Meta plans to deploy its proprietary MTIA 450 AI chip in data centers during the first half of 2027. According to engineering executive Lee-Jeong Song, the chip, internally codenamed Arke, is currently undergoing testing. The primary objective of this initiative is to lower energy consumption and operational costs for AI model workloads.
The company previously outlined a roadmap in March for four generations of MTIA silicon. Both the MTIA 450 and the upcoming MTIA 500 are designed specifically for generative AI inference tasks. Despite this shift toward in-house hardware, Meta intends to maintain its existing portfolio of external accelerator suppliers.
The MTIA 500 is scheduled for release later in 2027, following the initial rollout of the MTIA 450. While the roadmap is established, the actual operational value remains to be verified. Independent data regarding performance benchmarks, energy efficiency, latency, or the maturity of the software stack for these new chips has not yet been disclosed.
