24 May

Meta Ventures into Custom Chip and Data Center Design for AI Workloads

    Meta, the parent company of Facebook, has unveiled its plans to develop a proprietary chip specifically designed for running artificial intelligence (AI) models, along with a new data center architecture optimized for AI workloads.

    In a recent blog post, Santosh Janardhan, head of infrastructure at Meta, stated, “We are executing on an ambitious plan to build the next generation of Meta’s AI infrastructure and today, we’re sharing some details on our progress. This includes our first custom silicon chip for running AI models, a new AI-optimized data center design, and the second phase of our 16,000 GPU supercomputer for AI research.”

    Meta’s custom chip, known as the Meta Training and Inference Accelerator (MTIA), aims to deliver enhanced computational power and efficiency compared to existing CPUs. The chip is tailored for internal workloads such as content understanding, feeds, generative AI, and ad ranking. The initial version of MTIA was developed in 2020.

    The announcement from Meta arrives at a time when other major technology companies are also pursuing their own AI-specific chips. This trend has been driven by the increasing prevalence of large language models and generative AI. Microsoft, for instance, has reportedly partnered with chip-maker AMD to develop its own chip for AI workloads, while AWS has already released its own AI-focused chip.

    In addition to the custom chip, Meta revealed its plans for an AI-optimized data center design. The new data center will be specifically optimized for training AI models, enabling improved performance as the models ingest larger volumes of data. Janardhan emphasized that this new data center design will support liquid-cooled AI hardware and a high-performance AI network connecting thousands of AI chips to create data center-scale AI training clusters. The company expects the new data center systems to be faster and more cost-effective to deploy compared to previous facilities.

    Meta is also actively working on the development of AI supercomputers capable of training next-generation AI models, powering augmented reality tools, and supporting real-time translation technology.

    By venturing into custom chip and data center design for AI workloads, Meta aims to strengthen its AI infrastructure, enabling more advanced AI capabilities and further innovation in areas such as content understanding, augmented reality, and real-time translation.

    Share this