This report does not constitute a rating action.

Build it and they will come. This is the likeliest narrative for the rapid expansion of internet data centers in China. S&P Global Ratings believes the major operators have sufficient demand to warrant their capacity additions. For smaller and less competitive players, however, we see a potential asset bubble.

In our view, the push for AI development in the country's tech sector will significantly increase demand for computing power. Budgets for internet data centers (IDCs) are on the rise for cloud service providers, social media platforms, and other internet service providers. We estimate that Alibaba Group and Tencent Holdings, which are leading this charge, will in total ramp up annual spending in this area to above RMB200 billion (US$28 billion) in 2025-2026, from about RMB50 billion (US$7 billion) in 2023.

This is a significant opportunity for the biggest wholesale IDC operators such as GDS Holdings Ltd., VNET Group Inc., and WinTriX DC Group (familiarly known as Chindata). We anticipate strong revenues as well as enhanced EBITDA and cash flow visibility. The large players have already booked much of the capacity they are now building, and they can meet the service demands of big tech players. The smaller players could get pushed further to the margins.

Chart 1

For Top Players, The New Capacity Will Be Utilized

Chinese enterprises have been increasingly integrating large language models and generative AI into their operations. The success of AI-focused companies such as DeepSeek has accelerated this trend. As a result, we expect tech companies to invest heavily in AI infrastructure, including AI chips, data centers, and talent acquisition for AI software development (see China's DeepSeek Triggers Cycle Of Disruption," published on RatingsDirect on Feb. 10, 2025).

This trend in turn will drive demand for internet data centers. As such, we don't see the major expansion plans as particularly risky. This is because much of this capacity is pre-booked. For example, in a recent earnings call, VNET said it will deliver 400 megawatts (MW)-450MW of additional wholesale IDC capacity in 2025. That's a big jump from the 486MW in service at end-2024. But it has obtained orders for 83% of this capacity.

GDS, as another example, has reported a commitment rate of 91.9% for area in service and a pre-committed rate of 64.1% for area under construction in 2024.

The ability to expand quickly will help the leading players solidify market share (see chart 2). We estimate the new capacity could ramp up as quickly as six months upon delivery, compared with the typical 12-24 months historically, as customers are eager to speed up the integration of large language model and generative AIs into their operations to stay ahead of the AI race.

Chart 2

Strict Requirements Will Rule Out Most Smaller And New Players

We believe hyperscalers--such as Tencent, Alibaba and Huawei Technologies--are looking for wholesale IDC operators that have the capability to rapidly deliver large-scale data centers with low operating costs and high reliability. Key factors include anticipating hyperscale needs through pre-committed land acquisitions in prime locations with sufficient energy and water resources, network connectivity, and near compute demand.

Hyperscalers generally have a long list of operational criteria through service-level agreements (SLAs) with IDC operators. These criteria include guaranteed service uptime, support response and resolution times, performance metrics (such as latency and processing speed), power density and cooling solutions, other facility management metrics (such as security). Violation of SLAs usually results in significant financial penalties for IDC operators.

In addition, the operators need significant liquidity and access to various financing channels to fund the rapid development of new datacenters. These strict requirements rule out most smaller players from securing commitments from hyperscalers, which account for the lion’s share of AI workloads and hence datacenter demand. The smaller operators also lack a track record of building and running large-scale IDCs that they can showcase to the hyperscalers for contract bidding.

Location Is A Deciding Factor

Wholesale IDC demand is uneven across the country. Beijing, Shanghai and the surrounding region host the majority of China’s data center capacity and have the highest utilization rates, often around 70% or above. These cities benefit from close proximity to compute demand and limited access to key resources such as land, electricity, and water, thus reducing the supply of suitable data centers.

Demand in these regions should increase, due to the growth of AI inference workloads. Unlike training large language models, AI inferencing is less resource intensive and requires low latency--i.e., the time it takes for data to travel between server and end-user.

Chart 3

China’s leading IDCs players generally have solid presences in Beijing and other higher demand region. We believe it's challenging for smaller players and newcomers to expand in these areas due to land scarcity and strict energy requirements.

Leading IDCs Typically Have Much Lower Operating Costs

Data centers consume vast amount of electricity. Typically, electricity costs account for at least 40% to 50% of the total operating costs of data centers, with AI datacenters consuming a few times more electricity than traditional datacenters. Improving energy efficiency lowers operating costs for hyperscalers.

Moreover, China regulators are increasing energy efficiency requirements for data centers to reduce their environmental impact and energy consumption. Hyperscalers may also have sustainability goals and net-zero emissions targets that IDC operators will need to meet.

Leading IDC operators typically have much higher energy efficiency ratios relative to other datacenter operators, as measured by the ratio of power usage effectiveness (see chart 4). This is defined as the ratio of the total amount of power entering a data center facility to the power consumed by the IT equipment within it. Using this metric, VNET, GDS, and Chindata consume about 16% less energy than a typical data center in China.

Chart 4

For Smaller Players, There May Be A Supply Glut

Service fees and utilization rates across China have been declining since 2020 due to significant datacenter capacity additions, likely due to favorable government policies to incentivize datacenter construction and a desire to capture the growing AI demand.

However, much of this new supply likely does not meet the needs of major internet firms. Therefore, we are seeing a growing divergence in capacity utilization between the biggest players and the rest of the market (see chart 5). Leading players typically have utilization rates of 70%-75% compared with other datacenter operators that have an average utilization of about 40%-50%, based on our estimation. Moreover, much of the unutilized capacity for leading players are committed, meaning that the utilization rates for completed datacenters will likely exceed 90% within a year.

Chart 5

U.S. Chip Restrictions Remain A Key Risk For IDC Operators

The supply of advanced chips is a key risk for new datacenter demand in China. A lack of advanced chip supply could stifle internet companies’ ability to build out their AI infrastructure and hence reduce their demand for datacenters.

U.S. restrictions on exports of advanced chips to China continue to tighten, with new licensing requirements for Nvidia H20 chips. Such chips were modified specifically to meet previous U.S. export control requirements when exporting to China. Some Chinese firms have increased the use of domestically produced AI chips. However, It remains uncertain whether Chinese AI chip production can meet demand, especially given the limitations on acquiring advanced semiconductor equipment.

Major China tech firms have reportedly significantly increased their stock of H20 chips, limiting the risk on AI demand over the next year. However, chip availability remains a risk beyond that.

Related Research

• Tencent Can Absorb Higher AI Spending, March 24, 2025
• Alibaba To Maintain Strong Cash Amid AI Investments, Feb. 26, 2025
• China's DeepSeek Triggers Cycle Of Disruption, Feb. 10, 2025

Primary Contact:	Aras Poon, Hong Kong 852-2532-8069; aras.poon@spglobal.com
Secondary Contact:	Clifford Waits Kurz, CFA, Hong Kong 852-2533-3534; clifford.kurz@spglobal.com
Research Assistant:	Jenny Chan, Hong Kong 852-25333570; jenny.chan@spglobal.com