August 9, 2026 (InvestinChina.asia) – Moore Threads Intelligent Technology (Beijing) Co., Ltd. (688795.SH) — widely regarded as “China’s Nvidia alternative” and the country’s first publicly listed full-function GPU designer — reported on Sunday that first-half revenue jumped 147.42% year-on-year to 1.736 billion yuan, surpassing its entire 2025 full-year sales. On the same day, the company announced it is planning a Hong Kong IPO, setting the stage for an “A+H” dual-listing structure.
The Beijing-based company, founded in June 2020 by former Nvidia executive Zhang Jianzhong, designs and sells GPU chips capable of handling AI compute, graphics rendering, physical simulation, scientific computing and ultra-HD video codec on a single piece of silicon — a breadth of functionality that distinguishes it from most domestic AI accelerator peers that focus on narrow ASIC or DSA routes.
Half-Year Results: Hypergrowth, Sharper Loss Narrowing
For the six months ended June 30, Moore Threads delivered:
| Metric | H1 2026 | YoY Change |
|---|---|---|
| Revenue | 1.736 billion yuan | +147.42% |
| Gross profit | 989 million yuan | +103.78% |
| Gross margin | 56.95% | |
| Net loss attributable to shareholders | 11.56 million yuan | Narrowed by 259 million yuan (−95.73%) |
| Non-GAAP net loss | 150.8 million yuan | Narrowed by 52.37% |
| R&D investment | 769 million yuan | +38.16% (44.30% of revenue) |
The company noted that the revenue surge was driven by robust demand for full-function GPUs amid the AI boom, coupled with accelerated commercialization of its Kuae (夸娥) intelligent computing clusters and stable delivery to customers. It posted a net loss of just 11.56 million yuan, narrowing 95.73% from a year earlier, though it has yet to turn a full-half profit.
Cumulative R&D spending since 2022 now stands at nearly 5.9 billion yuan. As of end-June, the company had filed 2,167 patent applications and held 788 authorized patents — ranking among the top echelons of domestic GPU enterprises.
The Technology: MUSA Architecture and the MTT S5000 Flagship
Moore Threads’ core technical moat is its proprietary MUSA (Moore Threads Unified System Architecture) — a unified programming model that supports compute, graphics, multimedia and inter-chip communication engines on a single scalable architecture. The company has iterated through five chip generations in roughly five years, culminating in its fourth-generation “Pinghu” (平湖) architecture.
The flagship product, the MTT S5000 AI training-and-inference card, is built on the Pinghu architecture and the PH100 chip, which has passed the China Information Security Research Center’s safety and reliability evaluation (2026 No. 2):
- FP8-to-FP64 full precision — one of the few domestic GPU vendors to natively support the full precision range from FP8 to FP64
- Up to 1 PFLOPS dense AI compute (FP8), with 80 GB memory and 1.6 TB/s memory bandwidth
- 784 GB/s inter-card interconnect via second-generation MTLink, enabling linear scaling from single card to 10,000-card clusters
- 95% cluster training linearity achieved in ultra-large-scale AI model training in H1 2026
- CUDA-compatible migration — MUSA’s near-complete CUDA syntax compatibility and the proprietary MUSIFY porting tool allow developers to migrate existing code at near-zero cost
Benchmarked against international flagships, Moore Threads claims its FP8 GEMM utilization exceeds 90% — 15 percentage points above international flagship products — while Flash Attention efficiency reaches over 95%, 15–20 points higher than comparable overseas offerings.
In the first half, Moore Threads completed deep adaptation of DeepSeek-V4 on the MTT S5000 using native FP8 low-precision compute, and rapidly adapted a new batch of SOTA large models including MiniMax M3 and Zhipu GLM-5.2. The MTT S5000-powered Kuae clusters have achieved large-scale sales, with deployments and deliveries completed in Beijing, Wuxi, Hangzhou and other cities.
Full-Stack Product Matrix: Cloud, Edge and Terminal
Beyond the data-center S5000, Moore Threads has built a diversified product portfolio:
- Cloud AI compute — MTT S4000/S5000 cards and Kuae thousand-card/ten-thousand-card clusters for large-model training and inference
- Graphics acceleration — consumer gaming GPUs (MTT S80, S70, S50) and professional cards (MTT S3000, S2000) for PC, workstation, cloud gaming and cloud desktop scenarios
- Intelligent SoCs — the “Yangtze River” (长江) series SoC integrating GPU, CPU and NPU, targeting AI PCs, edge computing and intelligent terminals; the MTTAICUBE powered by Yangtze SoC launched on JD.com in H1
- Cloud PC — its full-function GPU and pioneering dynamic elastic vGPU slicing technology are penetrating telecom operators; multi-provincial commercial deployments with China Mobile are underway
- Software ecosystem — the MUSA software stack now supports over 800,000 developers
Customers span government agencies, State Grid, China Mobile, China Unicom, Shanghai Pudong Development Bank, Postal Savings Bank, Baidu, JD.com, 360 and ByteDance, among others.
Planning the Hong Kong IPO: An “A+H” Hard-Tech Play
On August 7, the board approved a proposal to issue H-shares and list on the Hong Kong Stock Exchange’s main board. The company said the move aims to deepen its international strategy, attract R&D and management talent, and improve corporate governance and core competitiveness.
Moore Threads listed on Shanghai’s STAR Market on December 5, 2025, at an issue price of 114.28 yuan per share, raising 8 billion yuan and becoming the “first share of domestic GPUs.” As of August 7, its A-share price stood at 597.89 yuan, valuing the company at 281.0 billion yuan.
The H-share issuance remains subject to shareholder approval and regulatory filing, and specific details including timing, size and pricing have not yet been determined. The company intends to complete the offering within 24 months of shareholder approval, or such extended period as shareholders may agree.
Notably, its H1 operating cash flow remained negative at −2.169 billion yuan, reflecting continued heavy upfront investment in production scale-up and inventory — a reminder that the path to sustainable profitability remains a work in progress despite the dramatic revenue acceleration.
If successfully executed, Moore Threads would join a growing roster of Chinese hard-tech “A+H” dual-listed companies, following peers such as Biren, MetaX and Enflame that together form China’s “four little dragons” of the domestic GPU sector. As the domestic GPU and AI chip industries transition “from an investment phase into a commercialization and profit-delivery phase,” in the words of Shenzhen-based First Seafront Fund chief economist Yang Delon, Moore Threads’ Hong Kong listing would provide a critical offshore capital channel to fund the next leg of its full-function GPU roadmap — including the next-generation “Huagang” architecture and 100,000-card cluster development already underway.