
According to multiple media reports, Meta plans to deploy its next-generation in-house AI chip, codenamed Arke (MTIA 450), in its data centers in the first half of 2027. The company is also expected to launch its subsequent-generation chip, codenamed Astrid (MTIA 500), by the end of 2027. The two chips represent the third generation and a follow-up generation of Meta’s in-house AI accelerator program, which was launched in 2023, and are primarily designed for AI inference workloads.
Yee Jiun Song, Meta’s vice president of engineering, said the new chips are expected to deliver better performance per watt and performance per dollar when running AI models than NVIDIA processors. Meta is reportedly aiming to complete the design of Astrid within about a month and expects the chip to be deployed at a larger scale than any previous generation of its in-house AI chips.
The development of Arke and Astrid is part of Meta’s broader effort to reduce the operating costs and energy consumption of its AI infrastructure while reducing its reliance on NVIDIA accelerators. By developing its own AI chips, Meta is seeking greater control over computing performance, power efficiency, and infrastructure costs as AI workloads continue to expand.