The PyTorch Foundation announced on September 8 that Alibaba Cloud and Cambricon have joined as Platinum members, with Ant Group joining at the Gold level. The announcement came from Shanghai, where KubeCon + CloudNativeCon + OpenInfra Summit + PyTorch Conference China 2026 is under way. Three companies that underpin much of China's AI infrastructure now hold formal seats at the table where the framework's direction is decided.
What a Platinum seat actually buys
The PyTorch Foundation operates under the Linux Foundation as a vendor-neutral home for open source AI. Alongside PyTorch itself, it hosts training and inference projects including vLLM, DeepSpeed, Ray, Helion and Safetensors.
Platinum membership gives Alibaba Cloud and Cambricon one seat each on the Governing Board and one seat each on the Technical Advisory Council. That is not a logo on a sponsor page; it is a vote on roadmaps and on which hardware gets first-class support. Both companies say they will work with the community to strengthen support for accelerators and hardware across the PyTorch ecosystem.
Ant Group, joining as a Gold member, is focused on production-grade platforms for AI workloads. The company has contributed to open source on both the cloud-native and AI framework sides for years.
The three companies occupy different layers
What makes this set of members interesting is that they barely overlap.
Alibaba Cloud is a full-stack cloud provider and the developer of Qwen, a family of large language and multimodal models. Since its debut in 2023, Qwen has grown into one of the most widely adopted open-weight model series anywhere. At the Shanghai keynote, the company plans to walk through how it serves Qwen at scale across multiple clusters using Karmada, in a talk delivered by Jionghang Cai, Senior Technical Expert on the AI Infrastructure team.
Cambricon designs AI chips and accelerators in China, with a Machine Learning Unit (MLU) ecosystem built around large-scale model training and inference. Its topic is hardening PyTorch's device-agnostic foundation: letting developers write the same code regardless of backend while still reaching each chip's native capabilities. With accelerators beyond NVIDIA now entering real production, that pair of goals has become difficult to postpone. Wei Li, Principal Software Engineer, is scheduled to present.
Ant Group takes the application layer. Its two linked talks cover assembling existing cloud-native building blocks such as Kubernetes Agent Sandbox and Kata Containers into a secure, on-demand runtime for AI agents. Once you let an agent execute arbitrary code, isolation becomes the practical pressure point, and reaching for proven virtualization technology is a reasonable answer.
Huawei, already inside
Huawei got there first, joining the PyTorch Foundation in 2023. The company helped start the Accelerator Integration Working Group and carried its Ascend NPU through to becoming the first additional platform in PyTorch. Fred Li, Head of Huawei's Computing Open Source Development Team, describes the goal as native development, native release and native distribution, so that developers get the same PyTorch experience whichever accelerator they pick up. In Shanghai, Liang Zhang, Vice President of the Ascend Computing Product Line, delivers the Huawei keynote.
More than 250 organizations across China contribute to PyTorch Foundation projects. The membership roster shows four companies, but the developer base behind it is considerably deeper.
Who gets to vote on the open foundation
Mark Collier, Executive Director of the PyTorch Foundation, said open source has become the default way the world builds AI, and that the new members strengthen the global ecosystem that developers rely on for training, optimizing, serving, distributing and running models. Feifei Li, Chief Technology Officer of Alibaba Cloud, pointed to years of running PyTorch at scale across heterogeneous hardware. Zhengyu He, Chief Technology Officer of Ant Group, framed the move around lowering the barriers to AI development and adoption. Elton Gong, Vice President of Software Engineering at Cambricon, spoke about delivering a native, out-of-the-box developer experience across a broader range of backends.
The pattern underneath these statements is that framework neutrality has stopped being a slogan and become an operational requirement. The more the compute market diversifies away from a single vendor, the less the layers above it can afford to lean on one. From the silicon side, a chip that does not run PyTorch cleanly never reaches the shortlist. Buying a board seat is what that shared interest looks like in practice.
The geopolitical backdrop is harder to set aside. As AI infrastructure splits along US-China lines, whether a common framework can keep working as a bridge will be tested through both membership and technical decisions. Huawei putting interoperability between Chinese and global AI stacks at the center of its keynote is a sign that the question remains open.
PyTorch Conference China 2026 runs September 7 to 9 in Shanghai. The North American edition is scheduled for October 20 and 21 in San Jose, with registration already open.
Summary
The PyTorch Foundation has added Alibaba Cloud and Cambricon as Platinum members and Ant Group as a Gold member. Platinum status carries one Governing Board seat and one Technical Advisory Council seat each, and together with Huawei, China's cloud, silicon and application layers are now represented in the framework's decision-making. How far the device-agnostic direction actually goes, and who drives it, is the question the Shanghai sessions start to answer.
Note: The thumbnail image is an AI-generated illustration.
