Meta Begins Production of New AI Chips in September

Meta is scheduled to commence production of its proprietary AI chips in September, according to an internal memo reported by Reuters. This initiative is part of Meta's broader strategy to decrease its reliance on external graphics processing unit (GPU) suppliers, such as Nvidia and AMD, and to manage costs amidst component shortages.

The company is collaborating with Broadcom on chip design, with manufacturing handled by Taiwan's TSMC. Meta is also sourcing components like RAM from Samsung and storage from Sandisk for its new chips, developed under the Meta Training and Inference Accelerator (MTIA) program.

In March, Meta detailed its four new MTIA chips, some of which are currently being deployed or will be later this year. The company has adopted a modular chiplet-based design approach, anticipating the rapidly evolving needs in artificial intelligence and aiming for shorter development cadences for future iterations.

While these custom-designed chips are expected to help Meta save on GPU expenditures, the company anticipates continuing to purchase substantial quantities from established vendors like Nvidia and AMD. The MTIA chips will primarily be used for training models for Meta's ranking and recommendation algorithms, as well as for broader AI workloads and inference tasks.