Meta Platforms will manufacture its own AI chips, which have been referred to as 'Iris,' starting in September as it steps up its efforts to build its own AI hardware. The effort will help the company to increase its AI computing capacity to 14 GW by 2027 while reducing its reliance on third-party chip manufacturers. This development came to light after Reuters reported on an internal Meta memo.
This new effort becomes a major part of Meta’s larger AI strategy. With demand for AI models growing, Meta is making huge investments in the necessary infrastructure.
Iris is one of the processors included in Meta's Training and Inference Accelerator (MTIA) project. The tech giant developed this chip for its AI workload on Facebook and Instagram.
The memo revealed that the processor was tested over six weeks, with no major problems found. The rapid testing process reveals greater momentum behind Meta's in-house chip design program, despite some delays over the past few years. Meta aims to launch four generations of its AI chip six months apart through 2027.
While most GPUs are general-purpose processors, Iris has been designed specifically for Meta's AI systems.
Meta is collaborating with Broadcom to design the Iris chip, while TSMC will manufacture it. This approach indicates the company's overall trend toward developing proprietary AI chips instead of depending on vendors such as NVIDIA and AMD.
Even though Meta will continue to use third-party GPUs, custom chips will help reduce costs and increase efficiency in the long run. This approach will make the company more competitive amid increasing AI investment across the technology industry.
Meta's latest investment highlights the importance of vertically integrated AI infrastructure. With the stiff competition among hyperscalers, it is becoming increasingly important to own AI models and hardware as well.
Also Read: Can Meta AI Use Your Instagram Photos? Muse Image Launch Triggers Privacy Scrutiny