Notably absent from the group is Nvidia, the largest producer of AI accelerators, which offers its own proprietary interconnect tech for linking GPUs within a data center server. Amazon Web Services (AWS) is also not contributing to UALink. The biggest beneficiaries of UALink are expected to be Microsoft, Meta, and Google, which have spent billions on Nvidia GPUs and are looking to reduce their dependence on a vendor they see as dominant in the AI hardware ecosystem.
Key takeaways:
- Several tech giants including Intel, Google, Microsoft, and Meta are forming a new industry group, the Ultra Accelerator Link (UALink) Promoter Group, to develop components that link AI accelerator chips in data centers.
- The UALink Promoter Group is proposing a new industry standard to connect AI accelerator chips within servers, with the first version of the standard, UALink 1.0, set to connect up to 1,024 AI accelerators.
- Nvidia, the largest producer of AI accelerators, is noticeably absent from the group's members and has not shown support for UALink, likely due to its own proprietary interconnect tech and strong market position.
- The UALink Consortium will be created in Q3 to oversee the development of the UALink spec, with UALink 1.0 available to companies that join the consortium and an updated spec, UALink 1.1, expected in Q4 2024.