China's Internet Giants Collectively Expand AI Capital Expenditure: Six Key Beneficiary Sectors in the Computing Infrastructure Supply Chain

Core Background: A Critical Shift from Expectations to Delivery

Since 2025, China's internet giants have collectively expanded their capital expenditure on AI infrastructure. During Tencent's Q1 2026 earnings call, the company explicitly stated that demand for AI-related services continues to grow, with capital spending set to increase year-over-year, particularly with accelerated investment in the second half. Alibaba was even more direct: the return on investment from heavy AI data center spending is "highly certain," and given the massive scale of computing center construction, capital expenditure may exceed the initially announced three-year plan of over 300 billion yuan.

This collective expansion in capital spending sends a critical signal — China's AI computing infrastructure has officially transitioned from the "promise" phase of expectations to the "delivery" phase of execution. The real money being invested by tech giants will transmit through the supply chain logic of hardware equipment and supporting systems, creating substantive order flow and earnings momentum for related industries.

The Transmission Mechanism of AI Capital Expenditure

Capital Expenditure (CapEx) refers to spending on purchasing, maintaining, or upgrading fixed assets. In the AI domain, this primarily includes procurement of GPU/TPU computing chips, data center civil engineering and mechanical/electrical works, and network equipment deployment. AI CapEx from internet giants carries a significant multiplier effect: every 1 yuan of direct investment typically drives 3-5 yuan of associated output across the upstream and downstream supply chain. The transmission path works as follows: tech giants issue procurement orders → system integrators organize supply → component manufacturers (chips, optical modules, power supplies) expand production → upstream materials and manufacturing companies benefit. Therefore, the scale and pace of tech giants' capital spending is the core leading indicator for gauging the prosperity of the entire AI hardware supply chain.

In-Depth Analysis of Six Beneficiary Sectors

AI Data Center Construction and Operations

Tech giants have significantly raised their AI data center construction budgets, with full-year cloud vendor IDC demand expected to more than triple compared to 2025. Currently, domestic daily Token call volumes have surged over 1,000-fold, and the computing power leasing market is heading toward the hundred-billion-yuan scale. This means two types of companies will directly benefit: first, enterprises focused on AI data center construction and operations will see rapid growth in orders and earnings; second, listed companies involved in computing power leasing will directly benefit from the explosive growth in computing demand.

The Relationship Between Token Call Volume and Computing Demand

A Token is the basic unit by which large language models process text — one Chinese character typically corresponds to 1-2 Tokens. When users interact with an AI model, every Token in both input and output requires computing power for inference calculations. A 1,000-fold surge in daily Token call volumes means that inference-side computing demand is growing exponentially. Unlike training, inference represents continuous computing consumption — model training only needs to happen once, but every user call requires real-time inference. This explains why computing demand continues to accelerate even as model architectures mature: expanding user bases and diversifying application scenarios jointly drive explosive demand for inference computing.

AI Transformation of IDCs

Traditional IDCs primarily provide basic services like server hosting and bandwidth leasing, but AI data centers are fundamentally different in architectural design. AI data centers need to support high-density GPU cluster deployment, with per-rack power increasing from the traditional 6-8kW to 30-50kW or higher; they require liquid cooling systems to handle the high thermal density of GPUs; they need 400G/800G high-speed networks to meet the communication requirements of distributed training; and they require large-scale high-performance storage systems to support data throughput. Consequently, the per-unit-area investment for AI data centers is 3-5 times that of traditional IDCs — one of the key reasons behind the sharp increase in tech giants' capital spending.

Domestic Large Models and AI Algorithm Companies

Token usage for domestic large models is growing exponentially. Increased AI computing investment by tech giants provides stronger computing support for large model companies, accelerating the commercialization of AI applications. From a supply chain logic perspective, computing power is the "fuel" for large model development. When fuel supply is abundant, model iteration speed and application deployment efficiency both improve significantly, creating an important development window for related large model and AI algorithm companies.

Computing Chips

Sustained growth in AI training and inference demand is pushing computing chip demand into a shortage state. Data shows that global semiconductor sales in March 2026 grew 79.2% year-over-year, with domestic sales growing 74.8% — staggering growth rates. As the core hardware of AI infrastructure, computing chip companies have extremely high earnings growth certainty, making this one of the most investment-worthy segments in the current supply chain.

The Division Between Training and Inference Chips

Computing chips can be categorized by purpose into training chips and inference chips. Training chips (such as NVIDIA H100/H200/B200) require extremely high floating-point computing power and large memory capacity for processing forward and backward propagation calculations on massive datasets; inference chips prioritize energy efficiency and low latency, needing to complete model inference quickly within limited power budgets. The current industry trend shows the training-to-inference computing ratio shifting from an early 8:2 to 3:7 or even 2:8, because the continuous inference demand after model deployment far exceeds one-time training requirements. Domestic computing chip companies (such as Huawei Ascend and Cambricon) are pushing forward on both fronts, seeking to establish their position in the domestic substitution wave.

Storage Chips and High-Speed Interconnects

The explosion of application scenarios like AI Agents and AI coding is driving rapid growth in Token call volumes, with storage chip demand continuing to climb. A telling example is ByteDance raising its capital expenditure due to rising memory chip costs, which indirectly confirms the tight supply-demand dynamics in storage chips. Additionally, AI training and inference create urgent demand for high-speed data transmission and efficient power management in computing clusters, making optical modules, liquid cooling, and high-efficiency power supplies noteworthy sub-sectors.

The Critical Role of Storage Chips in AI Scenarios

AI workloads place unprecedented demands on storage systems. During training, High Bandwidth Memory (HBM) is needed to feed data to GPUs, with HBM bandwidth reaching several times that of traditional DDR5 memory. During inference, the KV Cache (Key-Value Cache) mechanism requires substantial memory to store intermediate computation results — the larger the model parameters and the longer the context window, the more memory capacity is needed. For a GPT-4-class model, a single inference may require hundreds of GB of memory support. This is why companies like ByteDance are raising capital expenditure due to memory chip cost increases — HBM chips are currently supplied primarily by SK Hynix and Samsung, with capacity expansion far lagging behind demand growth.

Semiconductor Materials and Precision Manufacturing

The high prosperity of the semiconductor industry is transmitting upstream, with demand for photoresist, polishing slurry, and other semiconductor materials growing accordingly. Meanwhile, AI hardware equipment demands extremely high precision manufacturing standards, and companies involved in AI hardware manufacturing will see earnings improve as tech giants increase procurement volumes. This is a beneficiary direction that is easily overlooked by the market but carries strong certainty.

Vertical Applications and Domestic Substitution

AI vertical application scenarios across industries continue to expand, with tech giants' computing investments driving development in "AI+Energy," "AI+Manufacturing," "AI+Healthcare," and other fields. More notably, some domestic AI computing hardware currently relies on imports, and large-scale AI data center construction by tech giants will accelerate the domestic substitution process, creating enormous incremental market opportunities for domestic supply chain companies.

Technical Pathways and Market Potential for Domestic Substitution

In the AI computing supply chain, domestic substitution primarily involves three layers: the chip layer (GPU/NPU replacing NVIDIA products), the equipment layer (servers/switches replacing imported brands), and the materials layer (photoresist/electronic specialty gases replacing Japanese and Korean products). Due to U.S. chip export controls on China, high-end AI chips (H100 and above) cannot be directly exported to China, forcing domestic companies to accelerate independent R&D. Huawei's Ascend 910B/C series has entered tech giants' procurement lists, and while single-chip performance still lags behind NVIDIA's top products, the overall system efficiency gap is narrowing through cluster optimization and software adaptation. Estimates suggest that China's AI chip market could exceed 300 billion yuan by 2027, with the domestic substitution rate rising from the current sub-20% to over 40%.

Investment Logic and Risk Considerations

From an investment perspective, sustained attention should be paid to supply chain leaders with technological advantages, comprehensive capacity deployment, and quality customer resources. These companies, leveraging their technological moats and customer stickiness during the expansion of tech giants' AI capital spending, will capture greater market share and orders with stronger earnings growth certainty.

However, two types of risks warrant caution: first, companies with lagging technology that rely excessively on concept-driven speculation may be eliminated as industry competition intensifies; second, the pace of capital spending may fluctuate due to macroeconomic conditions, geopolitical factors, and other influences, requiring dynamic tracking of tech giants' actual investment progress.

Conclusion

The AI industry is currently at a critical inflection point from "technological breakthrough" to "scaled deployment." The collective increase in AI capital spending by internet giants is essentially a vote of confidence in AI's commercial prospects. This wave of infrastructure investment will profoundly reshape China's computing supply chain landscape — from chips and storage to data centers, from materials to applications — the entire supply chain will encounter structural growth opportunities.

Key Takeaways

• Internet giants including Tencent and Alibaba have collectively raised AI capital expenditure, with Alibaba's three-year investment potentially exceeding 300 billion yuan, marking China's AI computing infrastructure transition from expectations to delivery
• Full-year cloud vendor IDC demand is expected to more than triple versus 2025, with domestic daily Token call volumes surging over 1,000-fold and the computing leasing market heading toward the hundred-billion-yuan scale
• Global semiconductor sales grew 79.2% YoY in March 2026, with domestic growth at 74.8%, as computing chip demand enters a shortage state
• Six supply chain sectors stand to benefit: data center construction/operations, large model companies, computing chips, storage chips and high-speed interconnects, semiconductor materials, and vertical applications with domestic substitution
• Focus on technologically leading supply chain champions while remaining cautious of companies with lagging technology or excessive reliance on concept-driven speculation