Meta's MTIA v2 AI Chip Nears Mass Production

According to reports, Meta is close to completing the design of its second-generation in-house AI accelerator chip, the MTIA v2, drawing significant attention from the semiconductor and AI industry. The ASIC design is being led by Broadcom, a major global semiconductor company, with Taiwan-based Andes Technology playing a key role in the chip's architecture development. Mass production is expected to begin by the end of the year.

According to industry experts, the MTIA v2 is built on an open RISC-V architecture, incorporating a high volume of SRAM and mid-to-low-frequency cores to optimize AI processing performance. Andes Technology is reportedly contributing to the design of the processing elements (PEs), further strengthening its role in Meta's hardware ecosystem.

Designed to accelerate AI workloads powering Meta's social media platforms, the MTIA v2 chip will supplement the company's use of NVIDIA's GB series, creating new momentum in AI server infrastructure. Andes Technology has previously collaborated with Meta on multiple projects and has significantly enhanced CPU performance in the MTIA v2 development phase.

Technically, the MTIA v2 leverages TSMC's 5nm process node and integrates enhanced SRAM, increased memory bandwidth, and LPDDR5 support to boost inference capabilities. This architecture aims to deliver cost efficiency while supporting key applications such as machine learning inference and ad recommendation engines.