New industry standards for connectivity in AI data centers have been developed in collaboration with leading technology companies such as Meta, Microsoft, AMD and Broadcom. The move is aimed at challenging Nvidia’s dominance in the market. The “Ultra Accelerator Link” standard seeks to establish open protocols for communication between artificial intelligence accelerators, which are systems designed to handle large amounts of data used in AI projects.
Other major players include Alphabet-owned Google, Cisco Systems, Hewlett Packard Enterprise and Intel. Notably, Nvidia, a major player in the AI chip market, retains an estimated 80% stake and is not part of the consortium. Tech giants like Google and Meta are keen to reduce their reliance on Nvidia, whose AI dominance networking business plays a key role.
Marvell Technologies, Broadcom’s main competitor in the communications and custom chip markets, is also absent from the group. According to industry sources, standardizing interfaces for AI, machine learning, high-performance computing and cloud services is essential for the deployment of next-generation AI data centers
Tech companies are investing heavily in the hardware needed to support AI applications, increasing the demand for AI data centers and powerhouse chips. The Ultra Accelerator Link Group has created cases involving links between accelerators in the same data center. This information will be available for companies driving the Ultra Accelerator Link (UALink) industry in the third quarter of 2024.