AI demand drives rebalancing of GDS data center supply and demand, with Citi assigning a target price of US$58.1/share

Overview

This report is GDS Holdings meeting notes published by Citi after its 2026 Real Estate and Financial Conference. The core content focuses on GDS's domestic data center demand structure, incremental AI orders, supply-demand rebalancing, upgrades in data center operating metrics, and its C-REITs asset rotation strategy. The overall tone of the report is positive, arguing that AI-driven new demand is changing the industry's supply-demand dynamics and enhancing the leasing efficiency and asset return potential of GDS's new capacity.

Core views

Citi believes GDS's customer structure is dominated by hyperscalers, with AI-native customers becoming an important source of new demand, and single order size having increased from about 20MW in the CPU era to several hundred MW. The company believes chip supply remains the main bottleneck for growth in the domestic data center market, but as domestic chip production capacity expands and import channels selectively ease, incremental market demand may grow in 2027E. On the supply side, the previous oversupply phase has ended, new AI cabinets are near full utilization, traditional enterprise cabinets ramp more slowly but command higher unit prices, and the industry is showing K-shaped divergence.

Analysis framework

The report is mainly based on management discussions at the conference, providing qualitative and quantitative analysis of customer structure, industry contracted orders, utilization rates, cabinet power, liquid cooling penetration, project investment discipline, and asset exit returns, and uses an SOTP framework to value GDS's China business.

Methodology notes

• Valuation methodsSOTP and EV/EBITDA

  Valuation by business segment using FY27E EV/EBITDA multiples

  The target price of US$58.1/share is based on a 15x FY27E EV/EBITDA SOTP valuation for GDS's China business.

• Industry supply and demandData center orders and utilization tracking

  Use industry-contracted GW, company-addressable MW, utilization, and lease-up upon delivery to assess supply-demand conditions

  The report supports its supply-demand rebalancing view with 2026E industry contracted orders of 3-4GW, 2027E potential demand of 6-8GW, and utilization above 90% for new AI-ready data centers.

• Operating efficiencyCabinet power density and cooling technology

  Use kW/㎡, per-cabinet power, and liquid cooling penetration to measure a data center's ability to support AI workloads

  New site density is about 3.5-4kW/㎡ and liquid cooling penetration is about 30%, reflecting that demand for high-end AI computing power is driving upgrades in data center specifications.

• Capital returnsC-REITs asset rotation

  Enhance equity IRR and cash returns through mature project exits and capital recycling

  GDS is executing a C-REITs asset rotation strategy, targeting about 20% equity IRR for six-year project exits, while project cash returns remain at 10-11%.

Asset mapping & comparison

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

• GDS HOLDINGS LTD (GDS.O)

  Core company covered in the report

  Strengths

  Strong hyperscaler customer base, expanding AI-native order size, utilization of new AI-ready data centers exceeding 90%, disciplined project investment only after order confirmation, and C-REITs asset rotation can enhance capital returns.

  Weaknesses

  Overall utilization is still 77%, traditional enterprise cabinets ramp more slowly, there is a 1.5-2 year lag from order to revenue recognition, and growth is constrained by chip supply.

  Comparison

  AI cabinets are near full utilization with strong demand, while traditional enterprise cabinets ramp more slowly but have higher unit prices, creating K-shaped divergence.

  Risks

  Fluctuations in the pace of large customer move-ins, policy guidance causing potential oversupply, project delivery falling short of expectations, low-price expansion by competitors, rising utility costs, and upward interest rates.

• Domestic data center/IDC industry

  Main industry backdrop for GDS operations and order growth

  Strengths

  2026E contracted orders are about 3-4GW, 2027E potential demand could reach 6-8GW, and both AI training and inference demand are driving capacity upgrades.

  Weaknesses

  Chip supply remains the main bottleneck, construction and customer ramp-up cycles are long, and demand differs by workload and region.

  Comparison

  Remote server rooms in western regions are more geared toward AI training demand, while capacity in tier-1 cities serves more inference demand.

  Risks

  Policy-encouraged construction could lead to renewed oversupply, and NDRC window guidance and regional price changes require continued monitoring.

• ESSENT GROUP LTD (ESNT.US)

  Appears only in entity extraction; the main text provides no investment analysis

  Strengths

  No evidence in the main text

  Weaknesses

  No evidence in the main text

  Comparison

  No evidence in the main text

  Risks

  No evidence in the main text

Key data

• Target priceUS$58.1/shareBased on a 15x FY27E EV/EBITDA SOTP valuation for GDS China.
• 2026E domestic industry contracted ordersabout 3-4GWThe company observes domestic industry orders tracking this scale.
• 2027E potential incremental demand6-8GWDepends on expansion of domestic chip production capacity and selective easing of import channels.
• GDS potential capacity capture this year500-800MWThe company may capture this scale of new orders in 2026.
• Share of demand from large internet customers55-60%Large internet giants collectively contribute the majority of GDS's domestic IDC demand.
• Share of bookings from major internet companies>80%15-20 major internet companies account for more than 80% of the company's bookings combined.
• Overall utilization rate77%The company's current overall data center utilization rate.
• Utilization rate of new AI-ready data centers>90%New capacity is essentially fully leased upon delivery.
• Revenue recognition lag1.5-2 yearsAbout 1 year of construction plus about 1 year of customer ramp-up period.
• Power density of new sitesabout 3.5-4kW/㎡Higher than the 2.3-2.5kW/㎡ of traditional capacity.
• Liquid cooling penetration in new projectsabout 30%Mainly reserved for high-end AI computing power.
• Average power per cabinet10-15kWTier-1 cities still have many traditional 6-8kW CPU cabinets.
• Target equity IRR on project exitabout 20%Corresponding to six-year project exits and C-REITs asset rotation.
• Project cash return10-11%Measured by EBITDA/investment cost.

Impact & implications

If AI orders and chip supply continue to improve, GDS's new AI data center capacity may maintain high utilization and support pricing, revenue ramp-up, and valuation recovery. Order-driven project selection discipline helps reduce the risk of speculative expansion, while C-REITs asset rotation can improve capital recovery and equity returns. However, revenue recognition has a 1.5-2 year lag, and policy, competition, delivery, and energy costs may still affect profit realization.

Risks

• The pace of large customer move-ins may be uneven, and delays would significantly affect GDS's performance.
• The Chinese government is encouraging operators to build capacity, which could cause industry oversupply to reappear.
• There remains execution risk as to whether GDS can deliver its data center pipeline on time to serve pre-committed customers.
• Well-funded competitors may irrationally build data centers and compete at lower prices.
• Utility costs may rise, and a higher share of green energy may also bring higher costs.
• Rising interest rates may depress earnings and lead to a downward re-rating of the IDC sector.
• Chip supply is the main bottleneck for growth in the domestic data center market.

What to watch

• Progress in the expansion of domestic chip production capacity and selective easing of import channels.
• Whether 2026E domestic industry contracted orders of 3-4GW materialize.
• Whether 2027E incremental market demand can expand to 6-8GW.
• Whether GDS can capture 500-800MW of new orders this year.
• Whether utilization of new AI-ready data centers remains above 90%.
• Whether cabinet prices in Ulanqab rise in 2027, and whether price increases spread nationwide.
• The pace of improvement in liquid cooling penetration and cabinet power density in new projects.
• Execution of C-REITs asset rotation, project exits, and the target of about 20% equity IRR.