AINEWS
The two companies now have formal votes on the technical roadmap of the open source machine learning framework most researchers and engineers use; Ant Group joined at the tier below.
Alibaba Cloud and Cambricon have joined the PyTorch Foundation at Platinum tier, the level that gives them formal seats on the foundation's Governing Board and Technical Advisory Council alongside existing member Huawei. Ant Group joined at Gold, one tier below, at the same time. The four companies appeared together in Shanghai at PyTorch Conference China this week.
PyTorch is the open-source machine-learning framework most academic researchers and production engineering teams use to build and train neural networks. It originated at Facebook's AI lab in 2016 and is now stewarded by the PyTorch Foundation, a project under the Linux Foundation. The framework sits underneath a wide stack of related open-source projects including DeepSpeed (a Microsoft-backed training optimizer), Ray (a distributed-computing system), Safetensors (a model-weight format), Helion (a kernel compiler), and vLLM (an inference server), all of which are now part of the foundation's extended orbit.
Platinum is the highest member tier. Members at that level take seats on the Governing Board, which sets strategic direction, and on the Technical Advisory Council, which weighs in on the framework's technical roadmap. The seats are the mechanism by which hardware-vendor priorities enter the room where upstream support for new accelerators gets decided.
Cambricon, founded in 2016 as one of China's first dedicated AI chip designers, is joining on the strength of its MLU line, processors built specifically for machine-learning workloads. Cambricon's membership pitch centers on open development around the MLU hardware, including the toolkits, drivers, and model-pipeline integration work that makes a new accelerator usable from PyTorch. If MLU support lands in the framework's first-class paths, code written in standard PyTorch would run on Cambricon chips with less glue.
Alibaba Cloud's frame is up-stack. The cloud arm of Alibaba Group is one of the largest operators of Qwen, the LLM and multimodal model family developed by Alibaba. Membership gives Alibaba Cloud a formal voice in PyTorch governance at a moment when running large model training and inference efficiently, across multiple accelerator vendors, is a live technical problem across the Chinese cloud.
Ant Group, the fintech affiliate of Alibaba that operates Alipay, joined at Gold. Ant's focus is closer to the application layer: AI-agent software tools and a secure cloud-native runtime for deploying them. Gold members do not take board seats, but they participate in working groups and events.
The Chinese roster at the foundation is not new. Huawei was already a Platinum member. More than 250 organizations based in China already contribute code, reviews, or operational support to projects in the foundation's orbit, including PyTorch itself, DeepSpeed, Ray, Safetensors, Helion, and vLLM, according to the foundation. The Shanghai gathering is the formal point at which three more Chinese names appear on the membership list, with two of them now voting on the technical direction of the framework.
The practical question is whether Platinum membership translates into specific commitments on accelerator support. Foundation press materials and the project blog describe a membership and a gathering, not a roadmap, and there is no published Technical Advisory Council agenda tied to this announcement. The watch item is the next PyTorch release cycle: if MLU support, or broader Cambricon and Alibaba Cloud accelerator paths, shows up in first-class status in the next set of upstream changes, the Platinum seats did more than move logos around.