Meta Platforms ( NASDAQ:META ) is testing an in-house AI training chip , aiming to lower infrastructure costs and reduce reliance on Nvidia ( NASDAQ:NVDA ).
The move signals Meta's deeper push into artificial intelligence as it seeks greater control over its AI hardware. According to Reuters, the company has begun small-scale deployment of the chip, with plans to scale production if testing proves successful.
Meta currently spends billions on Nvidia's GPUs, a key component of its AI operations, and hopes its own chip can improve cost efficiency while optimizing tasks like recommendation systems and generative AI. The chip is part of Meta's Training and Inference Accelerator (MTIA) series and is being produced by Taiwan Semiconductor Manufacturing Co. ( NYSE:TSM ).
Meta recently completed the tape-out phase, a critical milestone in chip development, though success is not guaranteed. A failed test would force the company to troubleshoot and restart, adding months of delays and significant costs. Meta aims to integrate the chip into AI training systems by 2026. If successful, the initiative could weaken Nvidia's dominance in AI hardware and mark a shift in Big Tech's approach to AI infrastructure.
This article first appeared on
GuruFocus
.