Nvidia Corp. has bought $3.5 billion worth of MediaTek Inc. convertible bonds as part of an expanded partnership that the companies announced today.

A convertible bond is a debt instrument that the buyer can turn into shares later on. Nvidia regularly buys stakes in partners as part of technical collaborations. Earlier this year, it purchased $2 billion worth of shares in Marvell Technologies Inc., a MediaTek competitor. 

Taiwan-based MediaTek is best known as a supplier of smartphone and connected device processors. It also has a custom chip design business that specializes in artificial intelligence accelerators. According to Nvidia, that unit is the first focus of the newly expanded partnership.

Companies that hire MediaTek to design an AI accelerator can choose from a catalog of on-chip networking technologies. Some can connect the processor to other chips in the same rack, while others facilitate data movement across racks. Going forward, MediaTek will offer Nvidia’s NVLink Fusion networking technology as an interconnect option.

AI racks contain not only graphics processing units but also central processing units that perform supporting tasks such as organizing the data processed by AI models. NVLink Fusion enables Nvidia’s GPUs and CPUs to exchange data with third-party CPUs and CPUs, respectively. One of technology’s main selling points is its speed. NVLink, the interconnect that underpins NVLink Fusion, can move data more than 10 times faster than a standard implementation of PCIe 6.0.

Last week, Nvidia upgraded the technology by adding support for a new memory design approach called NVHBM. The onboard HBM memory in which graphics cards store data is managed by a tiny chip called a memory controller. NVHBM moves the controller from the GPU’s logic die to the die that contains the HBM memory. According to Nvidia, that change boosts data movement speeds while lowering power consumption.

The second focus of the company’s expanded MediaTek partnership is the personal computing market. In 2025, Nvidia launched a workstation called the DGX Spark that developers can use for tasks such as fine-tuning AI models. It’s powered by a system-on-chip that MediaTek helped design. The two companies subsequently collaborated on the RTX Spark, a laptop and desktop processor that debuted in June.

Nvidia stated today that it plans to collaborate with MediaTek on multiple future generations of the RTX Spark and DGX Spark. The latter company’s exact role was not detailed. However, the current-generation RTX Spark and DGX Spark chips might offer clues. MediaTek helped develop the chips’ onboard CPUs along with certain other components such as the RTX Spark’s memory controller.

The companies’ expanded partnership also prioritizes the auto sector. MediaTek sells a series of systems-on-chip called Dimensity Auto that carmakers use to power in-vehicle displays. The processors include, among others, an Nvidia graphics rendering chiplet. The companies stated today they plan to continue working together on software-defined vehicle chips.

“Together, we’re building platforms that bring Nvidia accelerated computing to new markets and give customers the freedom to create differentiated AI systems at enormous scale,” said Nvidia Chief Executive Officer Jensen Huang.

MediaTek, for its part, is strengthening a partnership that supports one of its fastest-growing businesses. The company recently doubled its full-year revenue forecast for its data center chip design group to $2 billion. It hopes to grow that number severalfold in the coming years.

Photo: Nvidia