Meta to Design Custom AI Chips, Joining Tech Giants Seeking Independence
The social media giant aims to reduce reliance on Nvidia and control costs with its in-house silicon initiative.
Meta plans to design its own artificial intelligence chips in-house, a move expected to begin in September as part of a broader industry trend among major technology companies to develop custom silicon amid high demand. The initiative, codenamed "Iris," is focused on creating specialized chips to enhance the artificial intelligence systems that power Meta's platforms, including Facebook and Instagram.
This strategic shift aims to reduce the company's dependence on external chip manufacturers like Nvidia, which currently dominates the AI chip market with its powerful graphics processing units (GPUs). Meta anticipates significant investment in AI infrastructure, projecting spending of up to $145 billion this year. For its custom chip development, Meta is collaborating with Broadcom on design and Taiwan Semiconductor Manufacturing on production.
The social media company is working with Broadcom on design and Taiwan Semiconductor Manufacturing on production. The development of the "Iris" chip has reportedly progressed rapidly, with initial testing completed in just six weeks without major issues. Meta intends to release new chip iterations approximately every six months through 2027, a faster cycle than the typical annual or longer release schedule for AI chips. The company aims to double its computing infrastructure by 2027.
These custom-designed chips are intended to supplement the substantial number of GPUs Meta purchases from Nvidia and AMD for its AI workloads. However, integrating these high-demand GPUs at Meta's scale has presented challenges. Developing custom chips offers the potential to lower expenses and diversify supply chains, according to industry analysis.
Meta joins other major technology firms, including Amazon, Google, and Microsoft, which have established their own in-house chip development programs. OpenAI has recently introduced its first custom inference chip designed with Broadcom, and Anthropic is reportedly in discussions with Samsung for similar development.
This push for in-house chip design reflects a wider industry effort to gain more control over AI infrastructure and mitigate the impact of strained global supply chains. Despite these efforts, the semiconductor industry continues to face demand that outpaces supply, with specialized manufacturing processes often controlled by a limited number of companies operating at full capacity while investing heavily in expansion. The complexity of achieving chip autonomy is further highlighted by reliance on specialized lithography equipment from a single Dutch company, ASML, for the most advanced AI chip production.
Apple also recently announced a $30 billion investment with Broadcom over five years to aid in expanding manufacturing capabilities. While Apple already designs its own chips for consumer devices, it is reportedly developing separate processors for AI servers. Samsung manufactures advanced chips for its products and external clients, and Intel is working to bolster its contract manufacturing business.