Key takeaway

1. Results grew strongly, while losses narrowed significantly. In 1H26, the company recorded operating revenue of RMB1.736bn, up 147.42% YoY; net loss attributable to shareholders of the parent company was RMB12mn, narrowing by 95.73% YoY. In 2Q26 alone, operating revenue reached RMB999mn, up 141.88% YoY. Net loss attributable to shareholders of the parent company was RMB41mn, narrowing by 74.17% YoY.

2. S5000 achieved large-scale shipment volume growth, while inventories increased substantially. In 1H26, the company's inventories reached RMB3.550bn, up 166.48% from end-2025, while accounts receivable reached RMB880mn, up 102.89% from the beginning of the year. This indicates that the company is responding to rapid growth in downstream demand while ensuring the supply chain's continued stability.

3. MTT S5000 clusters achieved large-scale delivery, while the model ecosystem and application scenarios continued to expand. In 1H26, the company completed delivery under its RMB660mn KUAE cluster contract, and its intelligent computing clusters were deployed for multiple key customers, including internet companies, telecom operators, and customers in Beijing and Wuxi. The company continued to advance R&D of the “Huagang” architecture and next-generation products such as Huashan and Lushan. The number of MUSA developers exceeded 800,000. The company's full-stack capabilities spanning chips, clusters, and ecosystem continued to improve, laying the foundation for subsequent product upgrades and customer expansion

Event

The company released 1H26 results. In 1H26, revenue reached RMB1.736bn, up 147.42% YoY; net loss attributable to shareholders of the parent company was RMB12mn, narrowing by 95.73% YoY; net loss attributable to shareholders of the parent company excluding non-recurring gains and losses was RMB151mn, narrowing by 52.37% YoY; and gross margin was 56.95%. The company generated revenue of RMB999mn in 2Q26, up 141.88% YoY and 35.40% QoQ.

Quick Take

1. Revenue surged and losses narrowed substantially, while business scale continued to expand in 2Q26.

In 1H26, the company generated revenue of RMB1.736bn, up 147.42% YoY, mainly driven by growth in demand for full-function GPUs amid the rapid development of the artificial intelligence industry, as well as accelerated commercialization and stable deliveries of KUAE intelligent computing clusters. Revenue from cloud products reached RMB1.693bn, accounting for 97.49% of total revenue. In 1H26, net loss attributable to shareholders of the parent company was RMB12mn, narrowing by 95.73% YoY; net loss attributable to shareholders of the parent company excluding non-recurring gains and losses was RMB151mn, narrowing by 52.37% YoY. In 2Q26, revenue reached RMB999mn, up 141.88% YoY and 35.40% QoQ; net loss attributable to shareholders of the parent company was RMB41mn, narrowing by 74.17% YoY; and net loss attributable to shareholders of the parent company excluding non-recurring gains and losses was RMB97mn, narrowing by 46.55% YoY. In 1H26, gross profit reached RMB989mn, up 103.78% YoY; gross margin was 56.95%, down 12.19 pcts YoY. Gross margin was 49.26% in 2Q26, down 13.94 pcts YoY and 18.09 pcts QoQ. The decline in gross margin was mainly due to higher semiconductor raw material prices, as rising procurement costs for raw materials placed temporary pressure on product gross margin. In 1H26, the difference between net profit attributable to shareholders of the parent company and the corresponding figure excluding non-recurring gains and losses was approximately RMB139mn, mainly due to non-recurring items such as government grants. During the same period, the company recognized other income of RMB91mn, investment income of RMB30mn, gains from changes in fair value of RMB28mn, and asset impairment losses of RMB39mn.

2. Expense ratios improved significantly while R&D investment remained high; capacity expansion and inventory buildup increased capital tied up in operations.

On the expense side, selling expense, administrative expense, R&D expense, and financing expense were RMB158mn, RMB151mn, RMB769mn, and RMB30mn, respectively, in 1H26. The corresponding expense ratios were 9.11%, 8.69%, 44.30%, and 1.71%, down 1.08 pcts, 13.05 pcts, 35.03 pcts, and 0.96 pcts YoY, respectively. In 2Q26, the selling expense, administrative expense, R&D expense, and financing expense ratios were 8.19%, 7.79%, 40.06%, and 1.92%, respectively, down 1.47 pcts, 11.08 pcts, 35.17 pcts, and 1.35 pcts YoY, respectively. The combined ratio of the four expenses was 63.81% in 1H26 and 57.96% in 2Q26, down 50.13 pcts and 49.07 pcts YoY, respectively, as economies of scale continued to materialize. The increase in selling expense was mainly attributable to team compensation and investment in customer service. The increase in R&D expense was mainly directed toward the “Huagang” architecture, the Huashan and Lushan chips, and related applications. The increase in financing expense was mainly due to interest on bank borrowings, while administrative expense remained broadly stable. Net cash outflow from operating activities was RMB2.169bn in 1H26, an increase of RMB1.005bn YoY. Period-end inventories stood at RMB3.550bn, up 166.48% from the beginning of the year. Accounts receivable stood at RMB880mn, up 102.89% from the beginning of the year, mainly because inventory buildup for capacity expansion and revenue growth increased working capital requirements.

3. Cloud products contributed the core revenue, while edge and endpoint products opened up incremental growth opportunities.

By business segment, cloud products generated revenue of RMB1.693bn in 1H26, accounting for 97.49% of total revenue, with a gross margin of 57.37%. Edge and endpoint products generated revenue of RMB41mn, accounting for 2.33% of total revenue, with a gross margin of 43.16%. Revenue from direct sales and distribution was RMB1.172bn and RMB564mn, respectively. Period-end contract liabilities stood at RMB71mn, up 230.79% from the beginning of the year. Revenue corresponding to outstanding performance obligations totaled RMB88mn, of which RMB87mn is expected to be recognized within the year. The cloud business achieved significant volume growth, but revenue and customer concentration remained high. Attention should be paid to the acceptance of major projects and the volume ramp-up of edge and endpoint products.

4. KUAE clusters achieved large-scale delivery, while the MUSA ecosystem accelerated its integration with mainstream large models.

In 1H26, the company continued to strengthen its foundation in the intelligent computing industry through MTT S5000, its fourth-generation chip architecture. The KUAE intelligent computing cluster built on MTT S5000 achieved model FLOPs utilization (MFU) of 60% and 40% in Dense and mixture-of-experts (MoE) model training, respectively. Effective training time exceeded 90%, while linear scaling efficiency reached 95%. In 1H26, the company completed delivery and recognized revenue for KUAE contracts worth RMB660mn, further validating its large-scale commercial delivery capabilities. Regarding the model ecosystem, the company continued to adapt mainstream large models to its MUSA computing platform. Leveraging technologies including native FP8 low-precision computing, it completed in-depth adaptation and performance optimization for DeepSeek-V4. It also progressively adapted state-of-the-art (SOTA) models such as MiniMax M3 and Zhipu GLM-5.2. As its model adaptation capabilities and integrated software-hardware ecosystem continue to improve, the company is expected to drive the commercial deployment of domestic computing power across more business scenarios.

5. Huagang architecture ushers in a new product cycle, while edge intelligence and cloud PCs broaden application scenarios.

The company continues to advance the R&D of its next-generation GPU architecture, Huagang, and related products. The architecture supports full-precision computing from FP4 to FP64, delivers a 50% increase in computing power density over the previous generation, and enables clusters with more than 100,000 GPUs through MT Link. The Huashan and Lushan chips, developed based on the Huagang archi tecture, will remain under development in 1H26. They will target integrated AI training and inference and high-performance graphics rendering, respectively. For edge and endpoint applications, the company continues to expand its E300, AIBOOK, AICUBE, and other product offerings based on the Yangtze River SoC. The E300 has made progress in scenarios including industry-specific AI agents, intelligent manufacturing, and smart education. In the cloud PC market, a single MTT S3000 card supports 32 concurrent sessions and has been commercially deployed by telecom operators including China Mobile. As the new architecture, new chips, and endpoint products are gradually commercialized, the company is expected to further broaden its product portfolio and revenue sources.

6. Earnings forecast and investment recommendation

We believe the company is evolving from a chip and board supplier into an integrated hardware -software intelligent computing platform provider, supported by the expansion of domestic computing infrastructure, domestic substitution of full-function GPUs, and large-scale deliveries of KUAE clusters. The KUAE cluster has demonstrated the company's ability to deliver large-scale intelligent computing systems. The MUSA computing platform continues to deepen its compatibility with mainstream large models, while applications such as edge intelligence and cloud PCs are also being deployed at a faster pace. Meanwhile, the Huagang architecture, Huashan and Lushan chips, and C256 supernode are expected to drive the next product cycle. Although high R&D investment, supply chain costs, and the delivery and acceptance schedules of major projects may continue to cause fluctuations in profit and gross margin in the near term, economies of scale are expected to gradually emerge as revenue grows, the product portfolio improves, and the software ecosystem matures.

We forecast revenue of RMB3.744bn, RMB5.861bn, and RMB8.461bn for 2026–2028, respectively, representing YoY growth of 148.7%, 56.6%, and 44.4%. Based on the current market capitalization, this corresponds to PS multiples of 75.1x, 47.9x, and 33.2x, respectively. We forecast net profit attributable to shareholders of the parent company of -RMB97mn, RMB663mn, and RMB1.684bn, respectively. Under the base-case scenario, we expect the company to achieve a full-year profit in 2027, corresponding to PE multiples of 424.1x and 166.9x for 2027 and 2028, respectively. As a rare domestic full-function GPU company, the company has integrated delivery capabilities covering chips, software ecosystems, and intelligent computing clusters, and is expected to continue benefiting from growing demand for domestically produced computing power. We remain optimistic about the company’s long-term competitiveness and initiate coverage with an “Overweight” rating.

Based on our sensitivity analysis, we forecast revenue for 2026-2028 at RMB3.744bn/RMB5.861bn/RMB8.461bn under the base-case scenario, RMB4.492bn/RMB7.034bn/RMB10.153bn under the optimistic scenario, and RMB2.995bn/RMB4.689bn/RMB6.769bn under the pessimistic scenario, respectively. We forecast net profit attributable to shareholders of the parent company for 2026-2028 at - RMB97mn/RMB663mn/RMB1.684bn under the base-case scenario, RMB333mn/RMB1.327bn/RMB2.627bn under the optimistic scenario, and -RMB528mn/-RMB2mn/RMB740mn under the pessimistic scenario, respectively.

Risks

1. Technology and product iteration risks: GPU architectures, advanced packaging, cluster interconnects, and software ecosystems are evolving rapidly. If progress in adapting new architectures, chips, or models falls short of expectations, product competitiveness may be affected.

2. Supply chain and cost risks: The company has been placed on the US Entity List, while semiconductor raw material prices are subject to fluctuations. If supplies are constrained, procurement prices continue to rise, or the onboarding of alternative suppliers does not proceed smoothly, gross margin may come under pressure and deliveries may be affected.

3. Customer concentration and project delivery risks: Individual orders for large-scale intelligent computing clusters are substantial. If demand from major customers, acceptance schedules, or payment arrangements change, revenue recognition and cash flow may fluctuate.

4. Working capital and asset impairment risks: Capacity expansion and inventory buildup have increased capital tied up in inventories and accounts receivable and placed greater pressure on operating cash flow. If demand, product prices, or customer payments fall short of expectations, pressure from inventory write-downs and credit impairment may increase.

5. Industry competition and demand risks: Leading international GPU vendors have significant advantages in performance, ecosystems, and costs, while domestic vendors are also accelerating investment. If the company’s market expansion or ecosystem development, or downstream demand for computing power, falls short of expectations, earnings growth may slow.