Moore Threads (摩尔线程), known as the Chinese version of Nvidia, has finished its mainland initial public offering (IPO) process in 88 days and is set to list. Counted as a leading high-performance graphics processing unit (GPU) player in China, Moore Threads was founded by a former global vice president at Nvidia with more than 10 years of experience there, has several experts from Nvidia, and has drawn attention with rapidly growing sales recently. However, there is still a technology gap before it can be compared with Nvidia, and it remains in the red.
On the 29th, according to China Business News, Caixin and others, Moore Threads applied for an IPO on Jun. 30 and passed the listing committee review on the 26th this month. It plans to list on the STAR Market, known as China's Nasdaq, and raise 8 billion yuan (about 1.5753 trillion won). The listing is underwritten by CSC Financial. The proceeds will be invested in ▲ next-generation artificial intelligence (AI) training and inference integrated chips ▲ next-generation graphics chips ▲ next-generation AI system-on-chip (SoC) development projects.
◇ In the U.S.-China conflict phase, left Nvidia and set out to localize GPUs
Moore Threads was founded in Beijing in June 2020. It researches, designs, and sells high-performance GPUs and related products. It is one of the few Chinese corporations that can perform graphics rendering and AI computation simultaneously. It aims for full-function products spanning AI, cloud, graphics, and supercomputing. To date, it has released four generations of its self-designed GPU architecture.
One reason the market is paying attention to Moore Threads' growth potential is that a significant portion of its workforce hails from Nvidia, the industry leader. Founder and CEO Zhang Jianzhong (张建中) served as a global vice president and China head at Nvidia from 2006 to 2020. As U.S.-China tensions intensified and the need to localize the GPU industry in China emerged, Zhang left the company and immediately founded a startup, leveraging roughly 14 years of experience and networks built at Nvidia.
Co-founder and Chief Technology Officer (CTO) Zhou Yuan (周苑) also previously oversaw the Nvidia ecosystem. Zhou worked at Nvidia for more than a decade, leading efforts to expand the GPU ecosystem, and now handles ecosystem and go-to-market strategy at Moore Threads. In addition, former GPU architects Zhang Yubo (张钰勃) and Yang Shangshan (杨上山), former sales director Wang Dong (王东), and former sales manager Song Xuejun (宋学军) have joined Moore Threads.
◇ Individual performance falls short of Nvidia … breakthrough with "clusters"
Although Moore Threads is a startup only five years after its founding, it has responded to market demand by releasing next-generation new products every year since 2021 through rapid research and development (R&D). It began with the first-generation GPU "Sudi (苏堤)" in 2021 and released the fourth-generation GPU "Pinghu (平湖)" last year. This chip supports the "FP8" precision needed to train large AI models. FP8 requires less computation and memory than FP32 and FP16, enabling faster processing. Nvidia's H100 GPU supports FP8 operations.
However, the company said in its prospectus, "There is still a gap in comprehensive R&D capabilities, accumulation of core technologies, and customer and ecosystem aspects compared to global leading corporations such as Nvidia."
Moore Threads' breakthrough is cluster technology. While the individual performance of its GPUs is not superior to Nvidia's, it achieves high performance by advancing technology that connects GPUs in large numbers, a strategy similar to Huawei's. After launching the "KUAE (夸娥) 1" cluster in Dec. 2023, it unveiled "KUAE (夸娥) 2" at the end of 2024, which can connect up to 10,000 GPUs. The product is already reported to have been delivered to multiple sites and is being used for large AI model training, inference, and computation.
Thanks to this strategy, revenue is also growing rapidly. According to China Business News, revenue in 2022–2024 was 46 million yuan (about 9.1 billion won) → 124 million yuan (about 24.4 billion won) → 438 million yuan (about 86.3 billion won), an average annual growth of 208%. First-half revenue this year was 702 million yuan (about 138.2 billion won), already exceeding the combined total of the previous three years. Notably, AI cluster product revenue (as of last year) accounted for 42% of total revenue.
◇ Not yet in the black … U.S. sanctions are also a risk
However, it remains in the red. Net losses from 2022 to the first half of 2025 were 1.84 billion yuan (about 362.4 billion won) → 1.673 billion yuan (about 329.5 billion won) → 1.492 billion yuan (about 293.9 billion won) → 271 million yuan (about 53.4 billion won), decreasing but likely to take more time to turn a profit. The company expects to break even in 2027, but Caixin projected that "excluding government subsidies, the result will in effect remain negligible."
Being added to the U.S. sanctions list in Oct. 2023, which restricts raw material procurement and other areas, is also a risk factor. Moore Threads has been restricted from using raw materials and semiconductor design tools that include U.S. technology, making it difficult to outsource production to Taiwan's TSMC, which in turn limits the use of advanced processes such as 5-nanometer (nm) and 7-nanometer.
In addition, it is not easy to procure high-bandwidth memory (HBM), mainly produced by companies under the influence of U.S. regulations, such as SK hynix, Samsung Electronics and Micron. HBM is a key component of high-performance GPUs. This is one of the main reasons Moore Threads cannot produce Nvidia-level GPUs. The company warned that "an increase in costs is inevitable during the process of supply chain transition," adding that if sanctions expand, it cannot rule out the possibility of a sharp drop in sales.