- Marvell Technology (MRVL, Financial) partners with NVIDIA to enhance AI infrastructure by integrating NVLink Fusion technology.
- The collaboration offers hyperscalers an accelerated deployment pathway, utilizing up to 1.8TB/s of bidirectional bandwidth.
- This partnership leverages Marvell's semiconductor expertise to efficiently scale AI operations to millions of custom XPUs.
Marvell Technology (MRVL) has announced a strategic partnership with NVIDIA to incorporate NVIDIA's NVLink Fusion technology into its custom cloud platform silicon solutions. This integration aims to provide hyperscalers with the capability to accelerate the deployment of custom AI infrastructure efficiently. NVLink Fusion offers a remarkable 1.8TB/s of bidirectional bandwidth, facilitating seamless integration of custom XPU silicon with NVIDIA's rack-scale systems.
The collaboration capitalizes on Marvell's comprehensive semiconductor capabilities, including SerDes, die-to-die interconnects, silicon photonics, and advanced packaging. By integrating these technologies, Marvell aims to deliver custom scale-up solutions that enhance AI model training and inference, supporting AI infrastructure deployment on a significant scale.
For cloud providers, this partnership offers the flexibility to scale AI factories to millions of custom XPUs, effectively integrating proprietary XPUs into NVIDIA's ecosystem. This alignment with NVIDIA's open platform, including the end-to-end networking platform, provides a strategic advantage in meeting the burgeoning demand for AI-driven intelligence in modern data centers.
