(Reuters) – Big technology companies including Meta, Microsoft, Advanced Micro Devices and Broadcom said on Thursday they had developed a new industry standard for AI data center networking, the latest effort to disrupt the dominance of market leader Nvidia.
The “Ultra Accelerator Link” is an attempt to establish an open standard for communication between artificial intelligence accelerators, systems that help process the vast amounts of data used in AI tasks.
Other members of the group include Alphabet Inc.'s Google, Cisco Systems Inc., Hewlett Packard Enterprise Co. and Intel Corp.
Why is this important?
Nvidia, the largest player in the AI ​​chip market with about 80% share, is not included in this group.
Tech giants like Google and Meta are eager to reduce their reliance on Nvidia, whose networking business is a key part of their AI dominance.
Marvell Technologies, Broadcom's biggest rival in the networking and custom chip markets, is not included in the group.
Important Quotes
“Industry specifications will be key to standardizing interfaces for AI and machine learning, HPC and cloud applications for next-generation AI data centers and deployments,” the companies said in a statement.
context
Technology companies are pouring billions of dollars into the hardware needed to support AI applications, driving demand for AI data centers and the chips that run them.
The Ultra Accelerator Link group designed the specifications to manage the connections between the various accelerators in a data center.
The specification is expected to be available to companies participating in the Ultra Accelerator Link (UALink) consortium in the third quarter of 2024.
reaction
Nvidia and Marvell did not immediately respond to Reuters requests for comment.
(Reporting by Arshiya Bajwa in Bengaluru; Editing by Tasim Zahid)
