US stock futures rise after Trump announces Japan trade deal
Investing.com -- Meta is testing its first in-house AI training chip, marking a major step in its effort to reduce reliance on external suppliers like Nvidia (NASDAQ:NVDA), according to a report from Reuters on Tuesday.
The small-scale deployment of the chip has begun, and if successful, Meta plans to ramp up production for wider use, said Reuters.
The move is said to be part of a broader strategy to cut costs, as Meta forecasts 2025 expenses of up to $119 billion, with $65 billion allocated to AI infrastructure.
According to Reuters, the new chip is a dedicated AI accelerator, optimized for AI-specific tasks rather than general computing, which could make it more power-efficient than traditional GPUs.
The company is working with Taiwanese chipmaker TSMC to produce the chip, a source reportedly told Reuters.
The chip is the latest in the Meta Training and Inference Accelerator (MTIA) series, a program that has faced past challenges, including scrapped designs.
However, Meta successfully launched an MTIA inference chip last year for content recommendations on Facebook (NASDAQ:META) and Instagram, and executives hope to apply its own chips to training models by 2026, said Reuters.
Chief Product Officer Chris Cox described the company’s chip development as “a walk, crawl, run situation”, adding that the inference chip had been a "big success."
While Meta remains a major Nvidia customer, the shift to in-house chips reflects broader concerns over the sustainability of large-scale AI models, Reuters noted.