Google to work with MediaTek to build AI server chips, the latter “not responded”

According to Taiwanese media “Economic Daily”, there are rumors that Google will cooperate with MediaTek to jointly produce AI server-specific chips, and set the goal to start mass production early next year. In this regard, MediaTek said it “does not respond.”

▲ Picture source MediaTek

The details of this rumor are as follows:

 The AI server chip completed by Google and MediaTek will be provided by MediaTek with serializer and deserializer (SerDes) solutions

 MediaTek assisted in the integration of Google's self-developed tensor processor (Tensor Processing Unit, referred to as TPU, specially used in the field of machine learning)

 Produced using TSMC's 5nm process

 It is expected to be delivered to the fab for manufacturing by the end of this year, with the goal of mass production at the beginning of next year

It is noticed that Chen Guanzhou, general manager of MediaTek, had previously stated publicly that how to meet the development trend of large language models, regardless of terminal or cloud AI, “is a problem that MediaTek is thinking about.” Chen Guanzhou also said that MediaTek has invested a lot in the field of AI computing. “A few years ago, I realized that future computing will not only rely on CPUs and GPUs, but AI-based APUs will also rise.”

In addition, MediaTek will launch the Dimensity 9300, the third-generation chip of the Dimensity series at the end of October this year, which will also have AI computing technology based on a large language model.

Google also laid out research on deep learning technology many years ago and released the first generation of deep learning dedicated processor chip TPU in 2015 to solve the problem of high computing costs derived from GPUs. At present, Google’s self-developed TPU chip has developed to the fourth generation.