Meta is moving one of its own AI chips from the lab to the factory floor.
According to MSN, the company plans to put a data center chip code-named "Iris" into production in September, part of an effort to double its computing capacity. The chip is one piece of a four-generation project called the Meta Training and Inference Accelerators, or MTIA, which Meta is designing in-house rather than buying entirely off the shelf.
The move follows a technical milestone. According to Tech Times, Iris has passed its testing phase, clearing the way toward the production step. The same report notes that Broadcom, which helps design the chip, now works on custom silicon for three rival firms — a sign of how much demand there is for companies to build their own AI hardware.
Designing chips internally is a strategic bet. Training and running large AI models requires enormous amounts of specialized computing power, and much of that capacity today comes from a small number of outside suppliers. By building MTIA accelerators itself, Meta aims to add capacity on its own terms and lean less heavily on those vendors.
Why it matters: if Meta's home-grown Iris chip reaches mass production and delivers the capacity boost the company is targeting, it would mark a meaningful step for one of the world's largest tech firms toward controlling the hardware that powers its AI ambitions — and reducing its dependence on external chip makers.