At a Shanghai conference that fused KubeCon, CloudNativeCon, OpenInfra Summit, and PyTorch Conference China 2026, Alibaba Cloud and Cambricon joined the PyTorch Foundation as Platinum members while Ant Group became a Gold member.

Keynote topics include building multi-cluster AI infrastructure, agent runtimes with open infrastructure, device-agnostic PyTorch, and collaboration between Ascend and PyTorch.

Representatives from Alibaba Cloud, Cambricon, Ant Group, and Huawei will lead keynote sessions, with Alibaba Cloud detailing Qwen deployment, Ant Group on AI agents, Cambricon on device-agnostic infrastructure, and Huawei on Ascend interoperability.

The strategic aim is to minimize code rewrites when switching hardware by promoting a unified PyTorch interface across devices and enabling broader multi-device support.

Cambricon is shifting from a downstream adapter to an active contributor and rule-maker in PyTorch, aiming to speed up optimization infrastructure, CI/CD, and cross-device support.

Cambricon will harden PyTorch’s device-agnostic foundation to broaden backend reach and enrich native capabilities for AI hardware.

The new memberships reinforce collaboration across hardware, models, and infrastructure to advance PyTorch and open-source AI in China.

These additions strengthen hardware and accelerator support within the PyTorch ecosystem and reinforce open-source AI infrastructure.

Cambricon adopts an Upstream First approach, submitting code to PyTorch’s main branch for community review to reduce migration costs across hardware.

The PyTorch Foundation, under the Linux Foundation, governs projects like PyTorch, vLLM, DeepSpeed, and Ray, with a Governing Board guiding priorities and main-branch integration.

Huawei will discuss device-agnostic PyTorch and Ascend interoperability across backends.

Cambricon has contributed to seven PyTorch modules and maintains Day 0 adaptation for models to run on Cambricon hardware from release day.