TL;DR
Meta's proprietary Iris AI chip entered production in September 2026, marking the company's first in-house silicon for AI workloads and anchoring a broader push to rival AWS, Azure, and Google Cloud.
What happened
- Iris chip production began September 2026, manufactured by TSMC and co-designed with Broadcom, per CEO Mark Zuckerberg's stated timeline.
- The chip targets inference workloads across Facebook, Instagram, and WhatsApp, including content ranking, recommendations, and generative AI services.
- Iris extends Meta's MTIA (Meta Training and Inference Accelerators) program, which already has the MTIA 300 deployed in production; the 450 and 500 variants target generative image and video inference through 2027.
- Meta plans a new chip iteration roughly every six months through 2027, an aggressive cadence versus traditional industry timelines.
- The Iris rollout accompanies Meta Compute, a new cloud infrastructure unit selling AI compute and foundation model access to third parties, launched the week of July 7 to 11, 2026.
Why it matters
- Reducing Nvidia and AMD dependence is the core strategic goal: shifting high-volume inference to custom silicon lowers per-unit costs and removes a critical supply-chain bottleneck.
- Wolfe Research estimates 20% EPS upside for every gigawatt of AI capacity Meta successfully monetizes at roughly $25 billion per gigawatt, a figure not yet in Wall Street consensus models.
- Meta is scaling from 7 gigawatts of compute capacity in 2026 to 14 gigawatts in 2027, a doubling that would consume more electricity than many small countries.
- 2026 capital expenditure guidance of $125 billion to $145 billion makes Meta one of the largest single-year infrastructure investors in corporate history, yet shares rose roughly 15% on the news.
- Meta has locked in long-term supply agreements with Samsung (high-bandwidth memory), SanDisk (flash storage), and Sumitomo Electric (fiber-optic networking), hardening the full stack beyond just the chip.
What to watch next
- July 29, 2026 Q2 earnings call: updated CapEx guidance and the first commercial timeline for Meta Compute will either validate or complicate the bull case; consensus revenue estimate sits at $59.8 to $60.2 billion.
- Iris bug-testing completion (roughly six weeks post-production start): successful clearance would confirm the chip can absorb meaningful inference load and begin displacing third-party GPU spend at scale.
- Meta Compute customer traction: any disclosed enterprise contracts or pricing would establish whether Meta can genuinely compete with AWS, Azure, and Google Cloud or remains a capacity-monetization side bet.
Originally published on Present of AI, a daily source-linked AI news timeline. Read the full timeline or browse the open dataset.