Meta plans to start mass production of custom AI chips in September

Meta Platforms is gearing up to start mass production of its custom-designed artificial intelligence chips, codenamed Iris, as early as September. This move is a critical step in the company's broader plan to significantly scale its AI computing infrastructure.

The Iris chip is part of Meta's AI chip development roadmap, aiming to increase its overall AI computing capacity to 14 gigawatts (GW) by 2027. By developing its own silicon, Meta seeks to enhance the computational power for AI features across its platforms, including Facebook and Instagram, while also reducing reliance on external chip suppliers.

An internal Meta memo indicates that testing for the Iris chip, designed in collaboration with Broadcom and manufactured by TSMC, was completed quickly and without major issues. This accelerated timeline is a positive development for the self-developed chip initiative, which has been underway for over five years. The chip is intended to complement, not replace, the company's large purchases of AI GPUs from companies like Nvidia and AMD.

This strategic push into custom silicon is expected to help Meta manage the substantial costs associated with its growing AI compute needs. The company announced Iris and other AI chips earlier this year and plans a rapid development cycle, releasing new AI chips approximately every six months through 2027. This pace is considerably faster than the typical industry update cycle.

Meta is projecting significant investments in its AI infrastructure, potentially reaching $145 billion this year. To support this expansion, the company has secured long-term supply agreements for essential components, including memory chips from Samsung Electronics and flash products from Sandisk, addressing potential supply chain constraints in the burgeoning AI hardware market.