摩尔线程 (688795:SSE) written Q&A, September 3, 2026 transcript

Transcript, summary, guidance, and what analysts pressed on from the 摩尔线程 call on September 3, 2026.

FreeSee 摩尔线程's financials in GloomOpen in browser →Jump to the transcript
Held
September 3, 2026
Tone
Positive0.48
Words
6,252
R&D investment, H1 2026
769 million yuanup 38.16% year on year
MUSA developers
more than 800000
Cloud revenue share, H1 2026
about 97.5%
AIBOOK AI computing power
50 TOPS
Lushan GPU launch
within this yearsubject to official release
MTT C256 interconnect
256 cardssub-microsecond card-to-card delay

Summary

Management framed the half as a shift from technical validation toward scaled commercial delivery, while emphasizing ecosystem breadth and deferring profitability and supply details.

  • 97.5%Cloud products accounted for about 97.5% of revenue, concentrating growth in intelligent-computing clusters.
  • Management tied architecture investment to reusable designs, shorter delivery cycles and lower future marginal costs.
  • Break-even timing remained unquantified, with management citing macro conditions, competition and business execution.

Guidance

  • GPU products related to Lushan Mountain expected launch within this year

What analysts pressed on

  • 01Profitability timing: management said fundamentals are improving but declined to specify a break-even date.
  • 02Gross-margin recovery: management attributed the decline to raw-material prices and delivery mix, without giving a repair path.
  • 03Foundry capacity: management said supply matches business needs but declined to quantify locked capacity or growth.
  • 04Customer wins: management confirmed new internet-head and telecom customers but deferred specific business progress to disclosures.
  • 05Share-based compensation: management said grant timing, coverage and assessment may be optimized, without quantifying future expense.

Notable disclosures

  • More than 800,000 developers were reported in the MUSA ecosystem, alongside SGLang mainline integration.
  • MTT S5000 has achieved Day-0 adaptation for MiniMax H3, Kimi-K3 and GLM-5.3-Flash, according to management.
  • Management described production-level deployment of its PD-separated inference solution but gave no customer names or deployment scale.
  • The Lushan-related GPU launch was reiterated as expected within this year, subject to official release information.

Risks raised

  • Second-quarter gross margin was 49.26%, and management cited rising semiconductor input prices and a changing delivery mix.
  • Management gave no break-even timetable, citing macro conditions, market competition and business execution.
  • Management declined to quantify wafer capacity secured for the second half, offering only that supply could match business needs.
  • Cloud products represented about 97.5% of revenue, leaving current growth concentrated in one product category.

Tone: Positive 0.48

Management highlighted accelerating commercialization, customer adoption and ecosystem progress, but avoided firm profitability, margin-recovery and capacity commitments.

Who spoke

Company

  • Zhang JianzhongFounder, Chairman, General Manager, Moore Threads Technology Co. Ltd.
  • Zhang YuboCo-founder, Director, Deputy General Manager, Moore Threads Technology Co. Ltd.
  • Xue YansongSecretary of the Board of Directors and Chief Financial Officer, Moore Threads Technology Co. Ltd.
  • Zhou yuanCo-founder and employee director, Moore Threads Technology Co. Ltd.

Full transcript

Analyst questions are marked with a bar. Timestamps are into the recording.

Prepared remarks

ModeratorSSE Roadshow Center

Ladies and gentlemen, investors and netizens, hello! Welcome to the Moore Thread 2026 Semi-Annual Performance Presentation. This activity is sponsored by Moore Thread Intelligent Technology (Beijing) Co., Ltd. and supported by the network platform provided by Shanghai Stock Exchange Information Network Co., Ltd. Thank you for your attention and participation. On August 10, Moore Thread Intelligence Technology (Beijing) Co., Ltd. published the Semi-annual Report for 2026. Investors can log on to the website of Shanghai Stock Exchange (https://www.sse.com.cn/) or this platform (https://roadshow.sseinfo.com/).

View related announcements. In order to facilitate investors to have a more comprehensive and in-depth understanding of the company's operating results and financial situation in the first half of 2026, the company decided to hold a presentation of Moore's performance in the first half of 2026. In today's event, the main members of the management of Moore Thread visited the Shanghai Roadshow Center and will communicate with investors. Everyone is welcome to participate actively. First of all, let's welcome guests from listed companies to make opening speeches.

Zhang JianzhongFounder, Chairman, General Manager, Moore Threads Technology Co. Ltd.

Dear investors, good afternoon. Welcome to Moore Thread Intelligent Technology (Beijing) Co., Ltd. 2026 Semi-annual Performance Presentation. On behalf of the company, I would like to express my heartfelt thanks to all the investors who have long been concerned about and supported Moore Thread! At the same time, I would like to thank the Shanghai Stock Exchange and the Shanghai Roadshow Center for setting up such a platform for us to communicate with investors and enhance their understanding of the company.

The company has disclosed its 2026 semi-annual report on the website of the Shanghai Stock Exchange and the statutory disclosure media. You are welcome to raise your concerns, thank you!

Questions and answers

Investor

Hello, how is the research and development of the company's next generation products? When is it expected to be available?

Zhang YuboCo-founder, Director, Deputy General Manager, Moore Threads Technology Co. Ltd.

Dear investors, hello! In the first half of 2026, the company's R & D focus includes product series based on the new generation of "Huagang" architecture. GPU products related to Lushan Mountain are expected to be launched within this year, subject to the company's official product release information, thank you for your attention!

Investor

Revenue in the first half of the year increased by 147.42% over the same period last year. What is the main driving force?

Xue YansongSecretary of the Board of Directors and Chief Financial Officer, Moore Threads Technology Co. Ltd.

Dear investors, hello! The company's revenue in the first half of 2026 was 1.736 billion yuan, a substantial increase of 147.42% over the same period last year, and has exceeded the annual revenue in 2025, which is in the leading level in the same industry. Among them, the revenue of cloud products was 1.693 billion yuan, accounting for about 97.5% of the total revenue, which is the core source of the current revenue growth.

Mainly benefiting from the rapid development of artificial intelligence industry, the market demand for domestic full-function GPU has increased significantly, and the company's MTT S5000 and Kua'e Intelligent Computing Cluster have been recognized by customers, and achieved stable supply, and the commercialization has continued to accelerate. Thank you for your interest!

Investor

Is the company's customer base expanding smoothly this year, and is it introduced into Internet manufacturers?

Zhang JianzhongFounder, Chairman, General Manager, Moore Threads Technology Co. Ltd.

Dear investors, hello! In the first half year of 2026, the company achieved operating income of 1.736 billion yuan, a substantial increase of 147.42% over the same period last year, and exceeded the annual revenue of 2025, which shows that the company's core competitiveness of products and customer recognition are being translated into actual operating results. The Company's products are competitive in the large model training and high-value and high-end reasoning scenarios of Internet head manufacturers. Internet head manufacturers are also important customer areas of the Company.

In the first half of 2026, the Company has introduced customers in key areas such as Internet head enterprises and telecom operators. The specific business progress shall be subject to the announcement officially issued by the company. Thank you for your interest in the company!

Investor

By the end of June, the company's R & D team had 1019 people, accounting for more than 74%, and R & D investment accounted for 44.3% of revenue. In the critical period of "Huagang" architecture and new generation chip R & D, how can company balance the resource allocation of "infrastructure R & D" and "short-term commercial delivery" to ensure that high R & D investment can be transformed into replicable commercial profits more quickly?

Zhou yuanCo-founder and employee director, Moore Threads Technology Co. Ltd.

Dear investors, hello! We believe that the long-term investment and short-term commercial delivery of the underlying architecture is never a "two-choice" issue, but a value chain that blends with each other-every breakthrough at the architectural level will directly reduce the R & D cost and delivery cycle of the next generation chips, and each generation of products in the market and ecological developers. Only by making the "base" solid enough, reusable and scalable, can a new generation of architecture like "Huagang" be truly transformed into a business profit curve that can be continuously replicated and has scale effect.

With the product rhythm of unified architecture, unified software stack and end-cloud isomorphism, we are precipitating every high-intensity R & D investment into long-term compound interest with shorter delivery cycle, lower marginal cost and higher customer stickiness in the future.

Investor

What are the actual cases of the company in large model reasoning? What are the advantages?

Zhang YuboCo-founder, Director, Deputy General Manager, Moore Threads Technology Co. Ltd.

Dear investors, hello! In terms of reasoning performance, the company's joint ecological partners have made important progress in both technology and business. The PD heterogeneous separation solution launched by the company and its ecological partners is the best cost-effective solution in the industry at present. MTT S5000 decouples Prefill from the traditional inference link by virtue of high-density AI computing power, native FP8 precision support, and MUSA's source-level compatibility with CUDA.

It is built as a computing resource pool that can be independently supplied, independently billed and independently optimized, and the overall solution is more cost-effective than international flagship GPU products. In terms of mode adaptation, the company's AI training and promotion integrated intelligent computing card MTT S5000 has achieved Day-0 adaptation to the industry SOTA model, including MiniMax H3, Kimi-K3, GLM-5.3-Flash, etc. Thank you for your interest in the company!

Investor

Secretary Dong, how does the company's business performance in the first half of 2026 compare with other domestic GPU companies?

Xue YansongSecretary of the Board of Directors and Chief Financial Officer, Moore Threads Technology Co. Ltd.

Dear investors, hello! The first half of the year is an important stage for the company to move from technical verification to large-scale commercial delivery. The Company achieved an operating income of RMB 1.736 billion, a significant year-on-year increase of 147.42%, and exceeded the annual income in 2025, which was in the leading position in the industry, and the commercialization process continued to accelerate; the total gross profit was RMB 989 million, a year-on-year increase of 103.78%.

The net loss attributable to the parent company narrowed by 95.73% year on year, and the net profit deducted from non-attributable parent company was-151 million yuan, which narrowed by 52.37% year on year. More importantly, revenue growth is mainly driven by the commercialization of cloud computing products represented by Kua'e Intelligent Computing Cluster. This shows that the core competitiveness of the company's products and customer recognition are being translated into actual business results. Thank you for your interest!

Investor

According to the financial report, if the impact of share-based payment is deducted, the actual net profit of the company in the first half of the year is 83.53 million yuan, which has realized a substantial turnaround in operation. While maintaining high-intensity R & D investment, how can company minimize the erosion of book profits by share-based payments by optimizing the equity incentive plan and ensuring the stability of the core technology team?

Zhou yuanCo-founder and employee director, Moore Threads Technology Co. Ltd.

Dear investors, hello! The company attaches great importance to the long-term stability of core talents, and continues to pay attention to the periodic impact of share-based payment on business performance. Equity incentive is an important mechanism for companies to attract and retain core technical talents and promote long-term value creation.

In the future, on the premise of meeting the accounting standards and relevant regulatory requirements, the Company will further optimize the granting rhythm, coverage and assessment mechanism of equity incentive in combination with business development, talent incentive needs and operating results, so as to better balance the long-term incentive effect and the impact of share-based payment expenses on current profits while ensuring the stability of the core team and the intensity of R & D investment. Thank you for your interest!

Investor

At present, what kind of market position is the company's full-function GPU chip in the training and reasoning of domestic large models compared with its competitors?

Zhang JianzhongFounder, Chairman, General Manager, Moore Threads Technology Co. Ltd.

Dear investors, hello! With excellent product performance and perfect software and hardware ecology, the company has established a solid leading position in the domestic computing power market, and its products have penetrated into many high-barrier core industries and application scenarios. Moore Thread MTT S5000 product has officially passed the evaluation of China Information Security Evaluation Center and National Secret Science and Technology Evaluation Center, which indicates that it has reached the high standard of the industry in terms of technology maturity, security and stability.

In the field of large-scale AI infrastructure, the company's MTT S series intelligent computing cards have been deployed in a number of intelligent computing centers and cloud service platforms, widely supporting the computing power needs of large model training, reasoning, agent applications and scientific computing. The company is one of the few GPU suppliers in the market that truly realizes the commercial application of large-scale clusters of kilocalories and ten thousand cards under a single network for training.

In December 2025, Moore Thread released the "Huagang" architecture, which supports the expansion of intelligent computing clusters with a scale of more than 100,000 cards. It aims at the post-training tasks such as pre-training of large models and reinforcement learning, as well as the frontier directions such as MoE large models, VLA embodied intelligent models and video generation models, to build the "model factory" infrastructure capability for large-scale model training. Thank you for your interest in the company!

Investor

What is the level of the company's current R & D investment and future planning?

Zhou yuanCo-founder and employee director, Moore Threads Technology Co. Ltd.

Dear investors, hello! In the first half of 2026, the company invested 769 million yuan in research and development, with a year-on-year increase of 38.16%, and the total amount of research and development investment still maintained a high intensity; by the end of June 2026, the company had 1,019 research and development personnel, of which 76.84% had a master's degree or above. In addition, in the first half of 2026, the number of intellectual property rights reached a new high, with a total of 2352 applications for intellectual property rights, including 2049 invention patents, 951 authorizations and 725 invention patents.

In the first half of 2026, the company focused on R & D investment, including Mount Hua, Lushan Mountain chips and related applications under the new generation of "Huagang" architecture, to provide support for subsequent product iteration and revenue growth. Thank you for your interest!

Investor

Hello, what customers does the company have at present? What is the customer structure?

Zhang JianzhongFounder, Chairman, General Manager, Moore Threads Technology Co. Ltd.

Dear investors, hello! The company's various forms of GPU products are widely used in the Internet, cloud services, low-altitude economy, games, scientific research, government and other core computing acceleration scenarios, covering a wide range of major technology innovation platforms, Internet vendors, cloud service vendors, telecom operators, AI enterprises, individual consumers and other customer types. In the first half of 2026, the company has introduced customers in key areas such as Internet head enterprises and telecom operators. Thank you for your interest in the company!

Investor

In the first half of the year, the net profit returned to the mother has narrowed to close to the break-even point, can it be understood that the profit turning point is coming? What does the company think? When is it expected to be profitable?

Xue YansongSecretary of the Board of Directors and Chief Financial Officer, Moore Threads Technology Co. Ltd.

Dear investors, hello! The company's revenue in the first half of 2026 was 1.736 billion yuan, a substantial increase of 147.42% over the same period last year, and has exceeded the annual revenue in 2025, which is in the leading level in the same industry. In the first half of the year, the net profit attributable to the parent company was -11.5631 million yuan, with a loss narrowing by 95.73% year-on-year; the net profit deducted from non-attributable parent company was -151 million yuan, with a year-on-year narrowing by 52.37%.

At present, the company's operating fundamentals continue to improve and the commercialization process continues to accelerate, but the specific break-even node is affected by the macro environment, market competition and business landing and other comprehensive factors, please refer to the company's subsequent disclosure announcement. Thank you for your interest!

Investor

Have you ever thought about adapting to Hongmeng system and making computer mainframe with Huawei?

Zhang JianzhongFounder, Chairman, General Manager, Moore Threads Technology Co. Ltd.

Dear investors, hello! The company has always been committed to creating an open and compatible computing base and software and hardware ecosystem. The company has always maintained an open attitude and actively promoted the adaptation and collaboration of products with mainstream operating systems, software frameworks and application ecosystems. Thank you for your interest!

Investor

How is the current MUSA's compatibility with the CUDA ecosystem?

Zhou yuanCo-founder and employee director, Moore Threads Technology Co. Ltd.

Dear investors, hello! Software ecology is the core barrier that determines the commercialization of the company's GPU products. The company has built a complete MUSA software stack, covering a complete tool chain from the underlying driver, compiler, operator library to the upper development framework. More importantly, the MUSA architecture adopts a programming model compatible with CUDA from the GPU architecture to the developer kit, and the self-developed AUTOMUSIFY tool can help developers quickly migrate the source code to the MUSA platform and replace the CUDA API with the self-developed MUSA-X computing library.

The MUSA Toolkit provides compilation and invocation support to ensure efficient and easy migration of applications to the MUSA architecture. MUSA provides a complete developer tool chain, including compilers, debuggers, and performance analysis tools, to match the capabilities of the CUDA developer suite and reduce migration costs. This ability of "full stack ecology + high compatibility" not only greatly shortens the adaptation cycle of downstream customers, but also builds an ecological moat for the company in the fierce domestic substitution process.

Thank you for your interest in the company!

Investor

At this year's WAIC, I saw the information that Jingdong and Moore Thread have in-depth cooperation. In the Jingdong Model Factory, the two sides carry out model training, adaptation and tuning around MTT S5000. What progress has been made?

Zhang JianzhongFounder, Chairman, General Manager, Moore Threads Technology Co. Ltd.

In the first half of 2026, the company has introduced customers in key areas such as Internet head enterprises and telecom operators to serve large model training and reasoning application scenarios. The specific business progress shall be subject to the information publicly disclosed by the Company. Thank you for your interest in the company!

Investor

In the first half of the year, the company's comprehensive gross profit rate dropped from 69.14% in the same period last year to 56.95%, and in the second quarter, the single-quarter gross profit rate further dropped to 49.26%. Is this mainly affected by the increase in the proportion of hardware in the delivery of intelligent computing clusters and the increase in the cost of upstream OEM? With the adjustment of product structure in the second half of the year, is there a clear repair path for gross profit margin?

Zhang YuboCo-founder, Director, Deputy General Manager, Moore Threads Technology Co. Ltd.

Dear investors, hello! In the first half of the year, the company's comprehensive gross profit rate was 56.95%, mainly related to the rising price of raw materials in the semiconductor industry and the change of product delivery structure. Thanks to the company's insistence on the research and development of full-function GPU, the company's comprehensive gross profit rate is still in the first echelon of the industry. Thank you for your interest!

Investor

Will there be a consumer graphics card available Lushan Mountain?

Zhang JianzhongFounder, Chairman, General Manager, Moore Threads Technology Co. Ltd.

Dear investors, GPU products related to Lushan Mountain are expected to be launched within this year, subject to the company's official product release information, thank you for your attention!

Investor

I saw a White Paper on the Practice Technology of PD Fractional Computing Power Mixing Department, which mentioned that domestic GPUs can support the same reasoning ability as Nvidia's high-end cards as long as they are deployed at a ratio of 2:1. How to understand this PD separation technology? Can domestic GPUs replace Nvidia? Is it only possible with a fully functional GPU?

Zhang YuboCo-founder, Director, Deputy General Manager, Moore Threads Technology Co. Ltd.

Dear investors, hello! Faced with the dual challenges of the surge in demand for computing power and the diversification of autonomous security, how to break through the bottleneck of cost, efficiency and localization of AI infrastructure has become the key. Large model reasoning can be divided into the Prefill stage and the Decode stage, in which the Prefill stage requires high computing power, and the Decode stage requires high video memory bandwidth. By combining GPUs with different characteristics, it is the core path to break the above bottleneck.

Moore Thread, in cooperation with Token Service Technology Partner and Big Model Company, has completed in-depth hardware adaptation and will achieve production-level deployment in the real production-level cluster environment based on the PD separation reasoning scheme constructed by Moore Thread AI training and pushing integrated intelligent computing card MTT S5000. Through the Prefill computing pool solution built by MTT S5000, the price is comparable to the international flagship GPU products currently on the market. Thank you for your interest in the company!

Investor

This year, the company has released edge and terminal products such as AI book and AI Cube. Will these become new growth points for the company?

Xue YansongSecretary of the Board of Directors and Chief Financial Officer, Moore Threads Technology Co. Ltd.

Dear investors, hello! The company has formed a matrix of edge and terminal products based on the "Yangtze River" SoC, including MTT E300 AI computing module, MTT AIBOOK computing base and MTT AICUBE home AI hub. The SoC integrates full-featured GPU, high-performance CPU, NPU and VPU, and can support end-side large model reasoning. Among them, AIBOOK provides 50 TOPS heterogeneous AI computing power, runs MTT AIOS based on Linux kernel, has multi-system compatibility, and presets AI development environment and tool chain.

MTT E300 continues to expand in industry agents, industrial intelligent manufacturing, smart education and other scenarios, AIBOOK online and offline sales have been orderly promoted, and AICUBE has been put on the shelves of Jingdong. In the short term, the revenue of this business is still in the early stage of volume expansion; in the medium and long term, it is an important layout for the company to extend from cloud computing to personal intelligence, family intelligence and embodied intelligence.

Thank you for your interest in the company!

Investor

What are the problems solved by the three "AI factories" mentioned by the company? What does it have to do with the company's income?

Zhang JianzhongFounder, Chairman, General Manager, Moore Threads Technology Co. Ltd.

Dear investors, hello! Relying on the continuous iteration of core technology, the company achieves continuous breakthroughs around key business scenarios such as generative artificial intelligence, intelligent computing infrastructure, cloud native graphics computing, end-to-end intelligence and edge computation, and further forms a "three factories" capability system for the AI era, namely, model training factories and word production factories. Continue to consolidate the company's leading position in domestic GPU technology.

Among them, the model training factory is mainly oriented to large model training and post-training scenarios, relying on the ability of 100000 card-level super-large-scale intelligent computing cluster, providing high-performance AI Infra and basic software support for post-training tasks such as trillion-level parameter large model pre-training, reinforcement learning, and so on. It covers the frontier model directions such as hybrid expert large language model, VLA embodied intelligence model, video generation model, etc; The token production factory is mainly oriented to the current fast-growing large model reasoning business, serving Internet large model customers and government and enterprise industry customers.

On the one hand, it provides higher quality Token services through lower reasoning delay, on the other hand, it improves the utilization efficiency of computing power through higher system throughput, and continuously reduces the unit Token cost; Agent production factories are oriented to the new generation of Agent systems and embodied intelligent applications, providing AI infrastructure capabilities needed to support complex agent computing, scheduling and deployment, and further oriented to the embodied intelligent robot ontology, building a cloud-end collaborative agent computing support system.

Through the capacity building of "three factories", the company will form a more complete AI infrastructure closed-loop from the underlying GPU computing power, cluster interconnection, basic software to industry application scenarios. Thank you for your interest in the company!

Investor

Hello, please introduce the necessity and advantages of the company's "cloud, edge and end" computing system?

Zhang JianzhongFounder, Chairman, General Manager, Moore Threads Technology Co. Ltd.

Dear investors, hello! At present, the explosion of AI Agent and embodied intelligence is pushing the focus of computing power from cloud training to edge reasoning and terminal intelligence, and a single cloud computing power has been unable to meet the needs of differentiated deployment. Based on the self-developed MUSA architecture, Moore Thread full-function GPU has taken the lead in achieving technological breakthroughs in supporting AI computing acceleration, graphics rendering, physical simulation and scientific computing, ultra-high definition video coding and decoding on a single chip, and has created a rare "computing, rendering and simulation" integrated universal computing base in China.

At present, Moore Thread has built a full-stack product layout covering "cloud-edge-end" to accelerate full-scene intelligent computing with high-quality general computing power. In the field of cloud intelligent computing, Moore Thread has established a full-stack intelligent computing product line from chips, intelligent computing cards, servers to Wanka-level clusters. MTT S5000, the company's flagship AI training and promotion integrated intelligent computing card, continues to be mass-produced on a large scale, with full-precision and full-function general computing capabilities.

It has passed the National Safety and Reliability Assessment in the first batch, and has reached the industry-leading level in terms of technical maturity, stability and security. Based on MTT S5000, the Kua'e Wanka Intelligent Computing Cluster has been successfully deployed and put into service, completing the complete training of super-large parameter models, and many key indicators have reached the international mainstream level.

In the field of terminal and edge intelligent computing, Moore Thread takes self-developed intelligent SoC "Yangtze River" as the core, and launches inclusive terminal products-AI computing power MTT AIBOOK and family AI center MTT AICUBE, as well as edge AI module MTT E300, so that AI can further enter millions of households and all walks of life. Thank you for your interest in the company!

Investor

What is the core difference between a full-featured GPU and a single AI acceleration chip?

Zhang YuboCo-founder, Director, Deputy General Manager, Moore Threads Technology Co. Ltd.

Dear investors, hello! The company's full-featured GPU is based on its own MUSA architecture, and a single chip supports AI computing, graphics rendering, physical simulation and scientific computing, ultra-high definition video coding and decoding capabilities. The company is the first to release a graphics acceleration engine supporting DirectX 12 in China, and is also one of the few domestic GPU manufacturers that support FP8 to FP64 full computing accuracy natively.

This universality enables the company's products to serve not only large model training and reasoning, but also support scenarios such as digital twins, industrial simulation, embodied intelligence, cloud computers, edge computation and intelligent terminals. Compared with single scene chip, full-function GPU has stronger scene adaptation ability, customer reuse value and long-term ecological extension space. Thank you for your interest!

Investor

At present, the parameter scale of the domestic large head model is constantly increasing, and a large number of training cards are needed for training. Does the company cooperate with the domestic mainstream large model? Thank you!

Zhang JianzhongFounder, Chairman, General Manager, Moore Threads Technology Co. Ltd.

Dear investors, hello!

The company and its ecological partners have completed a number of landmark training practices: based on the Moore thread MTT S5000 intelligent computing cluster, the MoE-236B super-large parameter basic model has been completely trained from scratch; Based on FlagOS-Robo framework and relying on MTT S5000 intelligent computing cluster, the company and Beijing Zhiyuan Artificial Intelligence Research Institute successfully completed the whole process training of RoboBrain2.5, a self-developed embodied brain model; The EvoPhys team of Peking University trained EvoPhys-World, a 5D world model with "people" as the center and "everything can be controlled at the scene level", on the MTT S5000 full-function GPU.

This cutting-edge research achievement ranked first in the "World Generation" track in the WorldScore public evaluation list of Stanford University, and remained at the top for 37 days. Aiming at DeepSeek and other cutting-edge models with industry benchmarking significance, the company, in conjunction with its ecological partners, took the lead in completing the deep adaptation and performance breakthrough of DeepSeek- V4 on MTT S5000 based on innovative technologies such as native FP8 low-precision computing power.

In addition, Moore Thread has realized the rapid adaptation of the latest batch of SOTA models, such as MiniMax H3, Kimi-K3, GLM-5.3-Flash and other latest release models, so as to achieve "release is adaptation". Thank you for your interest in the company!

Investor

What are the main application scenarios of the company's box AI Cube products? Are they similar to Mac mini?

Zhou yuanCo-founder and employee director, Moore Threads Technology Co. Ltd.

Dear investors, hello! In order to meet the demand for end-side computing power in the Token era, Moore Thread, based on its self-developed "Yangtze River" intelligent SoC, launched the first consumer product for home scenarios, MTT AICUBE, which realized the extension of computing power from cloud computing cluster to edge and terminal, and enabled more inclusive home intelligent scenarios. Home AI central device MTT AICUBE integrates three product forms of AI Agent, AI PC and AI NAS.

The product has a built-in global agent "wheat", which supports cross-application control of a large number of APPs and provides intelligent active services; the all-flash AI NAS module provides safe and efficient local storage and intelligent management for family data; At the same time, MTT AICUBE also has complete desktop AI PC capabilities, which can easily meet the all-round needs of home users, such as movie entertainment, efficient office, online learning, cloud games and local large model operation. Thank you for your interest in the company!

Investor

I would like to ask the current progress of MUSA software ecosystem construction, and what scale has the company's developer ecosystem reached?

Zhou yuanCo-founder and employee director, Moore Threads Technology Co. Ltd.

Dear investors, hello! The MUSA ecosystem is becoming more and more mature, and the number of developers in the company has exceeded 800000. The company relies on "Moore College" to build a systematic learning path, covering enterprise developers, scientific research institutions and students, to enhance developers'awareness and ability to use products. Moore College platform provides rich free course resources, and MUSA series courses have been launched on the national smart education platform "Enlightenment Learning Community".

SGLang, the world's top reasoning framework, has completed the integration of MUSA code mainline, marking that the domestic GPU open source ecosystem has entered the era of "native support". The company's open source TileLang-MUSA operator development tool, MUSA software stack continues to enrich. SimuMax, an open source cluster performance simulation tool, provides intelligent parallel strategy search and full-stack workflow support for large-scale distributed training. Thank you for your interest in the company!

Investor

Hello, I would like to ask about the capacity guarantee of your upstream wafer foundries in the second half of the year. How many foundry capacities (equivalent cards) have been locked in? How much is the increase compared with the first half of the year?

Zhang JianzhongFounder, Chairman, General Manager, Moore Threads Technology Co. Ltd.

Dear investors, hello! At present, the company's overall supply chain and production capacity can match the needs of business development. With the continuous development and improvement of the domestic semiconductor industry chain, the supply capacity of upstream manufacturing and supporting links is also constantly improving. The company will continue to deepen the synergy with industry chain partners to further enhance the stability and resilience of the supply chain. Thank you for your interest!

Investor

Mr. Zhang, please introduce Moore Thread's company positioning, long-term vision and plans for the next five years. Thank you!

Zhang JianzhongFounder, Chairman, General Manager, Moore Threads Technology Co. Ltd.

Dear investors, hello! With full-function GPU as the core, the company is committed to providing accelerated computing infrastructure and one-stop solutions to the world, providing AI computing support for the digital transformation of all walks of life. The company's goal is to become an internationally competitive GPU leader, to build an advanced accelerated computing platform for the digital world that integrates artificial intelligence and digital twins, with the vision of "accelerating for a better world".

The company will deepen the full-function GPU technology layout for the new generation of general super-intelligent computing center, and continue to strengthen the R & D investment in cutting-edge capabilities such as scientific computing, AI generative rendering and hardware-level ray tracing, so as to meet the urgent needs of emerging scenarios such as embodied intelligent simulation training and AI4S for compound computing power.

Facing the needs of large model training and reasoning, the company continues to promote the integrated layout of high-performance domestic GPU, advanced Internet, cluster software stack and computing power scheduling platform, and builds an autonomous and controllable technology base to support 100000 card-level intelligent computing power cluster. In the future, the company will continue to iterate around key capabilities such as high-density deployment, high-speed interconnection, stable operation and maintenance, and efficient energy consumption ratio, so as to promote the 100,000-card cluster from single-point product capability to systematic, engineering and large-scale landing.

In the direction of edge and terminal, the company will accelerate the iterative upgrade of SoC, focus on the development trend of agents, look forward to the layout of end-side computing power, lightweight model and software and hardware platform capabilities, and promote the upgrade of terminal devices from "connectable" to "perceptible, understandable and executable" agent carriers. Thank you for your interest!

Investor

What is the progress of the company's layout in the field of embodied intelligence and physical AI?

Zhang YuboCo-founder, Director, Deputy General Manager, Moore Threads Technology Co. Ltd.

Dear investors, hello! The company has opened up a complete link of embodied intelligence from synthetic data generation, large model training, simulation verification to end-side hardware deployment. MT Lambda is a full-stack intelligent simulation platform, and the company has also open sourced the GPU-accelerated physical simulation back-end MuJoCo Warp MUSA based on MUSA architecture, and completed the verification of four-legged robot dog and biped humanoid robot Sim2Real.

The company has set up an industrial embodied intelligence innovation center in Wuxi, jointly built with ecological partners, and established a "Joint Laboratory of Embodied Intelligence Computing and Simulation" with the National Embodied Intelligence Application Pilot Base. Embodied intelligence requires AI computing, physical simulation, graphics rendering and end-to-end reasoning collaboration, which just reflects the differentiation advantages of full-featured GPU. Thank you for your interest in the company!

Investor

What are the advantages of the company's S5000 products over other domestic competitors?

Zhang YuboCo-founder, Director, Deputy General Manager, Moore Threads Technology Co. Ltd.

Dear investors, hello! MTT S5000 is a full-featured GPU intelligent computing card designed for large model training, reasoning and high-performance computing. It is based on the fourth-generation MUSA architecture "Pinghu" and fully supports full-precision computing from FP8 to FP64. MTT S5000 relies on FP8 to improve the efficiency of training and pushing, FP64 to ensure accurate computing, ray tracing to achieve high-fidelity rendering, and provides safe, efficient and reliable core domestic computing support with autonomous and controllable strong underlying computing power.

The unique advantages of the company's products are mainly reflected in the following aspects: First, it insists on creating "full-function GPU" products. Moore Thread adheres to the full-function GPU route. A single chip supports AI training, graphics rendering (the first release in China to support DirectX 12), physical simulation and scientific computing, covering the whole scene of "cloud-edge-end", with strong product versatility.

Second, excellent system-level cluster performance, in the era of super-large-scale intelligent computing center, the overall performance of the "Kua'e" Wanka cluster based on MTT S5000 is comparable to the international advanced level of the same generation.

Third, the ecological base of low-cost migration, relying on the self-developed MUSA unified architecture, the product not only adapts to PyTorch, vLLM, SGLang, TileLang and other mainstream frameworks, but also has excellent compatibility with CUDA. The AUTOMUSIFY tool enables developers to migrate applications at a lower cost, significantly reducing customer deployment cycles. Thank you for your interest!

Investor

I heard that S6000 is popular in the Internet industry and has received more orders. When will its performance be reflected?

Zhang JianzhongFounder, Chairman, General Manager, Moore Threads Technology Co. Ltd.

Dear investors, thank you for your attention. Please refer to the company's announcement for details.

Investor

What is the strategic significance of the company's launch of MTT C256 super node, and what is the progress of super node research and development?

Zhang JianzhongFounder, Chairman, General Manager, Moore Threads Technology Co. Ltd.

Dear investors, hello! MTT C256 is a super-node architecture plan released by the company for the next generation of super-large-scale intelligent computing centers. Its innovative architecture is the first layer of Scale-up network, which realizes the high-speed interconnection of 256 cards in parallel cabinet expansion, and compresses the communication delay between cards to sub-microsecond level. Its design integrates high-efficiency liquid cooling, three-blind insertion design, perfect RAS mechanism and cluster management system, demonstrating the engineering ability based on Wanka training precipitation, which greatly improves the deployment density of intelligent computing center while ensuring long-term stable operation.

This product will mainly serve the training and promotion cluster of 10000 cards and even 100000 cards, which is an important hardware foundation for the company's "model training factory" and future super-large-scale AI infrastructure. Please pay attention to the company's relevant announcements on the latest R & D progress. Thank you for your interest in the company!

Investor

What are the representative achievements of the company in the "National Core Training Model"?

Zhang YuboCo-founder, Director, Deputy General Manager, Moore Threads Technology Co. Ltd.

Dear investors, hello!

The company and its ecological partners have completed a number of landmark training practices: based on the Moore Thread MTT S50 million card-level intelligent computing cluster, the partners have completely trained the MoE-236B super-large parameter basic model from scratch; Based on FlagOS-Robo framework and relying on MTT S5000 intelligent computing cluster, the company and Beijing Zhiyuan Artificial Intelligence Research Institute successfully completed the whole process training of RoboBrain2.5, a self-developed embodied brain model; The EvoPhys team of Peking University trained EvoPhys-World, a 5D world model with "people" as the center and "everything can be controlled at the scene level", on the MTT S5000 full-function GPU.

This cutting-edge research achievement ranked first in the "World Generation" track in the WorldScore public evaluation list of Stanford University, and continued to top the list for 37 days. Thank you for your interest in the company!

Investor

Moore thread every day to adapt to this game, that game, now have money, do their own game engine, do their own games, ah, or with the engine company, or deep cooperation with the game company

Zhang JianzhongFounder, Chairman, General Manager, Moore Threads Technology Co. Ltd.

Dear investors, hello! The company focuses on the R & D, design and industrialization of full-function GPU, and is committed to providing a unified computing base for AI computing, graphics rendering, physical simulation and scientific computing. Game and engine companies are important ecological partners of the company. At the first MUSA developer conference in December 2025, the company released a new generation of "Huagang" architecture and a high-performance graphics rendering "Lushan Mountain" chip based on the architecture, which will provide a new upgrade experience for game and graphics users in the future.

The chip is expected to achieve a 15-fold increase in 3A game rendering, a 64-fold increase in AI performance and a 50-fold increase in ray tracing performance. Thank you for your interest in the company!

Zhang JianzhongFounder, Chairman, General Manager, Moore Threads Technology Co. Ltd.

This is the end of today's meeting. Thank you very much for your participation and communication. At the same time, I would like to thank the Shanghai Stock Exchange for providing us with a platform and service for communication. After the meeting, you are welcome to continue to communicate with us through e-mail, e-interaction and other means! At the same time, I hope you will continue to care about and support the Moore thread. Thank you!

ModeratorSSE Roadshow Center

Dear investors and netizens, the semi-annual performance presentation of Moore Thread 2026 is drawing to a close. We invite guests from listed companies to make concluding remarks.

ModeratorSSE Roadshow Center

This is the successful conclusion of the Moore Thread 2026 semi-annual performance presentation. Thank you for the active participation of investors and netizens. If you have further questions, you can keep communicating with the company through the e interactive platform of Shanghai Stock Exchange. Thank you!

About this transcript

The call was held in another language and translated. The summary, guidance, analyst focus, and tone are automated and should be checked against the call before being relied on. Names and figures can be misheard. Listen to the replay.