NVIDIA announces an investment of $3.5 billion in Taiwan-based chip manufacturer MediaTek, and will also share with them AI technologies related to custom chip design. As large cloud providers and model companies accelerate their own chip development, NVIDIA is attempting to shift the focus of competition from a single GPU to...Data Center Interconnection StandardsWith the entire infrastructure ecosystem.
Betting on the expansion of custom chips
Amazon, Google, Microsoft, as well as OpenAI, Anthropic and other companies have been continuously advancing their own AI chip development in recent years to reduce their dependence on NVIDIA GPU. This investment indicates that NVIDIA has not avoided this trend but has chosen to become more deeply integrated with the custom chip supply chain.
According to the arrangements of both parties, MediaTek will be able to develop custom chips for AI and large cloud service providers in the future, and these chips will be directly integrated into data centers based on the NVIDIA architecture. For NVIDIA, even if customers do not use its own chips exclusively, they may still continue to adopt its underlying deployment system.
NVLink Fusion becomes the core
One key aspect of the collaboration is MediaTek's integration into NVIDIA's NVLink Fusion ecosystem. NVLink is NVIDIA's data interconnection technology, which is used to enhance high-speed communication between different chips. NVIDIA states that this system also supports non-NVIDIA chips to be integrated into the same data center architecture.
This means that the custom AI chips provided by MediaTek in the future can be deployed in parallel with NVIDIA platforms. For cloud providers and model developers, this will enable them to run GPU and ASIC simultaneously within the same rack-level infrastructure, reducing deployment costs associated with system fragmentation.
- NVIDIA's investment this time amounts to 3.5 billion US dollars.
- MediaTek will adopt NVIDIA-related design technologies
- One of the key areas of cooperation is to integrate with NVLink Fusion.
Cooperation has extended to more scenarios.
Just last week, NVIDIA also announced a similar partnership with Amazon Web Services. Although there was no direct investment, AWS will deploy an additional 2 million NVIDIA GPU in its infrastructure and integrate NVLink Fusion. These two consecutive partnerships indicate that NVIDIA is promoting a more open data center model, which is still led by them.
MediaTek is also expanding its data center ASIC business. The company stated in June this year that it expects this business to generate $2 billion in revenue by 2026. In addition to data centers, the two parties will continue to advance cooperation in DGX Spark, RTX Spark, as well as platform collaboration for AI software-defined vehicles, extending the scope of cooperation from servers to PC and automobiles.
