Nvidia invests $3.5 billion in MediaTek AI chip link

The announcement arrives as a financial deal with a hardware consequence: Nvidia is investing $3.5 billion in convertible bonds issued by Taiwan’s MediaTek. At the same time, MediaTek will adopt Nvidia’s NVLink Fusion platform, giving customers a route to develop custom AI chips that connect to Nvidia-powered systems and data centers.

The machinery behind that link is meant for more than a single chip. NVLink Fusion includes a chiplet that connects custom XPUs — processors designed for particular AI workloads — to Nvidia’s scale-up fabric, along with NVLink-C2C for connecting XPUs to Nvidia Vera CPUs and other processors. Nvidia also describes NVHBM, a memory approach intended to increase bandwidth and power efficiency. Advanced packaging, high-speed connections, memory and rack-scale networking still require engineering, qualification and supply-chain support.

That package changes the choice available to large infrastructure builders. Hyperscalers, cloud service providers and AI model developers can bring custom compute designs to MediaTek and configure connectivity, memory, packaging, performance and power for their workloads, while Nvidia and MediaTek provide the surrounding technologies and support. The companies say customers can also use Nvidia’s scale-up and scale-out technologies and MGX rack-scale architecture.

The partnership reaches closer to the user as well. Nvidia and MediaTek will continue developing multiple generations of RTX Spark and DGX Spark PC chips for consumer PCs, AI development systems and enterprise workstations. Their earlier work includes the GB10 Grace Blackwell Superchip used in DGX Spark, which combines an Nvidia Blackwell GPU and Grace CPU through NVLink-C2C for local AI computing.

And then there are vehicles. MediaTek Dimensity Auto platforms integrate Nvidia technologies for AI and RTX graphics in vehicle cockpits and can operate alongside Nvidia DRIVE AGX. The companies plan to keep developing platforms for AI-enabled, software-defined vehicles and physical AI applications.

So what changes, concretely? A company designing a specialized AI accelerator could have a clearer path to connecting it with Nvidia rack-scale infrastructure instead of building every layer alone. For PC makers and vehicle developers, the same partnership extends local AI and cockpit work into future platform generations. But the reports describe collaboration and development, not a customer deployment or production launch. Investors are also watching the financing structure: the Korea Herald says Nvidia’s support for businesses built around its technology could add to growing investor concerns about circular financing in AI.