Meta has started testing its first in-house chip for training artificial intelligence models, marking a major step in its strategy to reduce dependence on external suppliers like Nvidia. According to sources, the company has deployed the chip on a small scale and plans to expand production if testing proves successful. This initiative aligns with Meta’s broader efforts to lower infrastructure costs while heavily investing in AI technologies to drive future growth.
The chip, part of Meta’s Training and Inference Accelerator (MTIA) series, is designed specifically for AI workloads, making it more power-efficient than traditional graphics processing units (GPUs). Unlike GPUs, which handle a variety of tasks, this dedicated accelerator focuses solely on AI training, potentially improving efficiency and performance.
Taiwan-based chip manufacturer TSMC is producing the chip for Meta. The testing phase began after Meta successfully completed its first “tape-out,” a critical stage in chip development that involves creating an initial design and sending it through a factory for manufacturing. This process typically costs tens of millions of dollars and takes several months, with no guarantee of success. If the test fails, Meta would need to identify the issues and redo the process.
Meta has been working for years to develop custom AI chips, though progress has been inconsistent. A previous attempt at a similar phase of development was scrapped. However, last year, Meta successfully introduced an MTIA chip for inference tasks, helping power recommendation algorithms for Facebook and Instagram feeds.
Now, the company aims to extend its in-house chip capabilities to training AI models, a more computationally demanding task. Meta executives have outlined a long-term plan to transition its AI training workloads to its own chips by 2026. Initially, the chips will support recommendation systems, with eventual expansion into generative AI applications like Meta AI.
With AI infrastructure costs projected to reach up to $65 billion in 2025, Meta’s move toward custom chip development reflects a broader industry trend of tech giants seeking greater control over their AI ecosystems. If successful, this shift could enhance efficiency and solidify Meta’s position in the rapidly evolving AI landscape.
