Meta just expanded its partnership with Broadcom to co-develop multiple generations of its custom MTIA AI chips. Built on Broadcom's XPU platform and high-bandwidth Ethernet networking, the rollout kicks off with over 1 gigawatt of capacity before expanding to multi-gigawatt scale. Broadcom CEO Hock Tan will also step down from Meta's board to serve as a custom silicon advisor.
Meta and Arm just unveiled a partnership to co-develop multiple generations of custom data center CPUs, starting with the new Arm AGI CPU. The chip is engineered to pack more compute into tight spaces while running alongside Meta's custom MTIA silicon. Meta will also release its board and rack designs through the Open Compute Project later this year.
Meta just detailed an aggressive roadmap for its custom MTIA silicon, planning to launch four new chip generations over the next two years. MTIA 300 is already in production for recommendation training, while the 400, 450, and 500 series will focus heavily on GenAI inference through 2027. By taking an inference-first approach and a modular design that drops into existing racks, Meta aims to ship new chips roughly every six months.
Meta is pouring hundreds of billions into massive multi-gigawatt data centers to power its new Superintelligence Labs. The first cluster, Prometheus, arrives next year, followed by Hyperion—a 5-gigawatt behemoth that triples OpenAI's planned footprint. Zuck's goal is straightforward: give his newly poached researchers the most compute per head on the planet.
Meta just kicked off small-scale testing of its first custom AI training silicon, aiming to rein in its gargantuan infrastructure spending and ease its reliance on Nvidia. The MTIA chip is designed to be more power-efficient than general-purpose GPUs, and Meta hopes to deploy it for production training by 2026.
Meta just unveiled the next generation of its custom AI inference accelerator, MTIA. Built on TSMC's 5nm process, the new silicon more than doubles compute and memory bandwidth, delivering 3x the performance of MTIA v1 across key models. Why it matters: Meta is scaling its in-house chip lineup to slash inference costs and ease reliance on commercial GPUs.
Meta just introduced its first custom silicon for artificial intelligence, the Meta Training and Inference Accelerator (MTIA). Designed specifically for deep learning recommendation models and integrated with PyTorch, the chip is up to three times more power-efficient than GPUs per watt. While it trails Nvidia GPUs on more complex neural networks, it signals Meta's serious push to design its own compute infrastructure.
