Tech News: Meta to Test In-House AI Training Chip to Reduce NVIDIA Reliance

According to reports, Meta is collaborating with 台积公司 to produce its first in-house chip designed for AI training. The company has already begun testing the chip and aims to expand its development of custom silicon to reduce reliance on external suppliers like NVIDIA.

Sources indicate that Meta has initiated a small-scale deployment of the AI training chip, with plans to scale up production if testing proves successful. This move aligns with Meta’s long-term strategy to lower infrastructure costs while making significant investments in AI. The company projects total expenditures for this year to reach between $114 billion and $119 billion, with approximately $65 billion allocated to capital expenditures—much of which will support AI infrastructure.

The newly developed chip is a specialized accelerator designed to handle AI workloads with greater energy efficiency compared to traditional GPUs commonly used for AI processing. By working with TSMC, Meta successfully completed the initial tape-out phase of the chip’s development, a critical step where the preliminary design is sent to the foundry for manufacturing. This process can take three to six months and costs tens of millions of dollars, with no guaranteed success. Should the testing phase reveal issues, Meta would need to diagnose problems and initiate another tape-out.

As of now, both Meta and TSMC have not provided official comments on the project.

?

Stay tuned for updates: ?? https://www.smbom.com/ ??

If you like this article, please give us?a like! ?? ??