Meta is starting a new initiative called Meta Compute, aiming to build tens of gigawatts of energy infrastructure this decade, with plans to scale up to hundreds of gigawatts over time. Santosh Janardhan and Daniel Gross will lead this effort, which they see as a strategic advantage in their operations.

Mark Zuckerberg announced that Meta will unveil new AI models and products in the coming months, focusing on AI-driven commerce. He emphasized the unique value of Meta’s access to personal data for creating personalized shopping tools. The company plans significant infrastructure investments to support these efforts.

This article explains how Meta is using backend aggregation (BAG) to connect thousands of GPUs across multiple data centers for its Prometheus AI cluster. BAG facilitates high-capacity networking, enabling the infrastructure to meet the demands of large-scale AI applications. It details the technical aspects of BAG's design and implementation, emphasizing performance and reliability.

The article discusses Meta's significant investment of $75 billion in AI infrastructure, highlighting the strategic importance of this move in enhancing their technological capabilities and competing in the AI landscape. It analyzes the implications of this investment for both Meta and the broader tech industry.

Meta has entered a six-year agreement to spend over $10 billion on Google cloud services, focusing on artificial intelligence infrastructure. This deal comes as Google aims to compete with larger cloud providers like Amazon Web Services and Microsoft Azure, while Meta seeks to enhance its cloud capabilities amid heavy investments in AI.

Meta plans to invest up to $72 billion in AI infrastructure throughout 2025 as the competition for computing power intensifies among tech giants. This substantial investment is aimed at enhancing Meta's capabilities in artificial intelligence and maintaining its competitive edge in the rapidly evolving tech landscape.

Charlotte Qi discusses the challenges of serving large language models (LLMs) at Meta, focusing on the complexities of LLM inference and the need for efficient hardware and software solutions. She outlines the critical steps to optimize LLM serving, including fitting models to hardware, managing latency, and leveraging techniques like continuous batching and disaggregation to enhance performance.

Meta Platforms is moving forward with a strategy to share the financial burden of AI infrastructure by selling $2 billion in data center assets. The company aims to attract external partners for co-developing data centers, reflecting a trend among tech giants to mitigate the soaring costs associated with AI and data center operations.