Meta Tests In-House AI Chip with TSMC, Targets Nvidia Independence by 2026

Meta is testing its first in-house AI training chip, developed in collaboration with Taiwan Semiconductor Manufacturing Company (TSMC), as part of a strategy to reduce reliance on Nvidia and lower infrastructure costs. The company has projected AI-related expenses of up to $65 billion in 2025, as it aims to enhance its AI capabilities for applications like content recommendations and generative AI. This new chip, which is currently in small-scale deployment, is designed specifically for AI tasks, potentially offering better efficiency than traditional GPUs. Meta's previous attempts at custom chip development faced challenges, but the success of its earlier MTIA inference chip gives hope for this new initiative. The company plans to ramp up production if testing proves successful, with ambitions to fully utilize the chips for training by 2026. As the AI chip market grows rapidly, this move could significantly impact Meta's operational costs and competitive positioning in the AI landscape.