Meta reportedly plans to start production of its Iris AI chip in September

Meta plans to begin production of its in-house artificial-intelligence chip, code-named Iris, in September 2026, according to an internal company memo described this week.

Iris is one of four planned chip generations under Meta’s MTIA program, short for Meta Training and Inference Accelerator, and the memo reportedly says the chip cleared bug-testing in about six weeks without significant problems. Broadcom is the design partner and Taiwan Semiconductor Manufacturing Co., or TSMC, is set to fabricate it.

The chip is one piece of a much larger compute push. The memo reportedly outlines a two-step buildout: 7 gigawatts of capacity online in 2026, rising to 14 gigawatts in 2027. Meta’s 2026 spending on AI infrastructure is projected to run as high as $145 billion.

Moving a first-generation training-and-inference chip into production would give Meta an alternative to buying Nvidia accelerators for at least part of that expansion, trimming its dependence on a single supplier as costs climb. Other large cloud operators have followed the same path with custom silicon.

The account rests on a single internal memo and has not been confirmed by Meta, which has not published a chip roadmap or a production date. A September production start is a plan, not a shipped part, and semiconductor timelines slip often; the six-week testing figure and the gigawatt targets are the memo’s own claims and have not been independently verified.

Whether Iris reaches volume production on schedule will show how far Meta’s custom-silicon effort has closed the gap with the chips it still buys.