIDIA NVLink platform offers massive data transfer speeds—up to 1.8 TB/s per GPU, which is 14 times faster than PCIe Gen5. Leading cloud providers are already using NVIDIA NVLink racks and can easily adopt NVLink Fusion to create flexible AI data centers combining different silicon chips.

Software to Manage AI Factories

NVIDIA Mission Control™, a unified software platform, manages AI data centers powered by NVLink Fusion. It automates deployment, infrastructure validation, and workload orchestration to help enterprises run cutting-edge AI models faster and more efficiently.

Availability

NVLink Fusion silicon and solutions are now available through partners MediaTek, Marvell, Alchip, Astera Labs, Synopsys, and Cadence.

For more information, watch NVIDIA CEO Jensen Huang’s keynote at COMPUTEX or visit NVIDIA GTC Taipei.