At Computex 2025 in Taipei, MediaTek outlined its artificial intelligence strategy, spanning both edge devices and cloud infrastructure. Alongside plans for a 2nm chip process, the company also announced an expanded partnership with NVIDIA.
NVIDIA CEO Jensen Huang appeared alongside MediaTek CEO Rick Tsai during Tsai’s keynote to discuss the companies’ joint efforts. As Huang revealed a day earlier, MediaTek is adopting NVIDIA’s NVLink Fusion interconnect to develop custom silicon for AI data centres, drawing on its ASIC design expertise to meet the needs of evolving cloud workloads.
Huang explained that NVLink Fusion enables the development of semi-custom AI infrastructure, not just chips. This approach allows organisations to leverage NVIDIA’s broader ecosystem, especially important as data centres take on increasingly complex AI tasks. Tsai added that NVLink Fusion gives MediaTek the flexibility to build custom AI solutions integrated with NVIDIA’s platforms, helping to handle demanding tasks such as training and inference.
As part of the collaboration, the Taiwan-based chipmaker also highlighted its 20-core CPU architecture for NVIDIA’s DGX Spark workstation that was announced in March. Touted as a personal AI supercomputer, the system uses the jointly developed GB10 Grace Blackwell chip and is designed to run large language models with up to 200 billion parameters locally. The partnership also includes work on an “edge cloud” concept that merges computing resources across radio networks and devices to better support distributed AI workloads.
Additionally, MediaTek showcased its hybrid computing approach, which aims to better integrate processing and communications for AI systems. This concept is intended to improve how AI agents operate and collaborate across different platforms.
