- Major tech firms, including Meta, Microsoft, AMD, and Broadcom, have developed the “Ultra Accelerator Link” to standardise AI data centre networking.
- The initiative aims to reduce dependence on Nvidia, which holds around 80% of the AI chip market.
- Specifications for the new standard will be available in the third quarter of 2024 to companies joining the UALink Consortium.
Major tech firms, including Meta, Microsoft, AMD, and Broadcom, have launched the “Ultra Accelerator Link,” a new industry standard for networking in AI data centres. This initiative aims to reduce reliance on Nvidia, the current market leader in AI chips, by establishing open communication standards for AI accelerators.
Purpose
The effort is aimed at reducing the market dominance of Nvidia, which currently holds around 80% of the AI chip market. By creating this new standard, the consortium hopes to foster greater interoperability and innovation in AI data centres.
Also read: Stock grow rapidly, prospect of NVIDIA AI market is bright
Also read: Nvidia hits record high, nears Apple in market value
Industry response and members
The Ultra Accelerator Link consortium also includes other major industry players such as Alphabet-owned Google, Cisco Systems, Hewlett Packard Enterprise, and Intel. Notably absent from the group are Nvidia and Broadcom’s central rival in the networking and custom chip market, Marvell Technologies. The absence of Nvidia is significant, given its dominant position in the AI chip market. Tech giants like Google and Meta are particularly interested in reducing their dependence on Nvidia, whose networking business forms a vital part of the infrastructure that supports its AI dominance. The consortium’s formation underscores the industry’s desire for greater diversity and flexibility in AI hardware solutions.
Specifications and future plans
The Ultra Accelerator Link group has developed specifications to govern connections among different accelerators within a data centre. These specifications are intended to standardise interfaces for AI, Machine Learning, High-Performance Computing (HPC), and Cloud applications in the next generation of AI data centres. The consortium has announced that these specifications will be made available in the third quarter of 2024 to companies that join the Ultra Accelerator Link (UALink) Consortium. This move is expected to drive significant advancements in AI data centre infrastructure. A spokesperson for Nvidia declined to comment, and Marvell did not immediately respond to a request for comment.
