Arm custom chips get a boost with Nvidia partnership

In a significant development announced on Monday, Arm revealed that its central processing units (CPUs) will now seamlessly integrate with Nvidia's AI chips via the NVLink Fusion technology. This collaboration aims to facilitate a more customized infrastructure for hyperscalers, enabling them to pair Arm-based Neoverse CPUs with Nvidia's leading graphics processing units (GPUs). This partnership illustrates Nvidia's strategy of collaborating with major technology firms, positioning itself at the forefront of the AI industry. By opening its NVLink platform to a wider range of custom chips, Nvidia offers customers more flexibility rather than compelling them to utilize only its CPUs. Currently, Nvidia markets an AI product called Grace Blackwell, which combines multiple GPUs with an Nvidia-branded Arm-based CPU. Additionally, various configurations are available that utilize CPUs from Intel or AMD. Major players like Microsoft, Amazon, and Google are actively developing or implementing Arm-based CPUs within their cloud infrastructures, allowing them greater control and cost efficiency. While Arm does not manufacture CPUs, it licenses its instruction set technology essential for these chips and provides designs that expedite the development of Arm-based solutions. The recent announcement from Arm includes a new protocol for custom Neoverse chips, enhancing their ability to transfer data effortlessly alongside GPUs. Traditionally, the CPU has been the cornerstone of server architecture; however, the rise of generative AI has shifted this focus towards AI accelerator chips, predominantly Nvidia GPUs. In AI server setups, as many as eight GPUs can now be paired with a single CPU. In a related move, Nvidia announced a $5 billion investment in Intel, aiming to enhance the integration of Intel CPUs within AI servers utilizing Nvidia's NVLink technology. Although Nvidia intended to acquire Arm for $40 billion in 2020, the deal fell through in 2022 due to regulatory challenges in both the U.S. and U.K. As of February, Nvidia held a minor stake in Arm, which is primarily owned by Softbank. Notably, Softbank divested its entire stake in Nvidia earlier this month and is now supporting the OpenAI Stargate project, which plans to leverage both Arm technology and chips from Nvidia and AMD.