Meta unveils next-generation AI chip to boost content organisation

Meta has announced the launch of its latest silicon innovation, the Meta Training and Interference Accelerator (MTIA) chip, designed to enhance artificial intelligence (AI) capabilities across its platforms. This next-generation chip is set to revolutionise content organisation on Meta-owned apps and services like Facebook and Instagram through improved ranking and recommendation models.

Last year, Meta had introduced its initial MTIA chips, which were instrumental in powering indigenous AI models, particularly deep-learning recommendation algorithms managing content across its social media platforms. Building on this foundation, Meta now aims to augment its AI prowess further with the MTIA chip, enhancing its ability to handle complex workloads.

In addition to its proprietary Large Language Model (LLM), dubbed LLaMA, Meta has been actively expanding its suite of generative AI tools, including custom stickers in Messenger and AI-powered chatbots. 

This strategic focus on AI has prompted a surge in demand for computing power, compelling Meta to invest in infrastructure to support AI-driven initiatives, such as data centers and custom hardware solutions.

The unveiling of the next-generation MTIA chip marks a significant milestone in Meta’s strategy to develop domain-specific silicon chips tailored to its internal requirements. By reducing reliance on external providers like Nvidia, Meta aims to achieve greater autonomy and efficiency in AI hardware management.

The trend of major technology companies venturing into in-house AI chip development is gaining momentum. Just this week, Google revealed its custom Arm-based CPU, Axion, designed to handle general-purpose workloads, including CPU-based AI training and media processing. Similarly, Microsoft boasts its own pair of Arm-based custom AI chips, geared towards AI model training and supporting Azure Cloud services.