The PyTorch Foundation on September 8 said Alibaba Cloud and Cambricon have joined as Platinum members and Ant Group as a Gold member, marking the biggest expansion of Chinese AI heavyweights inside the Linux Foundation-hosted body since Huawei signed on in 2023. The announcement landed on the keynote stage of the first-ever KubeCon + CloudNativeCon + OpenInfra Summit + PyTorch Conference China in Shanghai.
Platinum Seats Change The Board
Platinum membership hands Alibaba Cloud and Cambricon one seat each on the Foundation's Governing Board and a matching seat on its Technical Advisory Council, giving them a direct hand in the framework's release cadence, project onboarding and backend-support roadmap. Ant Group's Gold seat routes it into the same working-group structure without a board seat.
Chips, Clouds, Agents
Alibaba Cloud CTO Feifei Li said in a supporting quote that the company has been running PyTorch at scale across heterogeneous hardware for years and now wants to formalize that work upstream; the team's opening keynote details how it serves the Qwen family of open-weight models on multi-cluster infrastructure via Karmada. Cambricon Vice President of Software Engineering Elton Gong is walking attendees through a device-agnostic PyTorch layer that will let its MLU accelerators — like Huawei's Ascend line — plug in as a first-class backend rather than a fork. Ant Group's slot spans two keynotes on assembling Kubernetes Agent Sandbox and Kata Containers into a secure runtime for AI agents.
Huawei Frames The Interoperability Bet
Huawei, which anchored the Foundation's first big Chinese endorsement in 2023, is using its keynote to argue for hardware-software co-design across Chinese and global AI stacks — a real interoperability question now that Ascend NPU has become the first non-CUDA platform natively supported inside PyTorch. Fred Li, head of Huawei's Computing Open Source Development Team, said Ascend's push toward native development, release and distribution "gets easier the more of us are pursuing it." Cambricon is signalling it plans to run the same play for the MLU family.
Why It Matters
The Foundation now has three Chinese Platinum-tier tenants pulling upstream at exactly the moment Beijing is codifying its 9,800-eflops national compute build-out around home-grown accelerators. That gives Chinese AI companies a formal channel into a framework that already powers the bulk of open-weight model training, and it hands global developers a cleaner runway to target Ascend, MLU and other domestic silicon without maintaining forks. More than 250 organizations across China now contribute to PyTorch Foundation projects including DeepSpeed, Helion, Ray, Safetensors and vLLM — a base the September 8 memberships are designed to formalize.
Reporting based on the PyTorch Foundation's September 7 announcement.