YICAI | Hard-Tech AI Leaders Post Earnings Beat: Supply Chain Synergies Fuel New Growth Cycle

Fueled by surging demand for AI computing power, Moore Threads, Biwin Storage, Huafeng Technology, and Taclink Optoelectronics achieved record-high performance in 2025 and sustained their strong growth momentum into the first quarter of 2026, demonstrating strong resilience in growth.

The year of 2025 marked the critical breakthroughs in China's hard-tech AI sector.As generative AI rapidly evolves toward agentic AI, the parameter count of large models has grown exponentially, giving rise to massive demand for computing, storage, and interconnection. The coordinated development and upgrade of computing, storage, and interconnection have thus become the core theme of the sector.The large-scale rollout of AI infrastructure construction has created unprecedented market opportunities across the entire semiconductor supply chain, while fully unlocking the value of hard-tech companies listed on the STAR Market.

Recently, Yicai launched its innovative dialogue program Dialogues on Growth • China's Sci-Tech Innovation in Annual Reports, a series of performance briefing sessions for STAR Market companies. The second episode focused on hard tech AI and featured senior executives from four industry-leading companies—Moore Threads (688795.SH), Biwin Storage (688525.SH), Huafeng Technology (688629.SH), and Taclink Optoelectronics (688205.SH)— in a joint dialogue with Liu Yang, Chief TMT Analyst at Shenwan HongyuanSecurities Company.According to the latest financial data, all four companies achieved explosive performance growth in 2025 and the first quarter of 2026, with sustained technological breakthroughs and accelerated domestic substitution progress.These results represent only outstanding operational achievements for the companies themselves, but also tangible fruits of capital market support for hard-tech development, underscoring the solid steps China's AI industry has taken in advancing from technology catch-up to global competition.

Performance Surges Across the Board as Technical Moats Solidify Core Competitiveness

Driven by surging demand for AI computing power, all four companies achieved record-high performance in 2025 and sustained their strong growth momentum into the first quarter of 2026, demonstrating strong growth resilience.Leveraging clear strategic positioning and deep technological accumulation, each company has established hard-to-replicate competitive advantages in their respective niche segments.

Moore Threads, a leading domestic full-function GPU player, delivered results that exceeded expectations.The company achieved operating revenue of RMB 1.51 billion in 2025, representing a significant year-on-year increase of 243.37%; in the first quarter of 2026, revenue further reached RMB 738 million, up 155.35% year-on-year, showing a clear acceleration trend.It has consistently maintained high-intensity R&D investment, with R&D expenses reaching RMB 1.31 billion in 2025.To date, Moore Threads has filed over 2,000 intellectual property applications, including more than 1,700 invention patent applications, building a solid intellectual property moat.

“We chose the difficult but right path at the start of our entrepreneurial journey. A full-function GPU should not only serve a single application scenario, but also empower a vast array of industries,"said Zhou Yuan, Co-founder and Chief Operating Officer of Moore Threads.In terms of computing infrastructure, the company has already deployed a ten-thousand-card cluster and is currently developing a one-hundred-thousand-card cluster, delivering robust computing power for large-model training.Notably, Moore Threads has pursued an ecosystem strategy that balances compatibility with independent innovation. Its self-developed MUSA architecture offers deep compatibility with the CUDA ecosystem, from hardware to the software stack, significantly reducing migration costs for developers.

"With many new models being launched recently, we are basically able to achieve Day-0 adaptation. In particular, with the recently released DeepSeek V4, we hadn't even adapted it ourselves yet when our clients officially announced that adaptation was complete,"Zhou Yuan revealed. The Moore Threads Academy now covers over 200 schools and has attracted 450,000 developers, with ecosystem development entering a positive feedback phase.

In the storage sector, Biwin Storage ushered in a period of explosive performance growth.The company achieved operating revenue of RMB 11.302 billion in 2025, up over 60% year-on-year; net profit attributable to shareholders was approximately RMB 853 million, a year-on-year increase of 429.07%, marking a profit peak since its listing.In the first quarter of 2026, the company sustained its high-growth momentum, with operating revenue reaching RMB 6.814 billion and net profit attributable to shareholders approaching RMB 2.9 billion.

"In recent years, the company has seen particularly prominent growth in emerging AI applications," said He Han, CEO of Biwin Storage. The company's products have secured a place in the supply chains of leading domestic and international clients including Google, Meta, Alibaba, and Rokid.In the automotive-grade storage segment, Biwin Storage has been included in the whitelist for the first batch of chip products certified by the State Administration for Market Regulation, making it one of only two storage manufacturers on that whitelist.

Biwin Storage has proactively deployed advanced wafer-level packaging technologies that help build an integrated AI storage, computing, and transmission packaging platform."Its value is not yet fully reflected in our current performance. We hope that through our efforts in 2026 and 2027, we can translate this strategic layout into tangible results,"said He Han. "Given the period of time for fab construction, our design orders have already surged ahead of full-scale mass production."

Huafeng Technology, a long-established manufacturer of optoelectronic connectors, is thriving in the AI era with renewed vitality.As China's first military connector manufacturer, the company has led the development of nearly 100 international, national, and industry standards, and holds over 600 patents."Reliable connection comes from Huafeng" is the company's tenet, and its products are widely applied in AI computing, new energy vehicles, humanoid robots, and other fields.

In terms of technological deployment, Huafeng Technology has maintained a forward-looking R&D pace."What we are truly working on now is 448G, while maintaining compatibility with 224G.Beyond PCIe 6.0, we are conducting research into PCIe 7.0 standard.Our technological research level has advanced to PCIe 7.0, but our products are currently still at PCIe 6.0.We recommend PCIe 6.0 to our customers, while making it clear that we will upgrade to PCIe 7.0 in the future,"said Liu Taiguo, CEO of Huafeng Technology.Speaking of localization progress, Liu Taiguo stated, "For storage below 56G, we have basically achieved full domestic production.For 112G and 224G, 112G is currently the mainstream, with perhaps 40% to 50% localization rate.In fact, China's chip computing power does not have much demand for 224G—domestic computing capabilities still lag behind the demand—so full domestic substitution for 224G has yet to be achieved.The core issue here is not really about the technology itself, but rather about confidence; there is still some concern about domestically produced products.When everyone works together in this field, confidence may grow stronger.From a technological readiness perspective, Huafeng Technology is fully prepared."

Taclink Optoelectronics, which has been deeply engaged in the optical communications field for nearly 30 years, has successfully seized the historic opportunity presented by AI computing interconnectivity."There is a saying in the optical communications industry, 'Don't just stand there—stand in the light.'For nearly three decades, Taclink Optoelectronics has been in the light and continues to carry it forward,"said Qu Jianping, Chairman of Taclink Optoelectronics, using an industry catchphrase to sum up the company's positioning.In 2025, the company's data communications business saw a significant increase in its share of total revenue, with DCI data interconnect-related products beginning to accelerate in volume, achieving a strategic transformation from a telecom-market-oriented approach to an equivalent approach covering both telecom and data communications markets.

To meet the strong demand from overseas customers, Taclink Optoelectronics established a production base in Thailand."Equipment has started to arrive in batches, and we expect to begin production in Q2, bringing limited capacity online initially. The main production ramp-up and most of the capacity will follow in the second half of the year.The base in Thailand primarily serves overseas markets, especially North America,"Qu Jianping explained.On the technology front, the company is focusing on three technology routes for 1.6T optical modules—EML, silicon photonics, and thin-film lithium niobate—as well as next-generation DCI data interconnect products. Meanwhile, it is developing high-dimensional, large-port OCS products such as 32×32 based on silicon-based optical waveguides, whose greatest advantage is low latency, making them highly suitable for meeting AI cluster requirements.

Emerging Sectors Unleash Long-Term Growth Potential as AI Momentum Continues

Executives at the conference unanimously agreed that the AI industry is still in its early stages of development and will maintain a high level of prosperity over the next three to five years.As AI rapidly permeates from training to inference, and with the rise of emerging applications such as embodied intelligence and humanoid robots, all segments of computing, storage, and interconnection will experience sustained demand upgrades, opening up broad growth space for companies across the industrial chain.

The explosive growth in AI demand is reshaping the demand structure of the semiconductor industry.He Han pointed out,"From a macro perspective, AI has a fundamental impact on demand for storage, computing, and transmission.For example, in the storage sector, mobile phones, PCs, and servers each used to account for 30% of demand, with all other categories making up the remaining 10%.But after the impact of AI became evident, this ratio changed significantly.The share of storage demand from servers and data centers rose from 30% to about 50% this year, and may reach 60% to 70% next year.More importantly, key players in the industrial chain have extended their assessment of industry prosperity from the short term to the long term.Our production capacity for 2026 has been fully subscribed, and most available capacity up for sale is allocated to 2027.

In addition, many large-scale computing providers, CSP cloud vendors, and memory manufacturers are signing three-to-five-year long-term contracts and are willing to pay substantial deposits to secure supply for the next three to five years.It is unprecedented in the storage sector,"said He Han. "From interviews with NVIDIA or some major U.S. CSPs, it can be seen that their outlook on industry prosperity extends to 2030.For industrial chainparticipants like us, it is relatively certain that we will see a high level of prosperity across the entire chain at least through 2026 and 2027."

The cutting-edge Chiplet packaging technology has emerged as a critical solutionto break through node scalinglimitations and enhance system performance,bringing new development opportunities to the industrial chain.Zhou Yuan stated,"Industrial chain collaboration is very important for domestic chip manufacturers, especially given that China's advanced process production is restricted. We must inevitably achieve breakthroughs in advanced packaging processes,including 2.5D and 3D packaging processes and Chiplet technology. From Moore Threads' perspective, we have always had a dedicated team working in this area, and corresponding chips are about to be launched."

He Han further noted, "Chiplet technology will blur the boundaries between computing power, storage, and interconnection. There is currently a hot direction in the industry, namely CPO.In fact, a co-packaged optics system integrates computing power chips, storage, and optical interconnects within a single chip packaged via Chiplet technology. This represents one of the most highly integrated optoelectronic packaging architecturesin the current industry landscape and also points to the future direction of the sector."

Embodied intelligence and humanoid robots are widely regarded as the next major application scenario for AI.Zhou Yuan stated,"We firmly believe that the next phase will trend toward embodied AI and physical world models.The future development of artificial intelligence will certainly involve cloud-based training and edge-side inference, but at the individual terminal level, every user will also need their own dedicated mobile AI platform."This will further expand computing power, storage, and interconnectiondemand.Huafeng Technology's high-speed connector products have already been applied in the humanoid robotics, and as the number of robot joints and the level of intelligence increase, demand for high-speed, highly reliable connections will grow substantially.

Collaborative innovation across the industrial chain has become an essential path for the development of China's hard-tech AI sector.Executives at the conference unanimously expressed that for China's AI industry to achieve global leadership, it cannot rely on the efforts of a single company in isolation; rather, upstream and downstream enterprises need to strengthen cooperation.Zhou Yuan stated,"We should be more proactive in reaching out to our upstream and downstream industry partners, by engaging in early-stage discussions and collaborative R&D planning, so as to ensure better delivery outcomes."He Han also noted,"The large-scale production of domestic computing power is an opportunity bestowed upon us by history. We must seize this opportunity to build a more competitive advanced packaging industrial chain."

Looking ahead, China's hard-tech AI sector is in its best period of development.The vast domestic market, a complete industrial chain, sustained policy support, and capital market assistance all provide a solid foundation for enterprise growth.Going forward, only by upholding a long-term perspective, continuously increasing R&D investment, and strengthening collaborative innovation across the industrial chain can enterprises truly propel China's AI industry into the global first tier and inject strong momentum into high-quality economic development.