NVIDIA Launches NVLink Fusion to Enable Semi-Custom AI Infrastructure at Scale

/ AI, Computer Hardware, Infrastructure
NVIDIA Launches NVLink Fusion to Enable Semi-Custom AI Infrastructure at Scale

The News

At COMPUTEX 2025, NVIDIA announced NVLink Fusion™, a new silicon architecture that enables partners to build semi-custom AI infrastructure by integrating directly into NVIDIA’s NVLink ecosystem. Launch partners include MediaTek, Marvell, Alchip Technologies, Astera Labs, Synopsys, and Cadence, with Fujitsu and Qualcomm Technologies planning to integrate custom CPUs with NVIDIA GPUs using NVLink scale-up and Spectrum-X scale-out technologies. Learn more at NVIDIA.com.

Analysis

NVLink Fusion is more than a connectivity upgrade — it’s a strategic move that modularizes NVIDIA’s AI infrastructure stack, giving silicon partners and cloud providers the tools to build AI factories tailored to their unique needs. As trillion-parameter models push the limits of monolithic systems, NVIDIA’s partner-first, rack-scale approach enables sustainable growth across industries, geographies, and use cases.

With Mission Control software, NVLink bandwidth, and a rapidly growing partner ecosystem, NVIDIA is turning the AI data center into a programmable, customizable, and globally distributed compute fabric for the frontier era of AI.

Opening the NVIDIA AI Platform for Semi-Customization

NVLink Fusion marks a strategic expansion of NVIDIA’s AI infrastructure playbook. By opening its NVLink™ computing fabric to silicon partners, NVIDIA is enabling:

  • Custom chipmakers to scale their own ASICs for high-performance inference and training
  • CPU vendors like Qualcomm and Fujitsu to interconnect proprietary CPUs with NVIDIA GPUs
  • Hyperscalers to adopt NVIDIA’s rack-scale systems without sacrificing architectural choice

This signals a new era of modular AI infrastructure, where compute can be semi-customized for specific performance, latency, and power efficiency needs.

Rack-Scale Architecture and Ecosystem Expansion

The NVLink Fusion rollout supports NVIDIA’s strategy to dominate AI infrastructure from silicon to software to systems. Key components include:

  • NVIDIA GB300 NVL72 and GB200 NVL72 systems, delivering 1.8TB/s bandwidth per GPU (14x PCIe Gen5)
  • ConnectX-8 SuperNICs, Quantum-X800 InfiniBand, and Spectrum-X Ethernet for high-speed I/O
  • NVIDIA Mission Control™, an orchestration suite to manage AI factory operations and workloads

By integrating these elements with third-party silicon, NVIDIA is creating a heterogeneous AI factory blueprint that can serve hyperscalers, sovereign clouds, and industrial R&D hubs alike.

Strategic Partner Commitments

Launch partners provide diverse capabilities:

  • MediaTek: Extending its automotive and datacenter silicon with high-speed interconnects
  • Marvell: Custom ASICs to meet trillion-parameter model demands
  • Alchip: Design + manufacturing ecosystem to scale NVLink Fusion access
  • Astera Labs: Low-latency, memory-semantic interconnect solutions
  • Synopsys & Cadence: EDA and IP ecosystems enabling rapid design and production

These integrations reinforce NVIDIA’s full-stack strategy while accelerating time-to-market for specialized AI infrastructure.

Fujitsu and Qualcomm: Sovereign and Efficient Compute

Fujitsu will integrate its 2nm Arm-based FUJITSU-MONAKA CPU with NVIDIA GPUs, offering sovereign, power-efficient performance for AI factories.

Qualcomm’s custom CPU roadmap, now compatible with NVIDIA rack-scale architecture, brings energy-efficient performance to data center-grade AI deployments — positioning both companies as leaders in next-gen, sustainable compute designs.

Sovereignty and Openness in AI Factories

As AI moves into mission-critical and regulated sectors, NVLink Fusion enables:

  • Vertical integration without vendor lock-in
  • Support for sovereign hardware configurations
  • Interoperability with NVIDIA DGX Cloud Lepton marketplace and tools

These attributes are increasingly important as countries and hyperscalers seek control over foundational AI infrastructure while leveraging best-in-class performance.

Author

  • Paul Nashawaty

    Paul Nashawaty, Practice Leader and Lead Principal Analyst, specializes in application modernization across build, release and operations. With a wealth of expertise in digital transformation initiatives spanning front-end and back-end systems, he also possesses comprehensive knowledge of the underlying infrastructure ecosystem crucial for supporting modernization endeavors. With over 25 years of experience, Paul has a proven track record in implementing effective go-to-market strategies, including the identification of new market channels, the growth and cultivation of partner ecosystems, and the successful execution of strategic plans resulting in positive business outcomes for his clients.

    View all posts