China Data Center Demand Outlook Exceeds Expectations, Official Forecasts 800TWh Electricity Usage by 2030

Overview

This research report comments on the Bloomberg story regarding 'China planning to invest 2 trillion RMB in data centers over the next 5 years'. Goldman Sachs does not express an opinion on the likelihood of policy implementation but verifies the strong demand outlook for China's data center industry by comparing official data from the National Data Administration (NDA) and the National Energy Administration (NEA). The core conclusion is: official predictions for data center power consumption are significantly higher than Goldman's previous estimates, and the investment scale aligns with early plans, indicating the industry is in an acceleration phase. Based on this, the report reiterates positive ratings on some leading IDC providers.

Core views

Official demand forecasts significantly revised upward: The National Energy Administration (NEA) stated at the end of May 2026 that China's data center electricity consumption is expected to reach 800TWh by 2030. This figure implies a Compound Annual Growth Rate (CAGR) of 36% for 2025-2030, far exceeding Goldman Sachs' previous prediction of 20%. In terms of share, data center electricity usage as a proportion of total social electricity consumption will rise from 1.6% in 2025 to 6% in 2030. Additionally, considering the continued improvement in data center energy efficiency ratios (PUE), the growth rate of actual IT equipment power demand may be even faster than the growth rate of total electricity consumption. Reversal of China-US computing power energy consumption comparison: According to NEA forecasts, China's data center power consumption scale will surpass that of the US by 2030. The Goldman Sachs Sustain team previously estimated that US data center electricity consumption would be 726TWh in 2030 (CAGR of 22% from 2025-2030), while China's 800TWh target indicates that domestic AI and digital infrastructure construction is catching up and potentially surpassing at a steeper slope. Recent high-frequency data also corroborates this trend; national internet data service sector electricity consumption surged 44% year-on-year in January-April 2026. 2 trillion investment not a new surprise: Regarding market concerns about the 2 trillion yuan investment rumor, the report points out that this actually aligns with calculations released by the National Data Administration at the beginning of 2025 entitled "Guidelines on National Data Infrastructure Construction". At that time, NDA already estimated direct investments in data infrastructure over the next 5 years would be around 2 trillion yuan, covering network, computing, security facility construction and upgrades, as well as data circulation and utilization links. This indicates the current investment heat is more a continuation and implementation of established national strategy rather than short-term sentiment speculation.

Analysis framework

The research report adopts a 'Policy/Official Data Cross-Verification' analysis method. Facing media reports on massive investments, analysts did not blindly accept or refute but reviewed top-level design documents from the National Development and Reform Commission, Data Bureau, etc., one year ago to confirm the continuity of funding sources and scale. Meanwhile, they selected 'electricity consumption' as a hard indicator to measure real data center demand, tri-comparing the latest forecast values from the Energy Bureau with internal institutional models (GSe) and overseas benchmark markets (US Sustain Team) to quantify the conclusion that 'demand prospects are stronger than expected'. This analytical framework of determining demand via energy consumption and supply via policy effectively filters noise from pure capital expenditure narratives.

Methodology notes

• Industry/Industrial Analysis FrameworkVolume-Price Splitting

  Using electricity consumption volume as a core proxy variable for data center industry prosperity

  Data center industry revenue is determined by cabinet quantity (volume) and rental unit price (price), but rentals lag due to contract cycles. Electricity consumption is a real-time physical indicator reflecting server rack-up rates and computing load, and is more sensitive in capturing marginal changes in industry demand than financial revenue.

• Industry/Industrial Analysis FrameworkSubstitution Effect Analysis

  Amplifying effect of PUE improvement on actual IT power growth

  A decrease in PUE (Power Usage Effectiveness) means non-IT equipment (such as air conditioning cooling) consumes less power. Given total electricity consumption is fixed, the better the PUE, the more power is allocated to servers and other IT equipment. Therefore, when observing high growth in total electricity and concurrent PUE improvement, the growth rate of actual computing power demand will be higher than the growth rate of total electricity.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• GDS Holdings (9698.HK)

  Beneficiary Target: As a leading independent IDC provider in China, directly absorbing AI computing infrastructure demand

  Strengths

  Order momentum is steady, although single-quarter expansion/move-in has slowed, long-term demand support is clear

  Weaknesses

  China region business short-term target price lowered, reflecting phased delivery rhythm pressure

  Comparison

  Same Buy rating as VNET, but need to watch divergence between them in different regions and customer structures

• VNET Group (VNET)

  Beneficiary Target: Maintain Buy rating, benefitting from domestic data center demand exceeding expectations

  Comparison

  Recommended alongside GDS, both are core industry assets

• Range Intelligent (300442.SZ)

  Beneficiary Target: First coverage gives Buy rating, positioned as intelligent O&M/supporting service provider

• Shanghai Athub (603881.SS)

  Neutral Observation: Give Neutral rating, not included in preferred beneficiary list

  Comparison

  Rating lower than GDS and VNET, may reflect differences in growth potential or service capability

• Beijing Sinnet (300383.SZ)

  Negative View: Give Sell rating

  Comparison

  Only stock among covered names receiving Sell rating, fundamentals or valuation may have obvious shortcomings

Key data

• Forecast for China Data Center Electricity Consumption in 2030800 TWhNEA forecast value, corresponding to 2025-2030 CAGR 36%, significantly higher than Goldman's expected 20%
• Proportion of Data Center Electricity Consumption in 20306%Significant increase from 1.6% in 2025, reflecting increased weight of digital economy in total social energy consumption
• Data Service Electricity Consumption Growth Jan-Apr 2026+44% YoYHigh-frequency data validates industry demand is in an accelerating release phase
• Direct Investment in Data Infrastructure Next 5 YearsApproximately 2 trillion RMBConsistent with National Data Bureau January 2025 guidelines, covering construction, upgrades, and data utilization

Impact & implications

For data center operators, the official upward revision of demand forecasts means mid-to-long-term rack-up rates and revenue ceilings have been opened up, especially benefiting head companies with high-quality computing power delivery capabilities in the AI infrastructure construction wave. Although some companies may face fluctuations in capacity expansion or customer move-in rhythms in the short term (e.g., GDS China business target price slightly lowered), the overall beta attribute of the industry has strengthened. For power equipment and supporting infrastructure suppliers, the 36% electricity CAGR also signals certain incremental opportunities in segmented fields such as transformers, switchgears, and liquid cooling equipment.

What to watch

• Whether more specific national-level data center construction rules or fiscal subsidies will be implemented subsequently
• Whether quarterly rack-up rates and new signed orders of major IDC providers match the 36% electricity growth expectation
• Whether stricter PUE regulatory policies will affect renovation progress and capital expenditures of some older data centers