On Monday, Arm announced that CPUs based on its technology will be able to integrate with AI chips using Nvidia’s NVLink Fusion technology.
This move will enable customers who prefer custom infrastructure—especially large-scale cloud service providers (such as hyperscale data centers)—to combine Arm-based Neoverse CPUs with Nvidia-dominated GPUs, simplifying the process of technological integration.
This announcement is the latest example of Nvidia leveraging mergers and collaborations to push partnerships with nearly all major tech companies amid the rise of the AI industry. It signifies that Nvidia is opening its NVLink platform, allowing for broader integration with various custom chips, not just limited to its own CPUs, providing customers with more options.
Currently, Nvidia has launched an AI product named Grace Blackwell, which combines multiple GPUs with Nvidia-branded CPUs based on the Arm architecture. Additionally, Nvidia supports server configurations using Intel or AMD CPUs.
Tech giants like Microsoft, Amazon, and Google are also developing or deploying Arm-based CPUs to gain more flexible control over infrastructure settings in cloud computing environments, thereby reducing operational costs.
Arm itself does not manufacture CPUs but licenses its instruction set architecture technology to chip manufacturers, who design and produce chips based on that architecture. Arm also sells design solutions to partners, helping them to create custom chips based on Arm more quickly.
In Monday’s announcement, Arm stated that future custom Neoverse chips will include a new protocol that allows for seamless data transfer with GPUs.
Traditionally, the CPU is the core component of servers, but with the rise of generative AI infrastructure, AI acceleration chips—especially Nvidia’s GPUs—have become the computational core. In AI servers, it is common to pair up to 8 GPUs with a CPU to meet large-scale computing demands.
Notably, Nvidia announced in September that it would invest $5 billion in Intel to promote better integration of Intel’s CPUs with Nvidia’s NVLink technology for AI servers.
Nvidia had previously announced in 2020 its intention to acquire Arm for $40 billion, but the deal was stalled until 2022 due to regulatory hurdles in the US and UK. As of February 2023, Arm is still held by SoftBank, while Nvidia holds a small stake.
Meanwhile, SoftBank also sold all its shares in Nvidia earlier this month and is supporting the OpenAI Stargate project, which plans to use Arm technology along with chips from Nvidia and AMD to drive innovation in the AI industry.