Alibaba Cloud, Ant Group and Cambricon Join PyTorch Foundation to Expand the Open AI Stack

Introduction

The PyTorch Foundation announced at PyTorch Conference China 2026 in Shanghai that Alibaba Cloud and Cambricon had joined as Platinum members and Ant Group as a Gold member. Huawei, a long-standing supporter and contributor, is also taking part in the event. Together, the companies are presenting work across model serving, AI accelerators, PyTorch backend integration, cloud-native infrastructure and agent runtimes. The agenda offers a view of how open-source AI is expanding from individual models into a full technology stack.

Key points

• Deeper governance participation: Platinum members receive one seat on the PyTorch Foundation Governing Board and one on the Technical Advisory Council. Alibaba Cloud and Cambricon are expected to work with the community on support for hardware and accelerators, while Ant Group will focus on production-oriented AI platforms and collaboration.
• Alibaba Cloud on large-scale serving: Its session covers serving Qwen across multiple clusters with Karmada-based AI infrastructure. The company is also positioning its experience running PyTorch across heterogeneous hardware as an input to the wider community.
• Cambricon on multi-backend PyTorch: Cambricon will discuss its MLU software ecosystem, including toolkits, drivers and machine-learning pipelines, as well as efforts to give developers a more consistent PyTorch experience across backends without losing native capabilities.
• Ant Group on agent execution: Ant Group’s talks connect cloud-native components such as Kubernetes Agent Sandbox and Kata Containers with AI agents, focusing on a secure and on-demand runtime model.
• Huawei on hardware-software co-design: Huawei will present work around Ascend and PyTorch, with an emphasis on upstream integration and more natural development, release and distribution experiences for its platform.

Why it matters

The significance of the announcement lies in the range of participants as much as in the membership changes. Alibaba Cloud brings cloud services and Qwen infrastructure; Cambricon contributes AI chips and accelerators; Ant Group has experience in cloud-native and AI open-source projects; and Huawei continues to develop the Ascend ecosystem. Their combined agenda addresses how chips, frameworks, models and deployment environments can work together.

For developers, backend compatibility and tooling determine whether a model can move from experimentation into production. The foundation says more than 250 organizations in China contribute to projects including PyTorch, DeepSpeed, Ray, Safetensors, vLLM and Helion. Broader participation from domestic hardware and infrastructure companies could add more device support and deployment knowledge to the ecosystem. The practical impact, however, will depend on sustained upstream contributions, interface standardization and real interoperability across platforms.

The announcement also reflects a shift in the AI infrastructure discussion. As open-weight models become more widely used, the layers around them—accelerators, frameworks, orchestration and secure runtimes—become central to both competition and collaboration. The Shanghai event’s progression from silicon to agent deployment illustrates how Chinese AI companies are using open-source communities to connect the stack.

Source: PyTorch Blog