Meta will start production of its custom AI chip, code-named Iris, in September 2026, according to an internal memo. The chip is part of the company's Meta Training and Inference Accelerator (MTIA) program, one of four planned chip generations designed to power AI systems across Facebook and Instagram.
At least one version of the chip cleared its bug-testing phase in roughly six weeks without turning up significant problems. Broadcom is serving as Meta's design partner, while Taiwan Semiconductor Manufacturing Co. (TSMC) will handle fabrication; Samsung is supplying memory, Sandisk storage, and Sumitomo Electric fiber-optic equipment. Meta describes its approach as a modular chiplet design, with each MTIA generation building on the last to incorporate the latest AI workload insights.
By building its own silicon, Meta aims to cut compute costs and reduce reliance on third-party vendors like Nvidia and AMD, even as it continues buying their GPUs. The company plans to deploy 7 gigawatts of compute capacity in 2026 and double that to 14 gigawatts by 2027, with 2026 capital expenditures projected at $125-145 billion, much of it earmarked for AI infrastructure.