NVIDIA and MediaTek announced an expansion of their long-standing partnership to develop AI platforms across data center infrastructure, client computers, and vehicles. Supported by NVIDIA's $3.5 billion investment in MediaTek convertible bonds, the deal pairs NVIDIA's accelerated computing, networking, and graphics software with MediaTek's custom silicon, system-on-chip design, advanced packaging, and connectivity capabilities.

For cloud service providers and frontier model developers, the primary shift centers on data center silicon. MediaTek is adopting the new NVIDIA NVLink Fusion design foundation, allowing customers to build custom multi-die accelerator chips—termed XPUs—designed to connect to NVIDIA's NVLink fabric and NVIDIA MGX rack-scale architectures.

According to NVIDIA, the NVLink Fusion framework provides prebuilt and prequalified building blocks to simplify chip-to-rack manufacturing:

  • An NVLink Fusion chiplet that connects custom XPUs to the scale-up fabric using optical photonics or electrical interconnects.
  • NVIDIA NVLink-C2C interfaces for high-bandwidth, energy-efficient links between customer XPUs, NVIDIA Rosa CPUs, and compatible processors.
  • NVIDIA NVHBM tailored memory integration intended to balance memory bandwidth and energy efficiency while reserving silicon area for compute.

Beyond cloud infrastructure, the companies outlined ongoing collaboration across two local edge segments. In client computing, the partners plan to develop multiple generations of NVIDIA RTX Spark and DGX Spark processors for consumer PCs, workstations, and developer systems, combining NVIDIA GPUs with MediaTek SoCs. For software-defined vehicles, MediaTek continues integrating NVIDIA RTX graphics and AI software into its Dimensity Auto platforms, designed to operate alongside NVIDIA DRIVE AGX.

The joint announcement mixes shipping silicon foundations with multi-year development programs. MediaTek previously partnered with NVIDIA on the GB10 Grace Blackwell Superchip, which links an NVIDIA Blackwell GPU and Grace CPU via NVLink-C2C. Conversely, custom XPUs using NVLink Fusion and future NVIDIA RTX Spark PC chips remain roadmap targets. The announcement omitted delivery timelines, manufacturing schedules, and commercial pricing for upcoming hardware built through the new frameworks.