Meta Platforms plans to begin producing its next in-house artificial intelligence chip, code-named “Iris,” in September, according to an internal company memo reviewed by Reuters. The move is part of a broader push to cut the company’s reliance on Nvidia GPUs and build out its own AI silicon supply chain.
Iris is the latest processor in Meta’s four-generation Meta Training and Inference Accelerator (MTIA) program, designed with help from Broadcom and manufactured by Taiwan Semiconductor Manufacturing Co. (TSMC). According to the memo, the chip completed roughly six weeks of testing without turning up significant problems, clearing the way for production to begin.
Scaling compute
The chip is meant to supplement – not replace – the large volumes of Nvidia and AMD GPUs Meta buys to train and run the AI systems behind Facebook, Instagram and its other products. The memo outlines plans to add about 1 gigawatt of computing capacity in the first half of 2026 and roughly 5.5 gigawatts more in the second half, bringing total capacity to around 7 gigawatts by year’s end. Meta is targeting 14 gigawatts by 2027, double this year’s total, as part of an AI infrastructure budget Reuters reports could reach $145 billion in 2026.
Meta also plans to release a new MTIA chip roughly every six months through 2027, a faster cadence than the roughly annual release cycles common among chipmakers.
Racing to cut reliance on Nvidia
Meta joins Google, Amazon and Microsoft in building custom AI accelerators to lower costs and reduce dependence on Nvidia, whose GPUs remain in short supply amid surging demand. “You can’t become an AI titan if you are dependent on another company for chips,” Forrester Research analyst Mike Gualtieri told Reuters.
To support the buildout, Meta has also lined up supply deals with Samsung Electronics for memory chips, SanDisk for flash storage and Sumitomo Electric for fiber-optic networking equipment, the memo shows.