Monday, August 17, 2026
DarkSubscribe
AI Infrastructure · News & Analysis
HomeChips & HardwareReport
Chips & Hardware · Report

A major Chinese domestic GPU company reported half-year revenue exceeding its full-year prior performance, signaling accelerated AI chip adoption in China.

Chinese GPU vendor growth validates the domestic AI chip market and demonstrates that geographic bifurcation is driving alternative silicon ecosystems.
Trade pressSlicast · August 12, 2026 · China · Source: Google News
importance 69

On August 9, Moore Threads (688795.SH), the leading Chinese GPU manufacturer, released its H1 2026 interim report. The core figures speak for themselves: the company achieved revenue of RMB 17.36 billion in the first half of 2026, a significant 147.42% year-over-year increase that already exceeds the full-year 2025 revenue. Gross profit reached RMB 9.89 billion, up 103.78% year-over-year. Profitability improved markedly, with net profit attributable to the parent company and adjusted net profit narrowing losses by 95.73% and 52.37% respectively compared to the prior year, as commercialization accelerated across the board.

Behind this impressive interim report lies a watershed moment: the concentrated realization of Moore Threads' commercialization capabilities following years of technological R&D accumulation, within an industry cycle of accelerating domestic computing power substitution. The growth logic revealed in this financial statement traces the complete path of the domestic GPU leader's "commercialization advancement."

To state the conclusion upfront: Moore Threads' profitability improvement in this interim report is not cosmetic cost-cutting, but reflects solid underlying industry fundamentals.

Why could revenue explode? Two core reasons: strong market demand for full-function GPUs and the scaled commercialization of intelligent computing cluster business.

The logic is straightforward: the AI industry is transitioning from large model R&D toward deep penetration in industry applications, as cloud training and inference computing demand continues to surge. In this environment, "full-function GPUs" capable of handling both training and inference simultaneously have become the absolute centerpiece of computing infrastructure.

Here's the key point worth reading thrice: Moore Threads' product layout precisely captured the inflection point of industry demand. Growth is not a short-term pulse but sustainable expansion supported by long-term industry trends.

First dimension: continuous scale effects and declining unit costs. GPU chips exhibit typical high fixed-cost, low marginal-cost characteristics. As revenue scale accelerates, prior R&D investments and fabrication costs are amortized across larger shipment volumes. Combined with simultaneous increases in gross profit, the product's profitability model continues to optimize. This profitability improvement driven by scale growth demonstrates strong sustainability; as subsequent shipment volumes increase further, profitability elasticity will continue to release.

Second dimension: R&D investment has not contracted but continues to accelerate. This is extremely important. The report shows H1 R&D investment reached RMB 769 million, up 38.16% year-over-year. Since 2022, cumulative R&D investment has approached RMB 5.9 billion, with R&D personnel representing 74.16% of total headcount. As of June 2026, the company has cumulatively applied for 2,167 patents, including 2,049 invention patents, with 725 invention patents granted.

In short: the company's profitability improvement is not achieved by cutting long-term investments for short-term profits, but through enhanced commercialization efficiency while maintaining technology iteration pace. The technological moat? They're still digging.

R&D investment provides technological confidence, while commercialization is the true test of hard technology value. In H1 2026, Moore Threads achieved breakthroughs in intelligent computing business scale-up, combined with continued MUSA ecosystem expansion, jointly supporting both earnings certainty and long-term competitive barriers.

On the product front, the flagship MTT S5000 unified training-and-inference intelligent computing card, serving as the core growth engine, has achieved continuous scaled mass production. Numbers don't lie:

In simple terms: the comprehensive performance combining high computing power and full precision makes this product competitive with mainstream international products in large model training and inference. Moreover, it became the first to pass the national "Security and Reliability Certification," achieving industry-leading standards in technical maturity, security and stability, naturally suited to computing demand in critical sectors like government, finance, and energy.

On commercialization, MTT S5000 intelligent computing clusters have achieved scaled sales with completed deployments and delivery across Beijing, Wuxi, Hangzhou and other locations. These multi-city commercial achievements have reassured the market—Moore Threads' engineering capability and product reliability withstand scrutiny.

Moore Threads is continuously expanding its product capability ceiling with notable achievements in "training national models with national chips." Based on the Kuawanu intelligent computing cluster, Moore Threads has completed full-stack training of a MoE-236B foundational large model on 250 trillion Token corpora, supporting Peking University's global-first 5D world model "EvoPhys-World" which topped Stanford's leaderboard (ranking first for 37 days). The open-source MusaCoder code model achieves operator performance on par with international SOTA levels.

Looking at product roadmap, Moore Threads is accelerating R&D of next-generation products based on the fifth-generation full-function GPU Huagang architecture: the Huashan and Lushan chips.

Huashan targets unified training-and-inference scenarios for data centers, with multi-dimensional advantages and even surpassing capabilities versus international mainstream chips in floating-point computing power, memory bandwidth, memory capacity, and high-speed interconnect bandwidth. Lushan focuses on high-performance graphics rendering while incorporating large model inference capability, achieving 15x improvement in 3A game rendering performance, 64x improvement in AI performance, and 50x improvement in ray-tracing performance compared to the previous generation. Their market introduction this year would further unlock new growth avenues for the company.

Simultaneously, there's an often-overlooked but extremely important signal: the company's customer structure continues optimizing with accelerating penetration toward core computing power customers like internet companies and telecom operators. These customers have large-scale procurement, high technical barriers, lengthy certification cycles—once integrated into supply chains, order continuity and stability are exceptionally strong, significantly enhancing the company's growth resilience. At WAIC 2026, Moore Threads unveiled collaborative achievements with industry leaders including Jiajia Vision and Querying Technology, demonstrating the company's implementation capabilities in cutting-edge scenarios like large model training and inference, embodied AI, and world models.

If hardware products determine current market competitiveness, the ecosystem defines the long-term growth ceiling. Moore Threads' MUSA ecosystem is constructing multi-layered competitive moats:

First, extreme migration cost advantages. MUSA achieves deep CUDA adaptation with 100% core mathematics library compatibility and 3,000+ PyTorch operator compatibility, receiving official support from mainstream inference frameworks like vLLM and SGLang. Mainstream large models including DeepSeek, Kimi, and GLM achieved Day-0 rapid adaptation, enabling customers to migrate from existing systems to domestic computing power platforms at near-zero cost, establishing extremely high replacement barriers.

Second, the developer flywheel has taken shape. As of H1 2026, the MUSA ecosystem developer community exceeded 800,000, formally entering a positive cycle of "developer expansion—enriched application scenarios—growing customer scale," with ecosystem stickiness continuously strengthening as scale expands.

Third, forward-looking positioning in next-generation computing infrastructure. Leveraging the unified computing advantages of "computation, rendering, simulation," the company has built through the complete ecosystem from "large model training—simulation—edge deployment," with layouts including the MT Lambda simulation platform and the Wuxi embodied AI innovation center, extending ecosystem boundaries from cloud intelligent computing to embodied AI and edge computing, preemptively capturing the industry's commanding position in next-generation computing infrastructure.

While attention focuses on business performance, Moore Threads' move toward Hong Kong listing deserves particular attention, viewed as the company's strategic positioning aligned with policy orientation and industry globalization trends.

Policy signals are already quite clear, with continuous signals supporting leading enterprises in leveraging "two markets and two types of resources." In August 2026, the China Securities Regulatory Commission publicly stated it "actively supports qualified mainland enterprises in listing in Hong Kong." Since 2024, over 270 mainland enterprises have completed Hong Kong listing registrations with cumulative fundraising exceeding HKD 650 billion. The Hong Kong Stock Exchange has simultaneously relaxed rules, lowering initial public float thresholds for A+H issuers, continuously widening listing channels.

Market feedback similarly confirms this trend: 2025 Hong Kong IPO fundraising totaled approximately 224% year-over-year growth, with A+H listings being the core growth engine. In just H1 2026, 24 A-share companies completed Hong Kong listings, with queued enterprises continuing to expand, making Hong Kong listing an increasingly common choice for mainland enterprises.

For hard technology enterprises, establishing dual A+H capital platforms has become standard strategic practice among industry leaders. On one hand, it leverages Hong Kong's international attributes to smooth single-market volatility impacts on long-term R&D investment, ensuring uninterrupted research momentum. On the other hand, it optimizes capital structure to hedge geopolitical uncertainty, significantly enhancing overall risk resilience.

Additionally, it paves global pathways for consumer-grade new products. The next-generation flagship gaming graphics card "Lushan," built on the "Huagang" architecture and expected to launch by year-end, will benefit from Hong Kong listing's international brand visibility and overseas channel resources as important catalysts for entering the global consumer market, helping the company truly unlock a second earnings growth curve.

The other side involves hard performance plus hard technology providing solid issuance fundamentals. H1 revenue doubling and continued loss narrowing, combined with long-term high-intensity R&D investment and patent reserves, constitute the company's most solid issuance foundation. This growth certainty and technological hard power are expected to gain recognition from global investors, while the dual capital platform boost will provide ample capital support for Moore Threads to capture the golden period of global computing infrastructure industry.

Revenue scale, loss improvement, and commercialization success across three dimensions all exceeded market expectations, confirming Moore Threads has accelerated entry into its commercialization realization cycle.

However, one must recognize that domestic GPU sector competition remains intense, with subsequent Huashan and Lushan mass production, overseas market expansion, and achieving full profitability remaining core market observation points. This impressive interim report is a milestone, far from the finish line.

Read the original
A major Chinese domestic GPU company reported… · Slicast