Cover Image for Nvidia Against Apple and the Rest of the World: Apple May Have Confirmed That Its Superchip ACDC Will Use UALink Technology.
Fri Jan 24 2025

Nvidia Against Apple and the Rest of the World: Apple May Have Confirmed That Its Superchip ACDC Will Use UALink Technology.

The main competitor of Nvidia's NVLink has just received a major boost.

Apple has joined the directory of the Ultra Accelerator Link (UALink) consortium, a coalition aimed at developing an industrial standard for high-speed, low-latency communication in artificial intelligence (AI) systems in data centers. UALink is positioned as a direct competitor to Nvidia's NVLink technology, seeking to challenge its dominance in the AI accelerator market.

In addition to Apple, other renowned companies such as Intel, AMD, Google, Microsoft, Meta, HPE, Cisco, and Broadcom already support UALink. Becky Loop, Director of Platform Architecture at Apple, stated that UALink has great potential to address connectivity challenges and create new opportunities in the field of AI. She also mentioned that the company has a long tradition of fostering innovations that drive the tech industry.

Apple's involvement in the consortium may be related to an initiative known as "Project ACDC" (Apple Chips in the Data Center), which aims to develop proprietary AI chips for the company's data centers in collaboration with TSMC and Broadcom. This effort seeks to enhance Apple's new intelligence capabilities.

In addition to Apple, other companies such as Alibaba Cloud Computing and Synopsys have also decided to join the consortium. The UALink 1.0 specification, which will allow connections of up to 200 Gbps per lane for a maximum of 1,024 accelerators in an AI pod, is expected to be launched in the first quarter of 2025.

Kurtis Bowman, chair of the consortium board, commented that they are pleased to welcome new members like Alibaba, Apple, and Synopsys. Since its inception, the consortium has grown to over 65 members representing various sectors, including cloud, silicon, and software providers, as well as system manufacturers. This ongoing support will contribute to accelerating the adoption of a critical standard for interconnecting AI workloads.