Marvell Technology, Inc., a leader in data infrastructure semiconductor solutions, announced a collaboration with NVIDIA to integrate NVIDIA’s NVLink Fusion technology into Marvell’s custom cloud platform silicon. Unveiled in Santa Clara, California, this partnership aims to provide hyperscalers with flexible, high-performance AI infrastructure, accelerating deployment for advanced AI model training and agentic AI inference.
Marvell and NVIDIA collaborate to offer NVLink Fusion for custom AI infrastructure, announced May 19, 2025.
NVLink Fusion chiplet delivers up to 1.8TB/s bidirectional bandwidth for AI factories.
Marvell’s custom silicon leverages expertise in SerDes, die-to-die interconnects, and advanced packaging.
Enables hyperscalers to scale AI factories to millions of custom XPUs with NVIDIA’s rack-scale systems.
Supports demanding workloads for trillion-parameter AI models.
Available now through Marvell and other NVIDIA partners like MediaTek and Cadence.
NVIDIA’s NVLink Fusion is a new silicon technology that integrates custom XPU (accelerated processing unit) silicon with NVIDIA’s NVLink connectivity, rack-scale architecture, and end-to-end networking platform. Its core chiplet provides up to 1.8TB/s of bidirectional bandwidth, offering 14 times the bandwidth of PCIe Gen5, enabling seamless scaling of AI factories to millions of custom XPUs.
Marvell’s custom platform strategy combines its expertise in:
Serializer/deserializers (SerDes) and die-to-die interconnects for 2D/3D devices.
Silicon photonics, co-packaged copper, and advanced packaging.
Custom high-bandwidth memory (HBM), system-on-chip (SoC) fabrics, and PCIe Gen 7 interfaces.
This allows Marvell to create tailored platforms that enhance infrastructure performance and efficiency, meeting the rigorous demands of AI model training and inference. “Marvell and NVIDIA are working together to advance AI factory integration,” said Nick Kucharewski, SVP and GM of Marvell’s Cloud Platform Business Unit.
The collaboration enables hyperscalers to deploy custom scale-up solutions rapidly, leveraging Marvell’s custom silicon and NVIDIA’s NVLink architecture. Key advantages include:
Flexibility: Hyperscalers can integrate proprietary XPUs into NVIDIA’s rack-scale systems, preserving existing NVLink investments.
Scalability: Supports scaling AI factories to millions of XPUs, addressing trillion-parameter AI models.
Performance: 1.8TB/s bandwidth ensures high-speed data transfers for compute-intensive AI workloads.
Efficiency: Reduces time-to-deployment by standardizing heterogeneous silicon on NVIDIA’s rack architecture.
“The computing landscape is being reshaped as AI is foundational to modern data centers,” said Shar Narasimhan, Director of Accelerated Computing at NVIDIA. NVLink Fusion extends NVIDIA’s platform to partners like Marvell, enabling hyperscalers to meet growing AI demands.
Announced at COMPUTEX in Taipei, NVLink Fusion is adopted by partners including MediaTek, Alchip Technologies, Astera Labs, Synopsys, and Cadence, with Fujitsu and Qualcomm integrating custom CPUs with NVIDIA GPUs. Marvell’s role leverages its 25-year history of delivering semiconductor solutions for data infrastructure, positioning it as a key player in the $100 billion AI infrastructure market.
Posts on X reflect strong sentiment, with @TipRanks noting the partnership’s potential to drive high-performance AI factories. However, competitors like Broadcom, AMD, and Intel, part of the Ultra Accelerator Link (UALink) consortium, are pushing open-standard interconnects to challenge NVLink’s proprietary approach, indicating a competitive landscape.
Marvell’s collaboration with NVIDIA positions it to capitalize on the surging demand for AI infrastructure, projected to double data center power consumption by 2026. By offering custom silicon with NVLink Fusion, Marvell enables cloud providers to build agile, scalable AI factories. “Marvell custom silicon with NVLink Fusion gives customers a flexible, high-performance foundation,” said Matt Murphy, Marvell’s Chairman and CEO.
NVLink Fusion silicon design services are available now from Marvell and other partners, with further details shared at NVIDIA’s GTC Taipei and COMPUTEX keynotes.
To deliver the data infrastructure technology that connects the world, we're building solutions on the most powerful foundation: our partnerships with our customers. Trusted by the world's leading technology companies for over 25 years, we move, store, process and secure the world's data with semiconductor solutions designed for our customers' current needs and future ambitions. Through a process of deep collaboration and transparency, we're ultimately changing the way tomorrow's enterprise, cloud, automotive, and carrier architectures transform for the better.
TechIntelPro empowers B2B technology decision-makers with the latest industry trends, deep insights, leadership perspectives and actionable intelligence to drive both organizational and professional milestones—and thrive on the tech pulse.