Moore Threads CEO Announces Launch of Next-Generation GPU Architecture and Full-Stack Strategy

Published:December 9, 2025

Reading Time:1 min read

Moore Threads is preparing to reveal its next-generation GPU architecture and a comprehensive MUSA-based full-stack roadmap as it advances its software and hardware ecosystem.

On December 9, Moore Threads founder, chairman and CEO Zhang Jianzhong announced that he will outline the company’s full-stack development strategy centered on MUSA, along with its long-term vision, at the first MUSA Developer Conference. The event will also feature the debut of a next-generation GPU architecture, a comprehensive product roadmap covering hardware, core technologies, and industry solutions, as well as updates on ecosystem building and multi-industry deployments.

MUSA (Meta-computing Unified System Architecture) is Moore Threads’ unified parallel-computing platform. The architecture enables GPUs to tackle complex computing tasks and consists of the MUSA instruction set architecture (ISA) together with the GPU’s internal parallel-computing engines.

Recently, the company also released Torch-MUSA v2.7.0, the MUSA extension library for the PyTorch deep-learning framework. The new version introduces advances in feature integration, performance optimization, and hardware support. Over the past month, Torch-MUSA has delivered two rapid updates—v2.5.0 and v2.7.0.

According to the company, the latest release further integrates accelerated computing libraries such as muSolver and muFFT, significantly improving execution efficiency for complex workloads. It also adds Unified Memory Management (UMM) support for unified-memory devices, improving memory-usage efficiency. As of November, Torch-MUSA supports more than 1,050 operators.