Sinolink: Domestic computing power chain is expected to usher in a golden age. We remain optimistic about the high prosperity of the computing power chain market.

Sinolink released a research report stating that the domestic computing power chain is the core engine of the digital economy, from independently developed underlying chips to globally leading key components such as optical modules and PCBs, to liquid cooling technology and ten thousand card-level computing power clusters. In the current high prosperity of the overseas computing power chain, the domestic computing power chain is also expected to usher in a golden age, continuing to be optimistic about the high prosperity of the computing power chain market. With the new products from TSMC Rubin, Google/Amazon, and other ASIC manufacturers driving growth in Q2, the outlook remains strong, particularly for AI industry chain hardware core companies. Currently, many AI-PCB companies have strong orders, are operating at full capacity, and are ramping up production, with high performance growth expected to continue. Key points from Sinolink are as follows: The focal point of the current domestic computing power market - domestic model capabilities exceed critical point 1) The essence of the current domestic computing power chain lies in the breakthrough of domestic large-scale model capabilities, with domestic models such as DeepSeek-V4, GLM-5, and other domestic model Agent capabilities advancing rapidly, supporting simple coding and Agent applications, thus truly accelerating domestic computing power demand. 2) On the other hand, domestic large-scale models have algorithm optimization engineering advantages, with overall reasoning power costs lower than overseas, with much lower API prices for top models in China compared to top models overseas. This high cost-performance ratio allows domestic large-scale models to successfully compete for market share of previous generation large models overseas, benefiting from the token bounty. Domestic model ARR and Token volume growing exponentially, computing power shortage intensifying Currently, domestic model manufacturers have seen significant growth in Token and ARR, with Minimax ARR surpassing $150 million; and KNOWLEDGE ATLASAPI ARR crossing $250 million, even after three price hikes, demand still exceeds supply, with a 400% increase in token calls in Q1 2026, translating commercial demand into rapid release of computing power. With strong demand on both supply and demand sides, the computing power industry chain in 2026 has substantially entered a stage of full-chain inflation, with industry prosperity shifting from core chips to AIDC, cloud and computing power services such as CPU/GPU facing capacity bottlenecks and rising rental prices, and major cloud computing companies such as Alibaba, Baidu, etc., starting multiple rounds of price hikes. The continuous price increases in multiple links strongly prove the explosion of real computing power demand and extreme supply scarcity. The golden age of domestic computing power is beginning, where usable computing power in itself is the most core asset under supply scarcity Domestic Chips: Major manufacturers speeding up adaptation, demand surge + supply improvement. After the release of Deepseek-V4, Ascend, CambriconDay 0, the subsequent Hygon Information Technology, Moore Threads, Muxi Shares, Baidu Kunlun Core and other 10 domestic chip manufacturers have officially announced support for Deepseek-V4, with domestic model manufacturers widely adapting to multiple domestic chip manufacturers on the reasoning side; based on Cambricon's first quarter report, prepayments and contract liabilities both saw a significant increase, confirming a demand explosion and supply improvement signal for domestic chips. Domestic Cabinets: Shipments of supernodes leading to improved ODM assembly manufacturers layout/profit margins. Units shipped transition from individual white box servers to full cabinets and even entire Pod-level systems, as complexity of deliveries and scarcity of qualified suppliers increase, leading ODMs have the actual capacity to shift premium charges to customers, with profit margins expected to shift to higher ranges. Computing Lease/Cloud/IDC: 1) Cloud: Model calls driving cloud computing demands to surge, token calls driving higher cloud prices for major players, with third-party clouds following suit; 2) Computing lease: Usable computing power in itself is the most scarce core asset, with a commercial ROI for computing leasing companies accelerating under the double affirmation of bursting demand and performance; 3) AIDC: Capital expenditures of major domestic and overseas companies have significantly increased, with the expectation of entering a phase of rising prices post abundant computing power supply. Overall Bullish Outlook on AI Core Hardware Industry Chain Google TPU rapid iterations, Anthropic revenue explosive growth, TSMCs advanced processes used for AI chips filled to capacity, ramping up production and GPU rental prices rising. Strong demand is anticipated in the short and medium term for AI. With the haulage of new products from TSMC Rubin, Google/Amazon, and other ASIC manufacturers, Q2 shows continued strong growth, maintaining a positive outlook on core hardware companies in the AI industry chain. Many AI-PCB companies are currently experiencing strong orders, running at full capacity, and are actively expanding production with high growth performance expected to continue. Demand for AI copper plates is also high, with leading mainland manufacturers expected to benefit positively due to slow expansion of overseas copper plates. The strong demand for AI copper plates/PCBs also drives demand for ancillary equipment (drilling machines, direct writing lithography equipment, drill bits, etc.) and upstream electronic fabric/copper foil. Optical Communication: Zhongji Innolight has ample material reserves, the first to begin production of 1.6T & silicon optical modules, with strong revenue certainty. Eoptolink Technology Inc., affected by material shortages and production pace, 1Q performance fell short of expectations, but the balance sheet shows positive expansion, with an expectation for steeper revenue and profit growth starting from 2Q. Suzhou TFC Optical Communications net profit is lower than market expectations, due to volume increases in active business, reduction in 800G volume, increased R&D and management expenses, but remains a core company in the CPO sector. Bochuang 1Q26 experienced reduced business from other sources, affecting production and confirmed income growth slightly below market expectations, but profitability exceeded expectations again, with the company's acquisition of remaining equity progressing faster than expected. Switches: AI computing power demands are driving data center switches to upgrade to 800G/1.6T high speed rates. Server Liquid Cooling: CSP manufacturers are fully embracing liquid cooling technology with server power upgrades driving increased demand for liquid cooling. Under the AI wave, there is great potential for liquid cooling. Nvidia's GB200 and GB300 series, Meta ASIC series cabinets adopt liquid cooling design, with a doubling in fast connector usage in GB300 cabinets, resulting in higher value per cabinet for liquid cooling. Domestic AI chips usually have higher power consumption at equivalent computing power, hence have more stringent requirements for heat dissipation efficiency. Connectors: Connectors represent about 3-5% of the value in communication equipment. Under the drive of AI computing power, there is a surge in high-density data center connectivity demands, with MPO and AEC becoming a core increment track. Fiberglass New Materials: Continuously bullish on AI electronic fabric, AI copper foil, and advanced packaging equipment and other high prosperous sub-sectors undergoing accelerated domestic substitutes, with expectations of continuous growth in performance quarter-to-quarter and annually. Risk Warning: Risks of increased industry competition, risks of technological research and development progress falling short of expectations, risks of cyclical fluctuations in downstream capital expenditure in specific industries, risks of weak recovery in end demand, risks of further external sanctions escalation, risks of slower-than-expected growth in IoT connections, risks of continued material shortages and price fluctuations of raw materials, risks of policy implementation falling short of expectations, risks of intense market competition.