Meta joins Big Tech race to build its own AI chips: Reports
Meta AI chips take shape as the firm readies its in-house Iris silicon for September production, targeting 14GW by 2027 and lower AI costs, per reports.
The next battle in artificial intelligence is shifting from software to silicon. Meta is preparing to manufacture its first production AI chip, code-named Iris, as part of a broader strategy to reduce dependence on third-party chipmakers and strengthen its AI infrastructure.
According to a Reuters-sourced memo, the company plans to begin production in September after completing successful early testing. Here's all that you need to know!
What Meta's Iris chip is designed to do
Iris is part of Meta's MTIA programme, short for Meta Training and Inference Accelerators. These are custom-designed processors built specifically to train and run AI models across Meta's platforms, including Facebook and Instagram.
The company reportedly completed initial testing in around six weeks without major issues, suggesting that development is progressing more smoothly than some of its earlier chip initiatives. By creating hardware tailored to its own AI workloads, Meta hopes to improve performance while gaining greater control over its computing infrastructure.
Why custom AI chips matter?
Developing advanced AI requires enormous computing power, and buying high-end graphics processing units (GPUs) has become increasingly expensive. Custom silicon offers an alternative.
Instead of relying entirely on suppliers such as Nvidia and AMD, Meta can optimise chips for its own applications, improving efficiency while reducing power consumption and long-term costs.
These chips are expected to support both AI training, where models learn from massive datasets, and inference, the stage where trained models generate responses, recommendations and other outputs for users. Meta also plans to accelerate development by releasing a new AI chip approximately every six months through 2027.
Building the infrastructure behind AI
The chip programme forms part of Meta's much larger AI investment strategy. The company reportedly plans to build around 7 gigawatts of AI computing capacity during 2026 before doubling that to 14 gigawatts in 2027. It also expects to invest up to $145 billion in AI infrastructure this year.
To support this expansion, Meta has secured long-term agreements with Samsung Electronics for memory, Sandisk for flash storage and Sumitomo Electric for fibre-optic equipment. Broadcom is supporting chip design, while Taiwan Semiconductor Manufacturing Co. will manufacture the processors.
What does it mean for the AI race?
Meta's Iris project hints that major AI companies are increasingly designing their own chips instead of relying solely on external suppliers. If successful, Iris could lower the cost of delivering AI features to billions of users while reducing dependence on third-party hardware.
It would also strengthen Meta's position as both an AI platform developer and a hardware innovator. The key milestones to watch are September's production launch, improvements in data centre efficiency and whether Meta can maintain its ambitious six-month chip development cycle through 2027.