According to Reuters and internal memos from Meta, Meta is collaborating with Broadcom and TSMC to develop a self-designed AI data center processor codenamed "Iris," which is planned to enter mass production in September 2026. The chip is part of Meta's custom-built infrastructure (MTIA) project and has successfully passed initial vulnerability testing within six weeks. Its core function is to deeply optimize the core content ranking, recommendation systems, and generative algorithms for platforms such as Facebook and Instagram, thereby reducing data center costs and reducing reliance on third-party GPUs.

QQ20260710-090955.png

Amid global hardware shortages and "chip inflation," Meta has signed long-term supply contracts with industry giants such as Samsung Electronics, SanDisk, and Sumitomo Electric to ensure strategic security of its supply chain. Looking at the larger picture of computing power deployment, Meta plans to deploy approximately 7 gigawatts of computing capacity by the end of 2026 and expects to double it to 14 gigawatts by 2027; for this purpose, Meta's infrastructure capital expenditure in 2026 will reach as high as 14.5 billion dollars.

The production of this self-developed chip marks a key turning point for Meta in the trillion-dollar computing power arms race among tech giants. Although "Iris" currently serves as a complement rather than a full replacement for existing NVIDIA and AMD third-party GPUs, this aggressive self-development cycle that occurs every six months will grant Meta greater hardware deployment autonomy. On an industry level, this not only effectively mitigates supply chain bottlenecks but also helps Meta deepen the construction of an independent AI ecosystem that does not rely on external software and hardware constraints.