On August 31, NVIDIA disclosed that it had put 3.5 billion USD (about 560 billion yen) into convertible bonds issued by MediaTek. The two companies have worked together on PC silicon for some time, but this adds a capital relationship on top of that. The scope covers three areas, AI infrastructure, local AI computing and automotive, under a framework for jointly building the next generations of AI computing platforms.

※1 USD = 160 JPY (based on the August 31, 2026 mid rate)

A Convertible Bond, Not an Acquisition

The structure of the money matters here. What NVIDIA took on is a convertible bond issued by MediaTek, not a direct equity stake that folds the company into its group. MediaTek stays an independent chipmaker, and the relationship simply moves up a level as a co-development partner.

NVIDIA has repeatedly placed capital around compute resources and supply chains in recent years. This deal sits in that pattern, but the intent looks somewhat different. With hyperscalers leaning toward building their own silicon in house, this investment reads as securing a place where those custom chips can plug into NVIDIA rack-scale infrastructure.

Rick Tsai, vice chairman and CEO of MediaTek, said the investment strengthens a collaboration that spans cloud AI infrastructure, local AI computing and automotive. Treating everything from edge to cloud as a continuum appears to be the shared premise.

Taking On What Comes After the Custom Chip

Of the three areas, AI infrastructure carries the largest structural weight. MediaTek will adopt the NVIDIA NVLink Fusion platform and offer it as the design foundation for customers developing custom XPUs.

The point is that designing a chip and running it inside a rack-scale AI factory are two different problems. Integrating multi-die architectures, advanced packaging, high-speed SerDes, HBM, I/O and scale-up networking into something manufacturable requires chip-to-rack qualification and supply chain support.

What MediaTek sells is that heavy layer, prepared in a prevalidated state. Customers can concentrate their resources on the differentiated compute that defines their platforms, and lean on NVIDIA and MediaTek for connectivity, memory architecture, packaging, manufacturing and the rack-scale technologies needed for production. Customers can bring their own XPU designs and tailor connectivity and power characteristics to their workloads.

The NVLink Fusion ecosystem already includes multiple partners. MediaTek now joins it with custom silicon implementation strength and a capital tie attached.

The Spark Line Continues Into Its Next Generations

The second area is AI computing that runs locally. The two companies previously collaborated on the GB10 Grace Blackwell Superchip, the part that powers NVIDIA DGX Spark. It links a Blackwell GPU and a Grace CPU over NVLink-C2C to bring AI processing to edge systems.

This announcement states that the collaboration on PC chips for NVIDIA RTX Spark and DGX Spark will continue across multiple generations. The targets are consumer PCs, AI developer supercomputers and enterprise-class workstations. Pairing NVIDIA GPUs with MediaTek SoCs is now positioned as a continuing product line rather than a one-off.

Power-efficient SoC design has long been MediaTek territory, so how much practical AI work these machines can sustain on battery is the interesting question for this pairing.

Vehicles Move Toward Being Defined by AI

The third area is automotive. The companies say they will develop platforms for AI-powered, software-defined vehicles across multiple generations.

The concrete starting point is MediaTek's Dimensity Auto platform. NVIDIA technologies are integrated into it to deliver advanced AI features and NVIDIA RTX graphics for intelligent vehicle cockpits, and the configuration is designed to work alongside NVIDIA DRIVE AGX on the driving side. Cockpit and driving functions can sit on separate silicon while the overall architecture stays scalable.

Jensen Huang, founder and CEO of NVIDIA, said AI is transforming every computing platform, from the world's largest AI factories to the PC and the car. The three-area lineup in this announcement reads like that sentence translated directly into an org chart.

Summary

NVIDIA's 3.5 billion USD convertible bond purchase looks less like plain funding and more like securing a home for its own architecture in an era of custom silicon. Customers building their own chips cannot be stopped, so the move is to supply the floor and walls those chips slot into. For MediaTek, a foothold in the data center market just solidified in one step. Concrete products from the PC and automotive collaboration are still some distance out, but the explicit commitment to continuing the Spark line is worth noting for anyone considering an AI PC.