Skip to main content

Meta's Custom AI Chip Iris to Start Mass Production in September

Meta is making a bold move in the AI hardware race. The company has partnered with Broadcom and TSMC to develop a custom AI data center processor, codenamed "Iris," which is slated to enter mass production in September 2026. This chip is part of Meta's broader MTIA (Meta Training and Inference Accelerator) project and has already passed initial vulnerability testing within six weeks.

Iris is designed to handle core tasks like content ranking, recommendation systems, and generative algorithms for Facebook and Instagram. By tailoring the chip to its specific needs, Meta hopes to reduce data center costs and lessen its dependence on third-party GPUs from companies like NVIDIA and AMD.

Image

But Meta isn't stopping there. To secure its supply chain amid global hardware shortages, the company has signed long-term contracts with Samsung Electronics, SanDisk, and Sumitomo Electric. Meta plans to deploy about 7 gigawatts of computing capacity by the end of 2026 and double that to 14 gigawatts by 2027. To fund this expansion, Meta's infrastructure capital expenditure for 2026 is expected to reach $14.5 billion.

While Iris won't immediately replace existing GPUs, it marks a significant step toward greater hardware autonomy. Meta's aggressive six-month development cycle will allow it to iterate quickly and reduce reliance on external suppliers. On an industry level, this move helps Meta build an independent AI ecosystem, free from the constraints of third-party software and hardware.

Key Points

  • Meta's custom AI chip, Iris, will enter mass production in September 2026.
  • The chip is designed to optimize ranking, recommendations, and generative AI on Facebook and Instagram.
  • Meta has secured long-term supply contracts with Samsung, SanDisk, and Sumitomo Electric.
  • The company plans to double its computing capacity to 14 gigawatts by 2027.
  • Iris represents a step toward reducing reliance on NVIDIA and AMD GPUs.