Meta to Develop Its Own AI Chips Starting September

Mark Zuckerberg’s Meta Inc. plans to start designing its own artificial intelligence chips in September, joining a trend among major AI companies to manufacture their own components amid rising demand.

Codenamed “Iris,” this initiative focuses on creating custom silicon for the AI systems that support platforms like Facebook and Instagram. This news, reported recently, highlights Meta’s commitment to enhancing its technological capabilities.

Meta is expected to invest around $145 billion in AI infrastructure this year and is working in partnership with Broadcom, based in Palo Alto, California, for design, and Taiwan Semiconductor Manufacturing for production.

The company is not alone; many technology firms are increasing their in-house chip development to lower costs and reduce reliance on Nvidia, which has been a major player in the AI chip market with its powerful semiconductors.

However, despite these moves, the semiconductor industry faces enduring demand challenges. The efforts to gain more independence among AI firms are not a cure-all for supply chain issues.

With demand for chip manufacturing, packaging, and other resources outpacing supply, some specialized processes remain under control of a few companies already operating at full capacity, even after significant investments to expand.

This initiative is part of Meta’s ongoing effort to create its own chips, with a training and inference accelerator program that began over five years ago. Progress has been slow, but recent developments are reportedly moving more quickly. A test phase only took six weeks without any major setbacks, and Meta aims to roll out a new chip roughly every six months through 2027, despite the typical cycle being over a year.

Meta’s goal is to double its computing infrastructure by 2027. The new chips are meant to supplement the extensive GPUs that the company currently sources from Nvidia and AMD for its AI tasks.

However, a memo highlighted that integrating the latest GPUs at Meta’s scale presents a complex and lengthy challenge.

Switching to custom chips could help cut costs and diversify the supply chain, according to industry insiders. One analyst expressed a desire to have leverage in negotiations with Nvidia’s CEO Jensen Huang, underscoring the competitive landscape.

Besides Meta, tech giants like Amazon, Google, and Microsoft are also developing their own chips. OpenAI has recently released its first custom inference chip with Broadcom, while Anthropic is reportedly discussing chip development with Samsung.

Apple recently unveiled plans to invest over $30 billion in Broadcom over the next five years to boost its manufacturing facilities in Fort Collins, Colorado. They already design their own chips for various products and are reportedly developing processors for AI servers.

Meanwhile, Samsung creates advanced chips for its own products as well as others, and Intel is trying to enhance its contract manufacturing as it has fallen behind in production technology, according to reports.

Achieving chip autonomy is complicated, as these manufacturers rely heavily on lithography equipment from ASML, the only supplier of the advanced machinery necessary for AI chip production.

The Post has reached out to Meta, Broadcom, and Taiwan Semiconductor Manufacturing for comments.