Apple joins UALink group to take on the task of acquiring Nvidia's AI hardware dominance.

Apple has officially joined the Ultra Accelerator Link Consortium, which is made up of more than 65 members developing the next-generation AI acceleration architecture. Apple's participation will influence new standards and promote adoption. This essentially means creating and promoting a standard that allows multiple GPUs to connect for AI tasks running as data center endpoints.

Backey Lo, the head of Alibaba Cloud Platform Architecture Department, stated, "We are optimistic about UALink because it addresses connectivity issues and creates new opportunities to expand AI capabilities and requirements. Alibaba has a long history of collaborating on innovative ideas to advance our industry. And I am looking forward to joining the Board of Directors of UALink."

"What is UALink? Simply put, UALink is an industrial effort in the AI hardware market that aims to compete with Nvidia's dominance. This standard specifically targets competition with Nvidia's proprietary NVLink technology. As described by an article from Tom's Hardware last October:

UALink seeks to become an industry standard for expanding connections of many AI accelerators and is aiming to be a competitor to Nvidia's patented technology, NVLink. [...] InfiniBand is challenging the new participant Ultra Ethernet. Ultra Ethernet is a major group of tech giants creating an open standard technology to break Nvidia's dominance.

UALink Consortium includes promoter members behind AMD and Intel, such as AWS, Google, Meta, Microsoft, and now Apple.

The initial specification for version 1.0 will be the 200Gbps UALink standard that can connect up to 1K accelerators within an AI pod. This standard is based on the IEEE P802.3dj PHY layer. According to the UALink Consortium, "This standard will be provided to contributing members in 2024 and made public by the first quarter of 2025."