Meta Platforms is preparing to kick off production of its own artificial intelligence chip this coming September, according to an internal company memo obtained by Reuters. The chip is a key piece of the social media giant’s ambitious plan to grow its total computing capacity to 14 gigawatts by next year.
The chip, which carries the internal code name “Iris,” is the third generation in a four-chip roadmap under Meta’s in-house program called Meta Training and Inference Accelerators, or MTIA. Meta intends to design the chips itself, with the goal of enhancing the AI systems that run its Facebook and Instagram platforms.
According to the memo, testing of the Iris chip wrapped up in just six weeks and turned up no significant problems — a notably fast turnaround that suggests the company is making real headway on a project that struggled to gain traction for more than five years after it was first launched.
Meta worked with Broadcom on the chip’s design and has partnered with Taiwan Semiconductor Manufacturing Co. to handle production. The custom chip approach is expected to help the company reduce its enormous computing expenses and lessen its reliance on chips purchased from outside vendors such as Nvidia and Advanced Micro Devices.
The Iris chip is intended to work alongside the large numbers of graphics processing units, or GPUs, that Meta currently buys from those suppliers for its AI operations. However, the memo noted that rolling out the latest GPUs at a company of Meta’s scale “has been a heavy lift, and it has cost us time.”
Meta publicly introduced Iris by its technical designation back in March, alongside three other AI processors. The company plans to release a new chip roughly every six months through 2027 — a much faster pace than the year-or-more intervals that most technology firms follow for AI chip releases.
The memo indicates Meta plans to deploy seven gigawatts of computing infrastructure during the current year, then double that figure in 2027. Overall, the company expects to spend up to $145 billion on AI infrastructure this year alone, representing a substantial share of the more than $700 billion that major technology companies are projected to invest in AI collectively.
To support that expansion, Meta has locked in long-term, multi-year supply deals with several companies. Those agreements include contracts with Samsung Electronics for memory chips, Sandisk for flash storage, and Sumitomo Electric for fiber-optic equipment.
Such supply agreements have taken on added importance as a shortage of memory chips has pushed some companies, including Apple, to raise prices. Analysts at Morgan Stanley have noted that rapid and significant price increases across memory and other chip categories have made what they call “chipflation” a broader economic concern.
Sandisk declined to comment on the report. Samsung Electronics and Sumitomo Electric did not respond to requests for comment. Meta also declined to comment.
