Meta to put its in-house “Iris” AI chip into production in September, aiming to double computing capacity
Category: AI Infrastructure
What it is
Meta plans to begin manufacturing its in-house AI chip, code-named Iris, from
September, as part of a plan to boost computing capacity to 14 gigawatts next
year, per an internal memo reviewed by Reuters. Iris is part of the four-generation
MTIA (Meta Training and Inference Accelerator) program, designed with Broadcom
and manufactured by TSMC. Testing took six weeks with no major issues found.
The chip aims to reduce Meta’s reliance on Nvidia and AMD GPUs as the company
guides 2026 AI capex to as much as $145 billion
Why it Matters for Enterprises
Custom silicon at hyperscaler scale signals long-term downward pressure on AI
compute costs. Enterprises planning multi-year infrastructure budgets should
track chip diversification trends when negotiating cloud and GPU capacity.
