Meta will begin manufacturing its in-house AI chip, codenamed Iris, in September 2026, according to an internal company memo reviewed by Reuters. The chip represents Meta’s first serious attempt to reduce dependence on external suppliers for the silicon powering its AI workloads.
The Iris chip was developed in-house and completed in under 18 months, marking an accelerated timeline for a company that has historically relied on Nvidia and AMD for AI compute. Internal documents show Meta plans to deploy seven gigawatts of computing infrastructure in 2026 to support its expanding AI ambitions, which include running large-scale training and inference across its social platforms and metaverse projects.
Iris is expected to handle both training and inference workloads, though initial production runs will focus on inference — the less technically demanding task of running already-trained models. The broader compute push aligns with Meta CEO Mark Zuckerberg’s stated vision of building artificial general intelligence capabilities internally.
The move places Meta among a growing group of tech giants — including Google, Amazon, and Microsoft — that have developed in-house chips to optimize AI performance and cost. Production is slated to begin through an external manufacturing partner, though Meta has not disclosed which foundry will handle the Iris chip’s fabrication.
