Arm announced on Monday that central processing units utilizing its technology will be capable of integrating with AI chips via Nvidia’s NVLink Fusion technology.
This development will facilitate customers of both firms who prefer a tailored approach to their infrastructure — particularly hyperscalers — in pairing Arm-based Neoverse CPUs with the leading graphics processing units from Nvidia.
This illustrates Nvidia’s recent pattern of leveraging partnerships with nearly every significant technology entity as it positions itself at the forefront of the AI sector. The announcement indicates that Nvidia is expanding its NVLink platform to accommodate a diverse range of custom chips, moving away from mandating the use of its own CPUs by customers.
Nvidia currently markets an AI product named Grace Blackwell that combines several GPUs with an Nvidia-branded Arm-based CPU. Alternative configurations consist of servers that employ CPUs from Intel or Advanced Micro Devices.
However, Microsoft, Amazon and Google are actively developing or implementing Arm-based CPUs in their cloud infrastructures to enhance control over their setups and lower costs.
Arm does not produce CPUs but licenses its instruction set technologies required for those chips. The company also offers designs that enable partners to expedite the development of Arm-based chips.
In conjunction with Monday’s announcement, Arm specified that custom Neoverse chips will feature a new protocol designed to allow seamless data movement with GPUs.
The CPU has traditionally been regarded as the most critical component in a server. Nevertheless, generative AI infrastructures are largely reliant on the AI accelerator chip, which is predominantly an Nvidia GPU. An AI server can accommodate as many as eight GPUs paired with a CPU.
In September, Nvidia disclosed plans to invest $5 billion in Intel, the leading CPU manufacturer. A significant aspect of the arrangement was to facilitate the integration of Intel CPUs into AI servers utilizing Nvidia’s NVLink technology.
Nvidia reached a deal to acquire Arm for $40 billion in 2020, but the agreement collapsed in 2022 due to regulatory challenges in the U.S. and U.K. Nvidia maintained a minor stake in Arm, which is primarily owned by Softbank, as of February.
At the same time, Softbank sold off its entire stake in Nvidia earlier this month and is financing the OpenAI Stargate initiative, which plans to utilize Arm technology alongside chips from Nvidia and AMD.
WATCH: Nvidia’s options pricing can swing 6-7% in either direction, says Susquehanna’s Murphy
