Arm, a major British semiconductor design company, said today that central processing units (CPUs) built with its technology will be integrated with the NVLink Fusion technology of artificial intelligence (AI) chip giant NVIDIA (Nvidia) in the future.

Financial media CNBC reported that this move will make it easier for customers of the two companies, especially those hyperscale data center operators who prefer customized infrastructure, to use Arm-based Neoverse CPUs with NVIDIA graphics processors (GPUs).

This is the latest example of NVIDIA integrating almost all major technology companies into its AI ecosystem through cooperation. It also shows that NVIDIA is gradually opening up its NVLink platform so that it can be integrated with more customized chips, rather than requiring customers to only use NVIDIA's own CPUs.

The AI product Grace Blackwell system currently sold by NVIDIA is composed of multiple GPUs paired with NVIDIA's own Arm architecture CPU; other server configurations include Intel or AMD CPUs.

However, Microsoft, Amazon and Google are all developing or deploying Arm-based CPUs in their cloud services to enhance control over the system and reduce costs.

Amou itself does not produce CPUs, but licenses its instruction set technology for use on chips. Arm also sells design drawings to help partners build Arm architecture processors more quickly.

Anmou also said today that the customized Neoverse chip will join a new protocol to enable it to smoothly transmit data with the GPU.

In traditional server architecture, the CPU has always been the most important component. But the infrastructure of generative AI is centered on AI accelerators, most of which are NVIDIA GPUs. Up to 8 GPUs and one CPU can be paired together in an AI server.

NVIDIA announced in September that it would invest US$5 billion (approximately NT$156 billion) in Intel, the leader in the CPU industry. One of the key points of the agreement is to enable Intel's CPUs to be integrated into AI servers through NVIDIA's NVLink technology.

NVIDIA reached an agreement in 2020 to acquire Arm for US$40 billion. However, due to regulatory concerns in the United States and the United Kingdom, the acquisition fell through in 2022. As of February this year, NVIDIA still holds a small stake in Arm, and Arm’s major shareholder is Japan’s Softbank.

SoftBank sold out all of its NVIDIA shares earlier this month and fully supports OpenAI’s “Stargate” project. The project is expected to use ARM technology and chips from NVIDIA and Advanced Micro Devices.

Further reading: Arm is ambitious and expects to reach 50% of the data center CPU market share by 2025 SoftBank Masayoshi Son sells out NVIDIA shares and places heavy bets on OpenAI. Whether AI is a bubble has caused controversy among many parties. Why can’t NVIDIA acquire Arm? It turns out that it was doomed to fail from the beginning. NVIDIA's acquisition of Arm officially announced that it was broken, and Arm planned an IPO.