The AI computing power industry chain is entering a notably dense period of catalysts for the year. The 2026 China Computing Power Conference is scheduled to take place from September 11 to 13 in Langfang City, Hebei Province. Under the theme "Weaving the Computing Power Network Together, Smartly Ushering in a New Future," the event will establish a comprehensive national platform for industrial exchange, innovation empowerment, supply-demand matching, and ecosystem co-construction, with the goal of driving high-quality, intensive, and inclusive growth of China's computing power sector.

At the end of August, Nvidia delivered a "perfect score" in its fiscal second-quarter earnings report for 2027. Quarterly revenue surged 106% year-over-year to $96.2 billion, with data center revenue climbing 117% to $89 billion, accounting for approximately 92.5% of total revenue. Even more electrifying for the market, Nvidia provided its first-ever mid-to-long-term outlook for fiscal 2028, projecting approximately 70% revenue growth. CEO Jensen Huang noted that Nvidia had never issued guidance a year in advance before, adding, "Now we have stronger visibility." Meanwhile, the company's backlog of unfinished orders from cloud service providers has exceeded $2 trillion.

The significance of this earnings report extends beyond quarterly metrics—it has raised the ceiling on global AI capital expenditure expectations. According to the latest AI server industry research from TrendForce, driven by surging demand for AI infrastructure, total capital expenditures of the world's nine major cloud service providers (CSPs) are projected to grow approximately 90% year-over-year in 2026, with an additional ~50% growth expected in 2027.

Where to Begin

On August 22, the 2026 Green Computing Power (Artificial Intelligence) Conference was held in Hohhot, Inner Mongolia. During the event, Hohhot and Ulanqab signed 13 projects with leading companies such as Volcano Engine and Cambricon, with total investment reaching RMB 136.1 billion. The projects span chip testing, equipment manufacturing, computing centers, and token factories. At the opening ceremony, Ao Li, Vice President of the China Academy of Information and Communications Technology (CAICT), released the "Green Computing Power Development Research Report (2026)." The report notes that a nationally integrated computing power network is accelerating. Driven by large models, the token economy is growing rapidly—daily token calls in China surged from approximately 100 billion in early 2024 to about 140 trillion by March 2026, a more than 1,000-fold increase.

On the policy front, seven government departments, including the Cyberspace Administration of China, recently jointly issued the "Implementation Plan for Promoting Coordinated Digital and Green Transformation and Development (2026-2030)." The plan emphasizes strengthening green design for computing facilities, supporting innovative technologies such as liquid cooling, waste heat recovery, and low-power chips, exploring 800V high-voltage DC power supply architectures, and promoting deployment of single-cabinet capacities exceeding 100 kilowatts. It also calls for optimizing green operations models, leveraging key technologies like digital twin simulation and multimodal sensing, and deepening AI algorithm applications in green operations to build an intelligent management system covering the entire lifecycle of computing facilities—enabling real-time energy monitoring, forecasting, dynamic adjustment, and intelligent data center operations.

Notably, AI large models continue to iterate rapidly, and AI applications are expanding from chat and search to complex scenarios such as intelligent agents, autonomous driving, and embodied intelligence. An increasingly clear trend is emerging: the AI industry is shifting from "model competition" toward "computing power competition." Within this shift, computing power leasing is becoming a critical bridge connecting infrastructure with AI applications.

On one hand, global AI-driven token consumption continues to rise, rapidly unleashing demand for computing resources. On the other, high-end GPU supply remains tight, with enormous investments needed for data center construction, GPU procurement, and network and power infrastructure. For many enterprises and institutions, the capital pressure and technical barriers of building in-house computing centers are growing. As a result, leasing computing resources is emerging as a more flexible option. Concurrently, recent reports indicate widespread price increases for Nvidia GPU leasing, further validating the global tight supply situation from the supply-demand perspective.

Why This Window Matters for Investors

China Securities Co., Ltd. (CSC) released a research report stating that AI computing power has fully entered a new cycle "dominated by inference." Application-side high token density scenarios are driving explosive demand growth, while the supply side shows a divergent pattern: high-end resources remain persistently tight, while mid-to-low-end segments face structural saturation. Geopolitical restrictions and rising costs across the entire supply chain are significantly raising the barriers for high-end computing capacity.

Computing power leasing is accelerating with both volume and price rising, ushering in a golden period of robust growth. The market size could surpass RMB 260 billion in 2026, with the industry transitioning from "extensive expansion" to "high-end premium and refined operations." The Matthew effect is concentrating resources among top players. Meanwhile, business models are evolving from bare-metal leasing to token revenue-sharing and integrated token factory services, with long-term contracts and financial leverage further reshaping industry certainty and broadening growth curves.

Bohai Securities pointed out that leading domestic large model companies are deepening chip-model collaborative development, using domestic AI chips as the core hardware foundation for model inference. With domestic cloud providers' capital expenditures recovering, the share of domestic computing power in total investment is expected to gradually rise. Computing power leasing and IDC providers are poised to benefit substantially from new build-outs and expansions of intelligent computing centers.

Great Wall Securities previously argued that the long-term growth logic for the AI computing power industry remains clear. Dawning Information Industry's domestic computing power solutions are accelerating deployment, with projects like the Hefei intelligent computing cluster validating industrial practices. Liquid cooling technology is expected to see rising penetration alongside high-density computing construction, and the firm is bullish on opportunities across the domestic computing power and liquid cooling supply chains.

Key Concept Stocks to Watch

Zhipu AI (02513): In early September, CMB International released a report stating that Zhipu's annualized recurring revenue (ARR) growth trajectory is steady and its outlook is strong, with newly disclosed user scale and engagement metrics reinforcing a positive view. Driven by stronger ARR growth, the firm raised its 2025-2028 revenue CAGR forecast from 199% to 275%, lifted its target price from HK$1,503.9 to HK$1,985, and maintained a "Buy" rating. CMB International believes the company has achieved an effective balance among model iteration pace, intelligence level, and per-task cost, allowing it to maintain its position on the Pareto frontier of the LLM industry and providing ample room for long-term monetization. The report notes that Zhipu completed six GLM iterations in 11 months, with its intelligence index rising from 32 to 60. MaaS platform registered users grew 144% to 7.4 million, paying daily active users increased 603%, token calls have grown more than 40-fold year-to-date, and average API selling prices rose approximately 101%. As of August 2026, ARR from its open platform and API reached $1.6 billion, up from $1 billion in early July, with management guiding toward $2.4 billion by year-end. Deployment of a 100,000-chip cluster and inference stack rebuild reduced per-token costs by approximately 80% year-to-date, with the compute multiplier up 14x year-over-year. Open platform gross margin reached 24.6% in H1 2026. CMB International raised its 2025-2028 revenue CAGR forecast from 199% to 275% and projects total 2028 revenue of RMB 38.1 billion.

Tianshu Zhixin (09903): The company's interim results show that, benefiting from concentrated demand for cloud-based AI inference computing, its Zhikai series achieved breakthrough growth. First-half revenue reached approximately RMB 654 million, surging 651.8% year-over-year, with gross profit of approximately RMB 233 million and a gross margin of 35.6%, making it the core growth driver. The Tiange series, targeting high-end training scenarios, maintained steady growth with first-half revenue of approximately RMB 262 million, up 38% year-over-year.

Kingsoft Cloud (03896): The company has been continuously strengthening its AI cloud service capabilities, upgrading its traditional cloud computing business toward AI infrastructure services. As one of the earliest independent cloud providers in China, Kingsoft Cloud has fully embraced AI over the past two years: leveraging the natural scenarios within the Xiaomi and Kingsoft ecosystems, AI-related revenue continues to scale, with intelligent computing cloud services becoming the primary growth engine. The company is also increasing AI infrastructure investment, reinforcing its full-stack capabilities from IaaS to MaaS. As enterprise demand for AI applications grows, companies with cloud platform capabilities, customer resources, and ecosystem advantages are well-positioned to capture new growth in the AI industrialization process.

Yuegangwan Zhisuan (01396): The company reported interim results for the six months ended June 30, 2026, showing explosive tenfold revenue growth: revenue reached approximately RMB 2.56 billion, surging 1,020.9% year-over-year, a major leap to a new revenue milestone with robust momentum. Profitability achieved a dramatic jump: interim profit reached approximately RMB 281 million—five times last year's total annual net profit (approximately RMB 56 million)—marking record-high profitability. Its computing power scale remains at the industry's forefront: FP16 dense computing power exceeds 50,000 PFLOPS, securing a leading industry position and underpinning the foundation for a 100,000-card-scale super computing cluster.