TL;DR
Meta's in-house Iris AI chip has entered mass production at TSMC's 3nm node, marking the end of Meta's reliance on third-party GPUs to serve four billion daily users.
What happened
- Meta's Iris chip (part of the MTIA family) cleared six weeks of validation testing before entering production
- TSMC's 3nm facility is the manufacturing site, placing Iris among the most advanced silicon in commercial production
- The chip is purpose-built for AI inference at Meta scale: Facebook, Instagram, and related services serving four billion daily users
- Validation clearance and production entry were confirmed as of September 21, 2026
Why it matters
- Third-party GPU dependency ends: Meta no longer needs to route inference workloads through Nvidia or AMD hardware for its core consumer products
- At four billion daily users, even marginal per-query cost reductions translate to hundreds of millions of dollars in annual savings
- 3nm production signals Iris is competitive on performance-per-watt with leading merchant silicon, not a fallback option
- Meta joins Google (TPU) and Amazon (Trainium/Inferentia) in owning the full inference stack, reshaping the merchant GPU market for hyperscale inference
- Reduced GPU procurement gives Meta supply chain independence during periods of chip shortage or geopolitical disruption
What to watch next
- Nvidia revenue exposure: watch for any guidance revision from Nvidia if Meta publicly quantifies GPU displacement at its scale
- Yield and ramp rate at TSMC 3nm will determine how quickly Iris can absorb Meta's full inference load versus a hybrid GPU-plus-Iris fleet
- Whether Meta licenses or publishes Iris architecture details, which would signal a broader platform ambition beyond internal use
Originally published on Present of AI, a daily source-linked AI news timeline. Read the full timeline or browse the open dataset.