Meta is developing its first in-house AI training processor. According to Reuters, the company has begun pilot testing a small batch of the new AI training chip and will order more if the testing phase yields positive results. Notably, this line of processors is currently manufactured by TSMC.
Meta’s decision to develop its own AI chips is also part of a long-term strategy to reduce its dependence on third-party suppliers, especially Nvidia, the world’s leading manufacturer of graphics processing units (GPUs) used in AI-related tasks. According to preliminary estimates, the project is expected to cost Meta a total investment of up to $119 billion by the end of 2025—most of which will be used to build AI infrastructure.
The chip designed by Meta is essentially a dedicated AI accelerator. It is optimized for AI-related tasks rather than general computations. Such a dedicated architecture could make the chip more energy efficient than GPUs currently used in AI training.

Meta initially plans to use the chip in its recommendation algorithms that determine what content appears on Facebook and Instagram. The company's ultimate goal is to scale the chip to support its generative AI products, such as an AI chatbot called Meta AI.
Meta’s journey into custom chip development has been mixed so far. The company previously scrapped an in-house inference chip after a failed pilot, opting instead to buy billions of dollars worth of Nvidia GPUs. The new project, however, appears to be working out better. Meta has now passed a key “tape-out” milestone in its development.
Meta isn’t the only software company deciding to build its own AI chips to reduce its reliance on third parties. OpenAI is also finalizing the design of its first custom AI training chip. The new chip will likely feature a systolic array architecture along with high-bandwidth memory, similar to Nvidia’s latest AI accelerator. This architecture is known for its high performance and efficiency in handling dense, complex calculations.