On July 9, an internal memo from Meta Platforms revealed that the company is slated to commence the production of a cutting-edge AI chip in September. This chip represents a significant milestone in Meta's four-generation self-developed initiative for the Meta Training and Inference Accelerator (MTIA), internally codenamed 'Iris.' The project's overarching goal is to boost the total computing power to an impressive 14 gigawatts by the following year.

Meta aims to leverage these custom chips to significantly enhance the AI capabilities embedded within its social media behemoths, Facebook and Instagram. The memo highlighted that the chip underwent a rigorous testing phase that spanned a mere six weeks, during which no major issues were encountered. This swift and successful testing period underscores the relatively rapid progress made on the project, offering a beacon of optimism for an initiative that has weathered its fair share of setbacks since its inception five years ago.

In crafting these bespoke chips, Meta tailors them to its unique requirements, collaborating closely with industry leaders Broadcom for design expertise and TSMC for manufacturing prowess. This strategic move is anticipated to not only slash computing costs but also diminish Meta's reliance on external chip suppliers, such as Nvidia and AMD.

Prior to this revelation, there had been no public reports regarding the completion of vulnerability testing or the precise production timeline for this chip. Meta, maintaining its customary discretion, declined to provide further comment on the matter.