Meta will begin manufacturing its in-house AI chip, code-named Iris, in September — a milestone in the company's multi-year push to reduce its dependence on NVIDIA for AI infrastructure.
Iris is the fourth generation of Meta's MTIA (Meta Training and Inference Accelerators) programme, designed in partnership with Broadcom and manufactured by TSMC. According to an internal memo, the chip cleared its bug-testing phase in roughly six weeks without turning up any significant problems — an unusually smooth validation cycle for custom silicon at this scale.
The chip is designed to handle Meta's core AI workloads: the ranking and recommendation systems that determine what appears in Instagram and Facebook feeds, and the generative AI tasks powering chatbots, content creation tools and the company's broader AI product suite. If successful, Iris could eventually power both training and inference across Meta's products.
The production timeline fits into a broader custom silicon strategy Meta formalised with Broadcom earlier this year, when the two companies agreed to expand their partnership on custom AI chips through 2029 covering multiple MTIA generations. Building custom chips does not mean Meta will immediately stop purchasing from NVIDIA, but each successful MTIA generation reduces dependence on external GPU suppliers and their pricing.
The infrastructure numbers are staggering. Meta targets 7 gigawatts of compute infrastructure by the end of 2026, doubling to 14 gigawatts in 2027. The company's 2026 capital expenditure guidance ranges from $125 to $145 billion — primarily allocated to data centres, GPUs and custom silicon. A recently announced $10 billion, 1-gigawatt data centre in Alberta, Canada is part of this expansion.
Meta's stock jumped more than 7 per cent on 10 July following an internal memo revealing plans to double total computing power by 2027. The company also implemented layoffs of approximately 8,000 employees — about 10 per cent of its workforce — as part of an AI-focused restructuring, with an additional 7,000 employees reassigned to AI-focused teams.
For context engineers, Meta's Iris chip joins OpenAI's Jalapeño and Google's TPU programme in a broader trend of frontier AI companies designing their own silicon. The economics are compelling: custom chips optimised for specific workloads can deliver significant cost reductions at scale, and the companies consuming the most AI compute have the strongest incentive to build rather than buy.
