Original Source
Meta Plans to Deploy Four In-House AI Chips by 2027
Meta Expands In-House AI Chip Development
Meta Platforms is set to sequentially deploy four types of internally developed artificial intelligence chips (AI chips) by 2027. This move is part of a strategy to effectively handle surging AI workloads and reduce reliance on external chip suppliers. Currently, Meta heavily purchases chips from Nvidia and AMD, but it aims to leverage the benefits of custom silicon to enhance cost-efficiency and optimize performance. This can be interpreted as a rebuttal to some recent reports suggesting that Meta had abandoned its own chip development efforts.
The Role of MTIA Chips and Future Plans
Meta is already in the production phase for its own chip called MTIA (Meta Training and Inference Accelerator) 300, which is used for training AI-driven ranking and recommendation systems, such as ad recommendations on Instagram feeds. While Meta primarily uses Nvidia and AMD hardware for model training, it believes that utilizing its own chips for the model execution (inference) phase is more economically advantageous. The company plans to introduce a total of four generations of MTIA chips by 2027, joining the trend of other hyperscalers like Google and Amazon that are already developing their own chips.
*Source: YouTube: Bloomberg (2026-03-11)*
