Nvidia Corp: Marvell to provide custom XPUs; Nvidia to supply supporting technologies

Marvell Technology Group and NVIDIA have announced a collaboration to provide custom solutions for advanced AI infrastructure, combining Marvell’s custom silicon with NVIDIA’s NVLink Fusion technology. This partnership aims to offer greater flexibility and scalability for hyperscalers developing next-generation AI systems. Marvell’s custom cloud platform silicon integrates with NVIDIA’s NVLink connectivity, rack-scale hardware, and end-to-end networking solutions, enabling seamless deployment of AI infrastructure.

The collaboration leverages Marvell’s expertise in semiconductor design and manufacturing, including advanced packaging, high-bandwidth memory (HBM), and system-on-chip (SoC) fabrics. These capabilities are combined with NVIDIA’s NVLink Fusion, which features a chiplet delivering up to 1.8TB/s of bidirectional bandwidth. This integration allows cloud providers to scale AI factories to millions of custom XPUs while leveraging existing architecture investments.

According to Marvell, the partnership accelerates time-to-deployment for AI infrastructure, addressing the growing demand for high-performance computing in AI model training and inference. NVIDIA emphasized that AI is now foundational to modern data centers, and NVLink Fusion expands its open platform to support custom silicon solutions.

This strategic alliance reflects the increasing importance of tailored semiconductor solutions in the AI-driven computing landscape, offering hyperscalers a competitive edge in performance and efficiency.