A-share market welcomes another newcomer, this time it's Moore Threads, which created a 'unicorn' in just 100 days.

On November 13, the official website of the China Securities Regulatory Commission showed that domestic GPU unicorn Moore Threads Intelligent Technology (Beijing) Co., Ltd. (hereinafter referred to as "Moore Threads") had registered for coaching and counseling with the Beijing Securities Regulatory Bureau, officially initiating the A-share listing process. The coaching institution is CITIC Securities Co., Ltd.

The listing coaching and counseling report shows that Moore Threads was established in June 2020, and its actual controller is Zhang Jianzhong, who controls 44.07% of the company's shares.

Zhang Jianzhong, founder and CEO of Moore Threads, was once the Global Vice President of NVIDIA and General Manager of NVIDIA China. He has been deeply involved in the GPU industry for nearly two decades and enjoys a high reputation.

According to the official website, Moore Threads, with its full-featured GPU at its core, is committed to providing the global market with accelerated computing infrastructure and one-stop solutions, providing AI computing support for the digital and intelligent transformation of various industries. It is also the only domestic GPU enterprise with layouts in both the B-end and C-end markets.

Before initiating the listing process, Moore Threads had completed several rounds of financing, raising a total of billions of yuan. Not long ago, Moore Threads also just completed its shareholding system reform, increasing the company's registered capital from 24.413 million yuan to 330 million yuan.

In addition to Moore Threads, domestic AI chip unicorns such as BiRen Technology and Suiyuan Technology have also recently submitted A-share listing applications, indicating that the domestic GPU industry is entering a period of rapid development and reshuffling.

Regarding the choice of A-shares, He Xiongsong, Executive General Manager of Shanghai Chentao Asset Management Co., Ltd., believes that currently, Chinese concept stocks face obstacles in listing on U.S. stock exchanges, and the liquidity of Hong Kong stocks is poor. The A-share market, on the other hand, tends to support such enterprises.

An investor also told Cyber Auto that for companies dedicated to the AI computing infrastructure industry, the A-share market is currently the best destination in terms of sensitivity, valuation, and support.

Whether for short-term capital needs or long-term development planning, domestic 'NVIDIA' needs to accelerate its listing efforts.

01 NVIDIA executive restarts his career, highly favored by the capital market

Like many AI chip companies, Moore Threads' founding team boasts impressive resumes, with special attention focused on founder Zhang Jianzhong.

Public information shows that Zhang Jianzhong graduated from the Department of Computer Science at Nanjing University of Science and Technology and later obtained a master's degree from the Automation Research Institute of the Ministry of Metallurgical Industry.

Zhang Jianzhong, founder and CEO of Moore Threads

He has held positions such as General Manager of the Computer Systems Business Unit and General Manager of the Government and Education Business Unit at Hewlett-Packard and Dell.

In May 2005, Zhang Jianzhong joined NVIDIA as Global Vice President and General Manager of NVIDIA China. Under his leadership, NVIDIA GPUs successfully expanded their ecosystem in China, making the Chinese market one of the most important global markets for NVIDIA.

Data shows that in 2008, NVIDIA's GPU market share in China was less than 50%. By 2020, when Zhang Jianzhong left, the market share had exceeded 80%.

Zhang Jianzhong was once known as the 'right-hand man' of NVIDIA founder Huang Renxun.

In addition to Zhang Jianzhong, other members of the Moore Threads team mainly come from chip giants such as Microsoft, Intel, AMD, and Arm.

Moore Threads has stated that it will be the only truly world-class and mature team in China capable of covering the entire GPU process, with team members familiar with all aspects of GPU chip design, production, packaging, testing, systems, and software application quality management.

Additionally, there are reports that Moore Threads' R&D director comes from Horizon Robotics.

Thanks to its strong founding team, Moore Threads was highly favored by the capital market from its inception.

According to official statements, Moore Threads was officially established in October 2020.

In December 2020 and February 2021, Moore Threads, which had been established for less than 100 days, received two consecutive rounds of financing totaling billions of yuan, catapulting it to unicorn status in the GPU industry.

This was far from the end, as Moore Threads subsequently completed four more rounds of financing.

With multiple funding rounds and large amounts raised, the investor lineup behind Moore Threads is also impressive.

The angel round included Sequoia Capital. The Pre-A round was led by China Capital, Sequoia Capital, and GGV Capital, with the participation of ByteDance. The A round attracted investment from Tencent, followed by bets from more than 20 luxury VC teams, including China Mobile Capital.

According to official data, by December 2022, when Moore Threads completed its Series B funding of 1.5 billion yuan, its valuation had reached 24 billion yuan. On April 9, 2024, Moore Threads ranked 261st on the "2024 Hurun Global Unicorn Index" with a valuation of 25.5 billion yuan.

The 'China Speed' was on display.

02 Full-line layout from software to hardware, from B-end to C-end

While continuously receiving funding, Moore Threads has not slackened its R&D efforts.

Moore Threads is one of the few companies that have announced plans to manufacture 'full-featured GPU' chips. A full-featured GPU can perform computations in areas such as graphics rendering, video codec, AI applications, and scientific computing.

According to official descriptions, Moore Threads aims to become a leading GPU enterprise with international competitiveness, building an advanced accelerated computing platform for the integrated AI and digital twin intelligent world.

In October 2021, while announcing the completion of its Series A funding of 2 billion yuan, Moore Threads also brought good news on the product front: it had successfully developed the first domestically produced full-featured GPU in just 300 days.

Moore Threads MUSA Architecture

Four months later, on March 30, 2022, Moore Threads officially launched the Moore Threads Unified System Architecture (MUSA), which is primarily targeted at computing, graphics, multimedia, and AI product lines. It includes a unified programming model, software runtime library, driver framework, instruction set architecture, and chip architecture.

Moore Threads claims that this architecture enhances application portability, allowing them to run simultaneously on computing platforms such as the cloud and edge, aligning with the design intention of reducing repetitive labor for software developers and unleashing the core capabilities of different engines.

After all, developing GPUs and graphics cards is extremely difficult, and software development and ecosystem promotion are even more challenging. Especially since the global GPU industry is almost monopolized by NVIDIA and its CUDA, with AMD and Intel struggling to shake their position, let alone domestic manufacturers, which are almost completely absent in this area.

MUSA has a direct comparison to CUDA, as it includes a unified programming model, software runtime library, driver framework, instruction set architecture, and chip architecture, providing a complete solution from the hardware level to software development.

Along with the launch of MUSA, Moore Threads simultaneously released and began mass production of two full-featured GPU chips, "Su Di" and "Chun Xiao".

Subsequently, Moore Threads entered a period of rapid development.

On the hardware side, products include the MTT S4000, MTT S3000, and MTT S2000 for metaverse computing; the MTT S80, MTT S70, and Zhiyu Mofang for entertainment and creation; the MTT X300 and MTT S50 for professional visual applications; and the desktop graphics cards MTT S30/S10 for digital office use.

Moore Threads hardware products

On the software side, there is the first metaverse computing platform MTVERSE, self-developed GPU physics engine AlphaCore, digital human solution DIGITALME, and AIGC content generation platform Mobimajiang.

In terms of AI large models, Moore Threads has launched the hardware and software integrated Kuae Intelligent Computing Cluster, significantly expanding from thousands to tens of thousands of cards to build advanced computing infrastructure for large models and general AI.

It can be said that Moore Threads has achieved a full-line layout, from software to hardware and from B-end to C-end.

At the same time, Moore Threads has also excelled in ecosystem development.

Currently, the main strategy for domestic GPUs is to first be compatible with the NVIDIA CUDA ecosystem to minimize user migration costs. Moore Threads' GPU chips are not only compatible with NVIDIA CUDA but also support almost all current open-source large models. Zhang Yubo, CTO of Moore Threads, once stated in an interview that "developers can migrate to the Kuae cluster with almost no code modifications, with migration costs approaching zero and the migration process completable within hours."

Soon, Moore Threads ascended to the first tier of domestic AI chips.

03 Significant gaps remain with NVIDIA in terms of both products and channels

Funding and product development are progressing, and Moore Threads is even referred to as the 'domestic NVIDIA,' but the gap between it and NVIDIA is still significant.

As the inventor of the GPU, NVIDIA's position in this field is undeniable.

From a technical perspective, compared to other major global competitors, NVIDIA leads in terms of product completeness and existing market share, and this leading position is likely to be maintained for a long time. From the perspective of software ecosystem development, NVIDIA's CUDA (Compute Unified Device Architecture) ecosystem has high barriers, and user migration requires high costs.

With these double "buffs," NVIDIA GPUs have long dominated the fields of AI training and high-performance computing. According to data from IoT Analytics, a well-known IoT research institution, the global data center market size was approximately $49 billion in 2023, an increase of 182% year-on-year, of which NVIDIA accounted for about 92%, and AMD accounted for about 3%.

Although there have been breakthroughs in China, there is still a significant gap in product performance parameters and user experience for Moore Threads.

For example, the FP32 computing power of the MTT S3000 is 15.2 TFLOPS, while that of the NVIDIA A100 is 19.5 TFLOPS, equivalent to 80% of the A100's performance; the paper performance of the desktop-grade MTT S80 graphics card is comparable to that of the NVIDIA RTX 3060.

Theoretically, Moore Threads should perform at NVIDIA's mid-range level in both computing cards and desktop-grade products, but actual performance may differ.

Taking the MTT S80 as an example, according to user tests, the usage experience is significantly different from the RTX 3060, and it has even been dubbed a "developmental" graphics card.

Not only is the product power inferior to NVIDIA, but the same is true for channels. Currently, Moore Threads' online channel still only includes JD.com, with a cumulative order volume of about 2,000 units for the MTT S80 graphics card on the JD.com flagship store. For offline channels, although core distributors such as Kuangfan Technology and Wangxin Turing have been developed, most business is focused on the B-end market.

Moore Threads MTT S80 graphics card

However, as a company that has only been established for four years, Moore Threads' overall performance has been good, but there is still a long way to go. As Zhang Jianzhong stated at the 2023 summer conference, "Surviving for at least ten years is Moore Threads' primary short-term goal."

To strive for long-term success and maintain advancement requires substantial capital.

After all, chip research and development is an extremely time-consuming and costly endeavor.

The IP core in the GPU industry occupies more than 80% of the area. However, IP development is not easy; GPU IP research and development require 36 to 48 months and 200 engineers. While purchasing IP can shorten the development cycle by 12-18 months, the frontend and subsequent design of high-end chips take 1-3 years, and tape-out takes 3-6 months. If tape-out fails, the process must be repeated. Even if tape-out is successful, it takes 3-12 months of product testing and optimization before mass production can begin.

At the same time, industry insiders have calculated that a single tape-out of a 14nm process chip costs approximately $3 million, or approximately 21.5 million yuan; for a 7nm process chip, it costs $30 million; and for a 5nm process chip, it costs an astonishing $47.25 million.

This is also the reason why AI chip companies are accelerating their listing efforts: a lack of funds.

Cambricon, an AI chip company, only recorded revenue of 64.7653 million yuan in the first half of this year

Regarding the choice of A-shares, it is because for domestic AI chips, the A-share market is considered a relatively ideal channel. It provides support for high-tech industries, especially those with technological breakthroughs.

However, He Xiongsong also noted that it is difficult to say how much support the A-share market can provide to these companies, as many are not yet profitable and have unstable performance. But the support is definitely more than under normal circumstances.

But obviously, in the face of many pressures such as formidable rivals and external sanctions, being able to go public and obtain long-term capital support is something that AI chip companies all need.

END-