Overview
- Reports on July 10 revealed an internal memo saying Meta expects to begin producing its Iris processor in September 2026 after a roughly six-week bug-testing window.
- Iris is part of the MTIA chip roadmap built to move high-volume inference tasks such as ranking, recommendations, and generative AI onto Meta’s own silicon.
- Meta has secured key partners for Iris design and manufacturing with Broadcom and TSMC and long-term deals for high-bandwidth memory, flash storage, and fiber networking from Samsung, SanDisk, and Sumitomo.
- The company plans to scale to about 7 gigawatts of AI compute in 2026 and roughly 14 gigawatts in 2027, is pushing lower-priced Muse Spark model access for developers, and is exploring renting excess data-center capacity to third parties.
- Investors have responded positively with a strong weekly stock gain, but the plan carries operational, supply-chain, energy and market risks that could tighten hardware availability and raise costs for smaller cloud and decentralized compute providers.