Story
Meta to Begin Production of In-House AI Chip in September, Memo Shows
Summary
Meta Platforms is set to begin manufacturing its custom AI chip, codenamed 'Iris,' in September as part of a major initiative to double its computing capacity to 14 gigawatts by 2027 and reduce reliance on external suppliers.
Meta Platforms plans to begin manufacturing its first-generation in-house artificial intelligence chip in September, a crucial step in its strategy to double its computing power and reduce dependence on third-party suppliers, according to an internal company memo reviewed by Reuters.
The move signals positive momentum for the tech giant's custom silicon efforts, which aim to power the AI features across its Facebook and Instagram platforms more efficiently.
Strategic Shift to Custom Silicon
The chip, codenamed "Iris," is part of a four-generation project for Meta Training and Inference Accelerators (MTIA). According to the memo, the chip completed a six-week testing phase without any major issues, clearing the path for production. The development and timing details have not been previously reported.
Meta is collaborating with Broadcom on the chip's design and with Taiwan Semiconductor Manufacturing Co (TSMC) for its manufacturing. This initiative is designed to augment the vast quantities of GPUs Meta purchases from market leaders like Nvidia and AMD. The memo noted that adopting the latest third-party GPUs has been a "heavy lift" and has "cost us time," highlighting the strategic imperative for developing bespoke hardware.
Aggressive Infrastructure Expansion
This chip development is a core component of Meta's ambitious plan to significantly scale its AI infrastructure. The company aims to deploy 7 gigawatts of computing infrastructure this year and intends to double that figure to 14 gigawatts in 2027, the memo stated.
AdTo support this rapid expansion amid intense demand for components, Meta has secured several long-term supply agreements. Key partnerships mentioned in the memo include:
- Samsung Electronics for memory chips
- Sandisk for flash storage
- Sumitomo Electric for fiber-optic equipment
Meta plans an aggressive release schedule, aiming to launch a new in-house AI processor approximately every six months through 2027, a much faster cadence than the typical industry cycle of a year or more.
Market Context and Implications
Meta's push for custom chips comes as the tech industry grapples with soaring demand for AI computing power, leading to component shortages and rising costs—a phenomenon Morgan Stanley analysts have termed "chipflation." Meta expects to spend as much as $145 billion on AI infrastructure this year alone.
By developing its own accelerators, Meta aims to lower its massive computing expenditures and gain greater control over its hardware supply chain. This vertical integration strategy is becoming increasingly common among Big Tech firms seeking a competitive edge in the AI race. Meta, Samsung, and Sumitomo did not respond to Reuters' requests for comment, while Sandisk declined to comment.