AI models will be prepared from its own chipset, Meta started testing its chipset

Meta has reportedly started testing its first in-house-designed chipsets, which would be used for training AI models. The company has reportedly released a limited quantity of processors for testing their stability and performance. In the case that the testing is successful, Meta can mass produce the same. These processors are in the Meta Training and Inference Accelerator (MTIA) family of chipsets.

According to the report, Meta has developed these AI chipsets in collaboration with Taiwan Semiconductor Manufacturing Company (TSMC). The company recently completed tape-out, the final stage of the chip design process, and has now started deploying these chips on a small scale.

This is not the first time Meta has designed AI-focused chipsets. Last year, the company unveiled Inference Accelerators, which were specifically designed for AI inference, however, Meta did not have any in-house hardware to train its Llama LLMs (Large Language Models) until now.

The report quoted company sources as saying that the main objective behind Meta developing its own chipsets is to reduce its AI infrastructure costs. This cost is mainly incurred in running internal AI systems, consumer-focused products, and developer tools.