Nvidia puts MediaTek at the center of its next connected AI chip strategy

Nvidia is expanding its relationship with MediaTek beyond individual components. The companies said August 31 that they will work across AI infrastructure, local computing and automotive systems. Nvidia is also investing $3.5 billion in MediaTek convertible bonds, adding a major financial commitment to a partnership that already covers several device categories.

The technical focus is Nvidia’s NVLink Fusion platform. MediaTek plans to use it to connect custom XPUs to rack-scale AI systems. In plain terms, the arrangement is meant to let a customer’s specialised processor work inside an Nvidia-connected AI factory instead of operating as an isolated chip. That could give manufacturers more room to tailor hardware for a particular workload while keeping a common high-speed interconnect.

Reuters reported that the investment is part of MediaTek’s record $3.9 billion overseas convertible bond offering. A convertible bond can later be exchanged for shares under its terms, so the deal is not the same as Nvidia simply buying MediaTek outright. The financing gives MediaTek capital and gives Nvidia a deeper position in a supplier with experience across phones, networking and automotive electronics.

The companies also pointed to their existing work on AI PCs. MediaTek has been involved in Nvidia’s RTX Spark PC chip effort, which aims to bring local AI processing to smaller computers. The automotive side of the partnership adds another route for chips that can handle perception, software and vehicle controls at the edge.

The announcement does not promise a finished product or a delivery date for a new MediaTek chip. It sets out a platform relationship and an investment. Customers will still need to evaluate performance, software support, power use and supply plans before a design reaches production.

MediaTek’s role is becoming more visible as chip companies look for ways to build AI systems that are not limited to one processor type. The Nvidia deal gives that effort a specific link: custom silicon on one side and Nvidia’s rack-scale network on the other. The next test will be whether the partnership produces hardware that customers can buy and deploy.