Meta is about to make its own AI brain. According to Reuters, which cited an internal company memo, Meta Platforms plans to begin manufacturing an in-house artificial intelligence chip in September. The chip is codenamed "Iris."
The move is part of a larger push to dramatically expand Meta's computing muscle. Reuters reports that the effort is designed to roughly double the company's overall computing capacity, with a goal of reaching 14 gigawatts of computing power in 2027.
The Reuters story — an exclusive reported by Katie Paul, Max A. Cherney and Stephen Nellis and dated July 9 — was picked up widely by outlets including CNBC, Yahoo Finance and Techmeme.
Why build your own chip? Today, companies training and running large AI models lean heavily on outside suppliers, chiefly Nvidia and to a lesser extent AMD. As NDTV Profit and CNBC-TV18 frame it, Meta's custom silicon is aimed at cutting that reliance on external chipmakers — and, potentially, the costs that come with it.
The timeline appears to be moving quickly. According to Cybernews, the Iris chip passed testing in just six weeks with no major issues, clearing the way for the September production start.
A few caveats are worth noting. The reporting is based on an internal memo rather than a formal Meta announcement, and the details describe plans that could still shift. "Production" here means the chip entering manufacturing, not that finished Meta data centers are running on Iris tomorrow.
Why it matters: if Meta can reliably build and deploy its own AI chips at scale, it joins a small group of tech giants trying to break the industry's dependence on Nvidia — a shift that could reshape who controls the cost and supply of AI computing power.