Meta is testing its first in-house chip for training artificial intelligence (AI) systems, marking a significant step in its efforts to design custom silicon and reduce reliance on suppliers like Nvidia, two sources told Reuters.
The social media giant has started a small deployment of the chip and plans to scale production if the test is successful. This initiative is part of Meta’s long-term strategy to cut infrastructure costs as it heavily invests in AI-driven growth. The company, which owns Facebook, Instagram, and WhatsApp, has projected 2025 expenses between $114 billion and $119 billion, with up to $65 billion allocated to AI infrastructure.
One source said Meta’s new chip is a dedicated AI accelerator, optimized for handling AI-specific tasks. Unlike general-purpose GPUs, this specialized chip is expected to be more power-efficient. Meta is partnering with Taiwan-based semiconductor giant TSMC for manufacturing.
The test phase began after Meta completed the chip’s first “tape-out,” a key milestone in semiconductor development. A tape-out—sending an initial design to a chip factory—can cost tens of millions of dollars and take several months. If the test fails, Meta would need to diagnose issues and redo the process, further delaying production.
Meta’s chip is the latest in its Meta Training and Inference Accelerator series, highlighting its ambition to build proprietary AI hardware for the future.
