Nvidia is expanding its product offerings to include standalone CPUs as part of a strategic shift to reach customers needing less compute-intensive solutions. This move follows a significant investment by Nvidia in a startup specializing in low-latency AI computing technology. The company’s efforts are underscored by a recent multiyear agreement with Meta, which plans to purchase billions of dollars’ worth of Nvidia chips to enhance its computing infrastructure.
Meta has previously indicated plans to acquire 350,000 H100 chips by the end of 2024, aiming for a total of approximately 1.3 million GPUs by 2025, although not all will necessarily come from Nvidia. The latest deal involves Meta building hyperscale data centers designed for AI training and inference, utilizing millions of Nvidia’s Blackwell and Rubin GPUs alongside significant CPU deployment.
This partnership marks Meta as the first major tech company to order Nvidia's Grace CPU as a standalone product. Nvidia's new Vera Rubin superchip is designed to integrate various chip technologies, reflecting the evolving landscape of AI applications that increasingly demand CPU capabilities similar to traditional cloud services.