Meta to begin production of in-house AI chip in September

Meta Platforms plans to begin manufacturing its in-house artificial intelligence chip, code-named Iris, in September as it seeks to reduce computing costs and expand its AI infrastructure, according to an internal memo.

The chip forms part of a four-generation programme of Meta Training and Inference Accelerators, or MTIA, designed to improve the AI systems used across Facebook and Instagram.

Meta is working with Broadcom on the chip’s design, while Taiwan Semiconductor Manufacturing Co will handle production. Iris is intended to complement, rather than replace, the large number of graphics processing units Meta buys from Nvidia and Advanced Micro Devices.

Testing of the chip was completed in six weeks without any major problems, the memo showed. Meta declined to comment on the production timeline and testing results.

The company has faced difficulties in deploying the latest GPUs across its operations, with the memo describing the process as costly and time-consuming. Developing custom chips could give Meta greater control over its computing systems and reduce its reliance on external suppliers.

Meta disclosed Iris under its technical name in March alongside three other AI processors. The company plans to introduce a new chip roughly every six months through 2027, compared with the annual or longer release cycles commonly followed by other chip developers.

Meta also plans to double its computing capacity to 14 gigawatts in 2027, after targeting seven gigawatts of infrastructure deployment during 2026.

The company added one gigawatt of capacity during the first half of this year and expects to deploy another 5.5 gigawatts by the end of 2026, according to the memo.

Meta expects to spend as much as $145 billion on AI infrastructure this year. The amount represents a sizeable share of more than $700 billion in projected AI spending by major technology companies.

To support its data centre expansion, Meta has entered into long-term supply agreements with Samsung Electronics for memory chips, Sandisk for flash storage and Sumitomo Electric for fibre-optic equipment.

The agreements are aimed at securing critical components as demand from technology companies places pressure on global supplies of memory, processors and data centre equipment.

Sandisk declined to comment, while Samsung Electronics and Sumitomo Electric did not respond to requests for comment.

Meta shares initially declined following the report but later recovered after the company announced that developers would be given access to an AI coding model designed to compete with products from OpenAI and Anthropic. The shares were trading 4.6% higher in late afternoon trading.