Meta's first public self-research AI chip, estimated to be officially launched in 2025
Meta, the parent company of Facebook, has joined the semiconductor war between Google, Amazon, and Microsoft, and is making its first public progress in developing its own AI chips.
Meta is building an infrastructure architecture specifically designed for AI, covering all aspects of the hardware and software stack, as well as a customized network to connect these technologies, including Meta's first custom chip for executing AI models, a data center design optimized for AI, and an AI supercomputer currently in its second stage of development.
The new ASIC chip called Meta Trainingand Inference Accelerator (MTIA) is Meta's first self-developed custom chip that claims to outperform CPUs in terms of computing performance and processing efficiency, and is specifically designed to handle tasks related to AI recommendation systems, helping users find the best content to post and present it to you faster. Meta developed the first generation of MTIA (MTIA v1) in 2020, using TSMC's 7nm process. According to foreign media TechCrunch, MTIA will not be officially launched and put into service until at least 2025.
In addition to MTIA, Meta is developing another chip, an ASIC chip called Meta Scalable Video Processor (MSVP), to support the growing demand for audio and video content processing, and eventually hopes to hand over most of the mature and stable audio and video content processing work to MSVP for execution.
As for Meta's next generation data center design, it will not only support existing products, but also assist future new AI hardware to start training and reasoning. The new data center is optimized for AI, supporting liquid-cooled AI hardware devices and efficient AI networks, and tying thousands of AI chips together to form a data center-scale AI training set that can complement new hardware devices such as MSVP.
Meta used to mainly use CPUs and custom chips designed to accelerate AI algorithms to handle AI computing work. In order to turn the situation around, Meta started to develop its own custom chips and compete with Google, Amazon, Microsoft and others who also invest a lot of resources into the AI field.