Meta Tests In-House AI Chip to Reduce Nvidia Dependency

The social media giant is piloting a custom AI training chip designed with TSMC, aiming to cut costs and scale adoption by 2026.

Meta has initiated a limited deployment of its first in-house AI training chip, part of its Meta Training and Inference Accelerator (MTIA) program.
The chip, developed in collaboration with Taiwan Semiconductor Manufacturing Co. (TSMC), is tailored for AI-specific tasks to improve power efficiency and reduce reliance on Nvidia GPUs.
If testing proves successful, Meta plans to scale up production and integrate the chip into recommendation systems and generative AI products like chatbots by 2026.
Meta's AI infrastructure spending is projected to reach up to $65 billion in 2025, driving broader efforts to lower long-term operational costs.
The chip development follows earlier challenges, including the discontinuation of a previous in-house chip, as Meta remains one of Nvidia's largest customers.