Meta is testing its own chip for training AI systems, Reuters reports, citing sources.
According to the agency, the new processor is a specialized accelerator aimed at solving specific problems for artificial intelligence. This approach makes the chip more energy efficient compared to integrated graphics processors traditionally used for AI workloads.
The company is collaborating with Taiwan’s TSMC. At the moment, the initial stage of development has been completed, which includes creating and sending prototypes to the chip factory for testing.
Developing its own processors is part of Meta’s plan to reduce infrastructure costs. It is betting on artificial intelligence to ensure growth.
The corporation predicts spending in 2025 in the amount of $114-119 billion, of which $65 billion will be directed to the artificial intelligence sector.
Meta wants to start using its own chips for AI tasks in 2026, Reuters writes.
In May 2023, the company introduced two specialised processors for artificial intelligence and video processing tasks.
Earlier, the media learned about OpenAI working on its own AI processor in partnership with Broadcom and TSMC.
The Chinese company ByteDance is developing a similar product in collaboration with Broadcom.