Meta’s planned September mass production of the AI chip, codenamed Iris, will increase overall computing power to 14 gigawatts by next year. According to Reuters, the company is collaborating with Broadcom on the design, with TSMC handling manufacturing.

Iris is part of the fourth-generation Meta Training and Inference Accelerator (MTIA) chip project, which hopes to improve the computing performance supporting AI functions on its social media platforms, such as Facebook and Instagram. The chip underwent testing in just six weeks without any major issues, representing significant progress for this self-developed chip project, which began more than five years ago and whose progress had been slow to meet expectations.

According to a Meta memo, Iris is intended to complement Meta's current heavy reliance on graphics processing units (GPUs) from NVIDIA and AMD. However, the memo acknowledges that for a large company like Meta, implementing the latest GPUs "has always been a challenging and time-consuming task."

Meta estimates that AI hardware infrastructure spending among tech companies will reach as high as US$145 billion this year, a significant portion of the over US$700 billion in total expected AI investment. As global players race to expand their data centers to meet the ever-growing computing power demands of AI, demand for components, such as memory and AI chips, continues to surge as well.