MediaTek collaborates with NVIDIA to develop custom AI silicon
At Computex 2025, MediaTek confirmed it is working closely with NVIDIA to develop custom AI silicon designed to meet the demands of hyperscale Cloud infrastructure.
The announcement, made by Vice Chairman and CEO Rick Tsai, revealed that MediaTek is among the first adopters of NVIDIA’s NVLink Fusion – a new silicon technology that aims to standardise and scale AI performance across diverse hardware.
NVLink Fusion: a new model for AI data centre infrastructure
NVLink Fusion is a silicon architecture developed by NVIDIA that enables hyperscalers to build custom compute platforms using NVIDIA’s high-speed interconnect ecosystem. It supports flexible integration with various processing units, including NVIDIA GPUs, Grace CPUs, and other custom accelerators. With its rack-scale design, co-packaged optics switches, and control software, the technology is designed to allow different silicon types to operate efficiently within a common infrastructure standard.
It aims to support intensive workloads such as AI model training and real-time inference by simplifying integration and improving performance at scale. NVLink Fusion is supported by NVIDIA’s long-term product roadmap and ecosystem partnerships, enabling faster deployment of custom silicon in Cloud environments.
MediaTek’s role in NVLink Fusion
MediaTek brings its established background in ASIC design and high-speed interconnects to the partnership. The company’s work in this space is central to creating custom silicon solutions for data centres looking to integrate NVLink Fusion-based platforms. The collaboration also builds on previous joint efforts with NVIDIA, including work on automotive and IoT products, as well as the Grace Blackwell GB10 superchip, used in the DGX Spark personal AI supercomputer.
MediaTek’s contributions to NVLink Fusion focus on:
- ASIC design and system integration: the company provides application-specific design services tailored for complex AI workloads
- IP portfolio: MediaTek brings a wide range of technology IP, including SerDes, high-speed I/O, memory integration, and die-to-die interconnects, which are critical for building scalable AI systems
- Ecosystem alignment: the company maintains partnerships with key technology providers such as TSMC, Cadence, Synopsys, and memory suppliers to support advanced process nodes and packaging technologies
- Flexible engagement: MediaTek offers bespoke collaboration models that allow customers to begin system design now, with target tape-outs for early 2026
Enabling hyperscale AI growth
This collaboration positions MediaTek as a key partner for Cloud service providers aiming to scale their AI infrastructure. By integrating its silicon expertise with NVIDIA’s AI connectivity framework, MediaTek supports the development of energy-efficient, scalable AI platforms while reducing development time and cost risk for hyperscale deployments.
A full version of the original blog can be read here.