Meta plans September production start for its in-house Iris AI chip to expand compute capacity.

Meta is reportedly preparing to begin manufacturing an artificial intelligence chip called Iris in September, as part of a broader initiative to expand its computing infrastructure and reduce dependence on external suppliers. The custom data center chip is part of Meta's multi-generation Meta Training and Inference Accelerator (MTIA) program and is being designed in-house to support AI workloads across Facebook, Instagram, and other Meta platforms.

According to an internal memo reported by Reuters, Meta is targeting 14 gigawatts of computing power by next year. The company plans to deploy 7 gigawatts of computing infrastructure this year and double that capacity in 2027.

Iris is designed to supplement the large quantities of graphics processing units Meta purchases from Nvidia and AMD for AI applications. While not intended to replace GPUs entirely, the chip could help Meta reduce infrastructure costs and gain greater control over its AI hardware roadmap. Reuters reported that Iris completed testing in six weeks without major issues—a significant milestone for Meta's internal chip program, which has encountered challenges since its launch over five years ago.

Meta is working with Broadcom on chip design and Taiwan Semiconductor Manufacturing Company (TSMC) for production. This custom silicon strategy reflects a broader trend among large technology companies seeking independence from external AI chip suppliers.

Meta expects to spend as much as $145 billion on AI infrastructure this year as it expands model training, inference, coding tools, and AI-powered product experiences. An internal memo noted that adopting the latest GPUs at Meta's scale has proven costly and time-consuming; custom chips tailored to Meta's workloads could optimize performance, manage costs, and enable more efficient scaling of AI services.

Meta publicly unveiled Iris in March alongside three other AI processors. The company plans to launch a new chip approximately every six months through 2027—a significantly faster cadence than the annual or longer release cycles typical in the AI chip market.

To support its infrastructure expansion, Meta has secured long-term supply agreements with Samsung Electronics for memory chips, SanDisk for flash storage, and Sumitomo Electric for fiber-optic equipment. These agreements reflect surging demand across AI infrastructure components, including memory, storage, networking equipment, and accelerators, as major technology companies race to build sufficient data center capacity for AI training and inference. Meta's approach to custom silicon underscores the increasingly strategic role of proprietary hardware in the AI competition, as rising compute demands drive companies to control more of the hardware stack underlying their AI products.