"The Chinese version of NVIDIA" is going public

Author | Liu Baodan

Editor | Huang Yu

In this wave of the AI industry, the shovel sellers, NVIDIA, are undoubtedly the biggest winners in the capital market. Now, domestic GPU unicorns have also begun their journey to go public.

On November 13, the official website of the China Securities Regulatory Commission showed that the domestic full-function GPU unicorn, Moore Threads, has completed the counseling registration with the Beijing Securities Regulatory Bureau, officially starting the A-share listing process. The counseling institution is CITIC Securities Co., Ltd.

Following Biren Technology and Suiruan Technology, Moore Threads has become the third domestic AI chip unicorn to complete the registration, which means that GPU companies benchmarking NVIDIA are gradually knocking on the door of the capital market.

Moore Threads was established in October 2020, and its founding team comes from the global GPU giant NVIDIA. Its founder and CEO, Zhang Jianzhong, previously served as NVIDIA's global vice president and general manager for China, having been deeply involved in the GPU industry for nearly twenty years. The listing counseling registration report shows that Zhang Jianzhong controls 44.07% of the company's shares.

The company focuses on full-function GPUs, providing accelerated computing infrastructure and one-stop solutions. Technically, Moore Threads is one of the few domestic companies that can functionally benchmark NVIDIA's GPUs.

Currently, Moore Threads is one of the only domestic GPU companies with layouts in both the B-end and C-end markets. Its chips adopt the advanced MUSA architecture, integrating four major engines: AI computing acceleration, graphics rendering, video encoding and decoding, physical simulation, and scientific computing.

The company's core product, the KUAE intelligent computing center full-stack solution, includes infrastructure centered around the KUAE computing cluster, a cluster management platform (KUAE Platform), and a large model service platform (KUAE ModelStudio), aiming to solve the construction and operation management issues of large-scale GPU computing power through integrated delivery.

According to industry insiders close to Moore Threads, with the change in computing paradigms, GPUs that integrate graphics and AI computing have become mainstream general-purpose computing power, widely empowering cutting-edge industries such as artificial intelligence, autonomous driving, digital twins, and embodied intelligence. "For large model training, Moore Threads has implemented a fully domestically produced kilowatt intelligent computing cluster and launched its own ten-thousand-watt cluster solution, currently ranking in the first tier of domestic AI chips."

Wall Street Journal has learned that the company has completed adaptation and testing with several large model companies, including Zhiyuan Research Institute, Zhipu AI, and Wuwen Xinqiong, providing computing power support for large models with hundreds of billions and trillions of parameters.

In addition, Moore Threads launched the domestically produced gaming graphics card MTT S80 in 2022, hailed as the "first domestic gaming card," and is also the only consumer-grade graphics card in the country that supports DirectX 12. Public information shows that Moore Threads' clients include leading operators, large state-owned banks, and large model startups.

Beyond products, around its self-developed MUSA unified system architecture and software platform, Moore Threads is committed to building a domestic GPU application ecosystem. The software ecosystem is key to promoting the popularization of GPU computing. Centered on the MUSA architecture, Moore Threads has released a complete MUSA software stack to serve developers and end users Moore Threads' MUSA architecture and software stack are compatible with CUDA. Through the MUSIFY code automatic conversion tool, users can quickly and seamlessly migrate their existing CUDA code to the MUSA platform. In addition, Moore Threads has launched a series of software for open-source project MUSA porting based on domestically produced full-featured GPUs, including the deep learning framework Torch-MUSA.

In recent years, with the global explosion of AI large models, Moore Threads' valuation has also begun to rise rapidly.

Currently, Moore Threads has completed six rounds of investment, with investors including China Mobile, Shenzhen Capital Group, Shanghai Guosheng, Bank of China International, Jianyin International, China Merchants Venture Capital, Zhongguancun Science City, Sequoia Capital, and other well-known state-owned and venture capital institutions. The last round of financing occurred in November 2023, where Moore Threads completed financing of several hundred million yuan, with a post-investment valuation of approximately 30 billion yuan.

From Hurun's annual global unicorn list, we can also see the changes in Moore Threads' valuation.

In 2022, two years after its establishment, Moore Threads made its first appearance on the list with a valuation of 13 billion yuan, ranking 568th. In April of this year, Hurun released the latest global unicorn list, showing Moore Threads' valuation at 25.5 billion yuan, ranking 261st.

The revolutionary opportunities brought by AI will have a significant impact on social and economic development. As a participant in AI infrastructure, Moore Threads has caught the rhythm of the times and is ready to make significant strides.

However, the challenges are also evident. To narrow the gap with NVIDIA, Moore Threads still has a long way to go