Nvidia Invests $3.5 Billion in MediaTek to Power Custom AI Silicon
The strategic partnership allows hyperscalers to build custom AI accelerators using Nvidia's NVLink Fusion infrastructure.
Nvidia invested $3.5 billion in MediaTek via convertible bonds on August 31, 2026, to expand the reach of its AI infrastructure. The move signals a strategic shift as Nvidia seeks to embed its interconnect technology into the custom silicon being developed by the world's largest cloud providers.
Under the terms of the expanded partnership, MediaTek will adopt Nvidia's NVLink Fusion platform. This integration allows MediaTek to provide a foundation for developing custom AI accelerators, known as XPUs, for third-party customers. By utilizing NVLink Fusion, these custom chips can integrate seamlessly with Nvidia's rack-scale systems and AI factories, ensuring that third-party compute dies remain compatible with Nvidia's broader ecosystem.
The Rise of Custom Silicon
This partnership arrives as major hyperscalers—including Google, Amazon, Microsoft, and OpenAI—accelerate the development of their own proprietary AI silicon, such as TPUs, Trainium, Maia, and Jalapeño. These companies are increasingly designing their own chips to reduce their total reliance on Nvidia's proprietary GPUs and to optimize hardware for specific internal workloads. While MediaTek is widely recognized for its dominance in smartphone System-on-Chips (SoCs), the company possesses deep expertise in custom silicon design for the data center, making it a critical partner for Nvidia's new direction.
Becoming the AI Glue
The investment represents a pivot in Nvidia's market strategy. Rather than focusing solely on selling its own branded GPUs, Nvidia is positioning itself as the essential infrastructure provider for the entire AI chip industry. By allowing partners like MediaTek to build custom compute dies that still rely on Nvidia's NVLink and packaging standards, Nvidia ensures its technology remains the "glue" that holds AI data centers together. Even as customers move away from Nvidia-branded accelerators, they remain locked into Nvidia's interconnect and infrastructure ecosystem.
Future Outlook
Industry observers will now watch how quickly hyperscalers adopt this hybrid approach—combining custom compute logic with Nvidia's interconnects. The success of the venture depends on whether cloud providers are willing to trade full silicon independence for the efficiency and compatibility of the NVLink Fusion platform. While the financial terms of the convertible bonds are clear, the long-term impact on Nvidia's GPU market share remains to be seen.