NVIDIA has committed $3.5 billion in convertible bonds to MediaTek, structuring a decade-long financing arrangement aimed at co-developing AI data center networking and edge computing solutions.
NVIDIA has agreed to purchase $3.5 billion in convertible bonds issued by MediaTek, formally expanding their alliance from client hardware into artificial intelligence data centers. The transaction establishes a 10-year product roadmap to integrate custom MediaTek processors directly into NVIDIA infrastructure via NVLink Fusion. While the partnership aims to accelerate custom chip deployments, it also tightly aligns both semiconductor leaders in the competition to build global AI manufacturing facilities.
Speaking to Bloomberg, NVIDIA CEO Jensen Huang identified MediaTek as one of his largest industry partners. At the core of this expanded alliance is NVLink Fusion, a framework that enables MediaTek’s custom processors to connect directly to NVIDIA data center hardware. This integration allows customers to deploy custom silicon alongside NVIDIA Spectrum X Ethernet, switch systems, and Vera central processors without constructing an entire server platform from the ground up. Consequently, whenever MediaTek secures a custom silicon contract, NVIDIA automatically supplies the accompanying network hardware and host processors.
The collaboration extends to consumer hardware as well. Previous efforts linked MediaTek mobile silicon with NVIDIA graphics chips via high-speed chip-to-chip interconnects, resulting in the DGX Spark desktop supercomputer. Building on this foundation, both companies intend to develop multiple generations of processors for Windows PCs, engineered to drive the next wave of AI agent software across desktop and laptop platforms.
Huang dismissed claims that the $3.5 billion convertible bond purchase constitutes circular financing, emphasizing that MediaTek is already a profitable company with robust cash flow. He clarified that the capital will fund long-term research initiatives rather than artificially inflate sales figures.
Huang highlighted his 25-year working relationship with MediaTek CEO Rick Tsai, citing personal trust as a cornerstone of the agreement. MediaTek intends to allocate the funds toward advanced chip packaging, high-speed interconnect intellectual property, and strengthening global supply chain resilience. Tsai added that MediaTek’s operations in Taiwan provide direct access to the manufacturing ecosystem essential for scaling AI hardware production.
Huang also outlined financial metrics illustrating how NVIDIA captures value in modern data centers. Every 1GW of power capacity drives increasing equipment revenue across product cycles. The Hopper architecture generated approximately $18 billion per gigawatt, while Grace Blackwell raised that figure to $25 billion. The forthcoming Vera Rubin generation is projected to exceed $40 billion per gigawatt. This revenue expansion stems from the fact that a modern AI data center requires seven distinct types of silicon, spanning switches to network interface cards. Consequently, even when a cloud provider opts for a custom MediaTek accelerator, NVIDIA retains its role as the primary supplier for the remainder of the server ecosystem.
To support this architecture, NVIDIA is extending its NVHBM memory platform to MediaTek. The technology pairs a custom NVIDIA base die with High Bandwidth Memory (HBM) utilizing advanced CoWoS packaging. Opening NVHBM access to strategic partners such as MediaTek and Amazon aims to standardize custom chip assembly processes while alleviating strain on global component suppliers.