Nvidia is investing $3.5 billion in MediaTek and giving the Taiwanese chip designer access to technology for building custom AI accelerators that plug into Nvidia-based data centers. The agreement centers on NVLink Fusion, which lets non-Nvidia processors communicate quickly with other chips in the same rack-scale system.
Large cloud and AI companies including Amazon, Google, Microsoft, OpenAI and Anthropic are designing more of their own silicon to reduce dependence on Nvidia GPUs. By supplying the networking and system architecture around those custom chips, Nvidia can remain part of the infrastructure even when another company provides the main processor.
MediaTek has been expanding from consumer chips into custom data-center application-specific integrated circuits. It said in June that the business could generate $2 billion in revenue during 2026, although it has not disclosed its custom AI-chip customers. The partnership gives those customers access to Nvidia’s scale-up and scale-out stack.
The companies will also continue work on compact AI computers, including DGX Spark and RTX Spark, and on automotive platforms that combine MediaTek systems with Nvidia graphics and Drive AGX technology. The investment therefore covers more than one chip project, but its strategic core is clear: Nvidia is opening its interconnect ecosystem so custom silicon can grow without displacing Nvidia from the surrounding AI factory.