Meta to Test In-House AI Chip to Cut NVIDIA Reliance

Meta is reportedly collaborating with TSMC to manufacture its first in-house chip designed for artificial intelligence (AI) training. The company has already begun testing the chip as part of a broader strategy to develop more customized processors, aiming to reduce dependence on external providers like NVIDIA.

According to sources familiar with the matter, Meta has initiated small-scale deployment of this AI accelerator and, if testing proceeds smoothly, plans to ramp up production for large-scale implementation. Developing proprietary chips aligns with Meta’s long-term goal of cutting infrastructure costs, as the company continues to invest heavily in AI-driven advancements.

Meta has allocated substantial resources to AI, projecting total expenditures between $114 billion and $119 billion this year, with $65 billion earmarked for capital investments. A significant portion of this budget increase is dedicated to AI infrastructure.

The newly developed training chip is a specialized AI accelerator optimized for handling AI-related workloads. Compared to traditional GPUs commonly used in AI applications, this chip is designed to offer greater energy efficiency.

Meta is currently working with TSMC on chip production and has completed the first fabrication process, known as "tape-out"—a critical stage in semiconductor development where the initial design is sent to the foundry for manufacturing. This process can take three to six months and requires an investment of tens of millions of dollars. However, there is no guarantee of success; if initial tests fail, Meta will need to diagnose issues and undergo another costly round of production.

As of now, both Meta and TSMC have declined to comment on the chip's development progress.