Nvidia will invest $3.5 billion (about 4.8 trillion won) in MediaTek, a Taiwanese fabless (chip design specialist) company. They plan to consolidation the custom artificial intelligence (AI) chips MediaTek designs with Nvidia's data center systems and expand joint development in PCs and automobiles.
Nvidia and MediaTek said on the 31st (local time) that they will expand cooperation in three areas: AI infrastructure, local AI computing, and automobiles. Nvidia purchased $3.5 billion worth of bonds issued by MediaTek that can be converted into shares.
The core of this partnership is Nvidia's "NVLink Fusion." MediaTek will use NVLink Fusion to design custom AI accelerators (XPUs) needed by big tech, cloud service providers, and AI model developers, and support consolidation of them to Nvidia's rack-scale AI data center systems.
NVLink is Nvidia's interconnect technology that links multiple AI Semiconductors at high speed so they operate like a single system. NVLink Fusion is an open platform that lets customers use their own AI chips while also leveraging Nvidia's network and rack-scale infrastructure.
When customers develop their own AI accelerators, MediaTek will provide NVLink consolidation technology, memory architecture, and advanced packaging together. Nvidia's "NVHBM" technology will also be applied. NVHBM is a technology that combines custom memory functions with AI chips to increase memory bandwidth and power efficiency.
As big tech and AI companies increase in-house AI chip development to reduce reliance on Nvidia graphics processing units (GPUs), Nvidia appears to be pursuing a strategy to pull even these custom chips into its data center ecosystem. Bloomberg assessed that the investment comes as Nvidia is encouraging more corporations to develop semiconductors that are consolidation to its data center ecosystem.
MediaTek is a Taiwanese fabless company focused on smartphone application processors (AP) and is recently expanding its custom chips for data centers business. Through this partnership, it plans to design the connectivity of customer-designed AI accelerators, as well as memory, packaging, performance, and power characteristics, to meet requirements.
The two companies will also continue cooperating on PC AI Semiconductors. MediaTek took part in developing the "GB10 Grace Blackwell Superchip" installed in Nvidia's "DGX Spark." Nvidia and MediaTek will jointly develop PC chips for "RTX Spark" and DGX Spark over multiple generations. The product combines Nvidia GPUs with MediaTek's system-on-chip (SoC).
In automobiles, MediaTek's in-vehicle SoC design technology will be combined with Nvidia's AI and graphics technologies. MediaTek's "Dimensity Auto" platform applies Nvidia AI and RTX graphics technologies and can interoperate with Nvidia's in-vehicle computing platform "DRIVE AGX." The two companies plan to continue developing software-defined vehicle (SDV) and physical AI automotive platforms.
Huang Jen-hsun, Nvidia chief executive officer (CEO), said, "AI is transforming every computing platform, from the world's largest AI factories to PCs and automobiles," and added, "Together with MediaTek, we are expanding Nvidia accelerated computing into new markets and creating a platform that enables customers to freely build differentiated AI systems at scale."