Meta’s Iris AI Chip Enters Production in September, Challenging Nvidia’s Dominance

Meta Platforms plans to begin manufacturing its in-house artificial intelligence chip, code-named Iris, from September as it works toward expanding its computing capacity to 14 gigawatts next year, according to an internal memo reviewed by Reuters. The chip is part of the company’s Meta Training and Inference Accelerators (MTIA) program, a multi-generation effort to develop custom AI silicon for the systems powering Facebook and Instagram.

The memo indicates that Iris recently completed testing after a six-week validation process that identified no major issues. Reuters reported that both the testing milestone and the planned production timeline had not previously been disclosed. Meta declined to comment.

Meta is developing Iris in collaboration with Broadcom, which is assisting with the chip’s design, while Taiwan Semiconductor Manufacturing Co. will manufacture the processor. The company is building the chip for its own infrastructure with the goal of strengthening the AI systems behind its products while relying more on custom-designed hardware.

Rather than replacing graphics processing units from Nvidia and AMD, Iris is intended to complement the large GPU deployments Meta already uses for AI workloads. According to the internal memo, integrating the latest GPU generations across Meta’s infrastructure “has been a heavy lift, and it has cost us time.”

Meta first introduced Iris under its technical name in March alongside three additional AI processors. The company plans to release new AI chips approximately every six months through 2027, a faster cadence than the annual refresh cycle commonly followed for AI hardware.

The memo also outlines Meta’s broader infrastructure expansion plans. The company expects to deploy seven gigawatts of computing infrastructure this year before increasing that figure to 14 gigawatts next year.

Supporting that buildout will require significant investment and long-term component supply agreements. Meta expects to spend as much as $145 billion on AI infrastructure this year, representing part of the more than $700 billion that major technology companies are projected to invest in AI infrastructure.

To secure critical components, Meta has entered multi-year agreements with Samsung Electronics for memory chips, Sandisk for flash storage, and Sumitomo Electric for fiber-optic equipment, according to the memo. Sandisk declined to comment, while Samsung Electronics and Sumitomo Electric did not respond to requests for comment.

The agreements come as demand for AI infrastructure continues to increase competition for key data center components. Reuters noted that shortages of memory chips have made long-term supply arrangements increasingly important, while Morgan Stanley analysts said rising prices for memory and other semiconductors have become significant enough that “chipflation” has emerged as a macroeconomic concern.

This analysis is based on reporting from CNBC.

Image courtesy of CNBC18TV.

This article was generated with AI assistance and reviewed for accuracy and quality.