Alibaba's Pingtouge Open-Sources T-Head SAIL AI Software Stack with 260+ Framework Support

At the World Artificial Intelligence Conference (WAIC) 2026, Alibaba's chip subsidiary Pingtouge announced the open-source release of its proprietary AI software stack, T-Head SAIL. This is the foundational software layer for the company's Zhenwu series AI chips, featuring independent intellectual property and full compatibility with mainstream AI ecosystems, covering the complete chain from the OS layer through SDK to interface layer. Source code, drivers, toolchains, and documentation are now available for download via the Pingtouge developer community.

Before the open-source release, SAIL had already accumulated a significant track record in production. As of April 2026, Zhenwu AI chips have shipped over 560,000 units, serving more than 400 enterprise customers across 20 industries. The software stack has been battle-tested in Alibaba Cloud's production environment, handling massive traffic surges comparable to Singles' Day scale.

By open-sourcing SAIL, Pingtouge has made its core technology accessible to every research team and individual developer. Users can now read the source code to understand chip operation logic, debug and fix issues themselves, and perform deep custom optimizations for specific business scenarios. The open-source release covers a complete five-layer technology stack: Driver & Runtime at the bottom, followed by a programming language layer (C/C++ and Python), a compiler layer (Compiler and Debugger), high-performance libraries (acBLAS, acDNN, acFFT, and other math/AI libraries), and the top-layer developer tools (Asight Compute and Asight Systems).

For most developers, migration cost matters more than feature completeness. Pingtouge's answer comes as three commitments: no code rewrite needed — SAIL is compatible with mainstream AI ecosystems; no waiting for adaptation — average adaptation time for mainstream AI inference frameworks is under 7 days; and no framework lock-in — SAIL supports over 260 mainstream training and inference frameworks including PyTorch, TensorFlow, vLLM, and SGLang.

SAIL's open-source marks a natural milestone in Pingtouge's eight-year development journey. Starting with the Hanguang 800 inference chip in 2019, followed by the Yitian 710 general-purpose CPU and the Zhenwu 810 AI chip, and this year's Zhenwu M890 training-inference integrated chip, Pingtouge now has self-developed capabilities across all three dimensions of computing, networking, and storage. SAIL stands for Seed of AI Library, and the company views every line of open-source code as a seed with unlimited potential for China's AI ecosystem.