According to a research report, the investment case for domestic computing power is becoming increasingly clear, and the financial performance of leading companies in the sector is expected to accelerate.
On June 30, 2026, Meituan released the first trillion-parameter model fully trained and inferred on a domestic 50,000-card computing power cluster. This marks an upgrade in domestic computing system capabilities from inference to full-scale training. Furthermore, the official release of DeepSeekV4 in mid-July will introduce a "peak and off-peak pricing" mechanism, which will double the API call price during peak hours, indicating that supply constraints on domestic computing power are intensifying. The report suggests that the clarity of orders for domestic computing power has significantly improved at this stage. Design companies with priority access to customer orders and production capacity are expected to be the first to benefit. The outlook remains positive for the entire domestic computing power industry chain, anticipating volume growth opportunities from scarce advanced process technology to a diverse range of design firms and super nodes. Simultaneously, advanced processes, advanced packaging, advanced memory, and supporting industrial chains are expected to experience strong growth momentum.
Meituan's Release of First Domestically Trained Trillion-Parameter Model
According to Meituan's official WeChat account, on June 30, 2026, Meituan officially released its new-generation trillion-parameter large model, LongCat-2.0. As the industry's first trillion-parameter model (total parameters 1.6T) to complete full-process training and inference on a domestic 50,000-card computing power cluster, LongCat-2.0 was pre-trained from scratch and natively supports a 1M ultra-long context. Its pre-training data scale exceeds 30T tokens. The development gradually addressed fundamental challenges such as operator adaptation, communication optimization, and distributed stability, ultimately achieving a steady-state daily throughput of over 1T tokens/day and enabling stable training of the trillion-parameter MoE model on domestic computing hardware. This signifies a comprehensive upgrade in domestic computing system capabilities from inference to training, demonstrating that domestic companies now possess the ability to conduct large-scale model training on domestic computing clusters.
DeepSeekV4 to Implement Peak Pricing, Tightening Supply Constraints
According to an official DeepSeek email, the official version of DeepSeekV4 is scheduled for release in mid-July 2026. To enhance service stability, this version will introduce a "peak and off-peak pricing" mechanism. The company announced that API call prices during peak hours will double compared to off-peak periods. This also indicates that supply constraints on domestic computing power are still tightening. Since 2026, the domestic computing power shortage has become increasingly severe. Against this backdrop, the pace of domestic computing power cards being adopted by major clients has noticeably accelerated.
Financial Statements of Domestic Computing Power Firms Set to Improve
Capital expenditures from tech giants continue to exceed expectations as they increase investments to accelerate the monetization of AI services. The chip iteration cycle is speeding up, driving synchronized acceleration in demand for computing clusters, network interconnects, and AI infrastructure. The report posits that the clarity of order expectations for domestic computing power has significantly strengthened at this stage, with supply chain inventory preparation accelerating. This is expected to lead to notable improvements in the financial statements of related companies, further solidifying performance expectations for the second half of the year. Given the rigid constraints on both supply and demand for domestic computing power this year, deliverable and implementable projects hold high priority. As domestic advanced process capacity gradually ramps up, design companies with priority access to customer orders and production capacity allocation are anticipated to be the first beneficiaries.
Risk Factors
Potential risks include a global macroeconomic downturn; changes in the international political environment and escalating trade frictions; weaker-than-expected downstream demand; slower-than-expected AI innovation; delays in AI commercialization; slower-than-expected progress in domestic substitution; slower-than-expected expansion of domestic wafer fabs; slower-than-expected development of advanced process technologies; intensified competition among downstream manufacturers; risks of significant raw material price increases due to inflation; escalation of U.S. sanctions against China; significant currency fluctuations; and slower-than-expected technological iteration.
Investment Strategy
The report maintains a comprehensively positive outlook on the domestic computing power industry chain. It is projected that volume growth opportunities will emerge across the board, from scarce advanced process capabilities to a flourishing array of design companies and super nodes. Concurrently, advanced processes, advanced packaging, advanced memory, and supporting industrial chains are poised for robust growth momentum. The suggested focus areas are: 1) AI chip design companies. 2) Wafer foundry and advanced packaging. 3) Other related sectors.