First Coverage of Range Intelligent Amid the AI Infrastructure Wave

Overview

Goldman Sachs provides its first in-depth coverage of China's AI infrastructure sector, focusing on how data center operators are leveraging the AI investment wave to expand into high-density computing and cutting-edge markets. The report details the profitability logic and risks of the GPU-as-a-Service (GPUaaS) business model, assesses the cost advantages of western computing clusters under the 'East Data West Computing' policy, and scores and values six core companies. Conclusion: Assigning a Buy rating to Range Intelligent (300442.SZ) with a target price of 117 yuan; a Neutral rating to Shanghai Athub (603881.SS) with a target price of 34 yuan; while maintaining Buy ratings for GDS, VNET, and SUNeVision.

Core views

Demand Side: Leading Chinese cloud providers are entering a period of rapid growth in AI capital expenditures, with projected 20% compound annual growth in data center demand from 2025–2028E, and surging AI inference demand driving GPU computing rental prices up roughly 30% since the beginning of the year. Supply Side: The National Development and Reform Commission has tightened approvals for new data centers, prioritizing projects that already hold power quotas, are progressing quickly, and have solid customer contracts, suggesting accelerated industry consolidation. Western computing clusters, benefiting from lower electricity and land costs, boast unit capex 20% lower than first-tier cities, with annual operating costs potentially reduced by one-third. GPUaaS Business Model: Taking an H100 cluster as an example, assuming a monthly rent of 60,000 yuan and 80% utilization, a five-year IRR could reach double digits, with a payback period of 3–4 years; success hinges on securing long-term demand and maintaining high utilization rates. Company Level: - Range Intelligent: Holds a 6 GW capacity reserve—the largest in the industry—with 60–70% already allocated power quotas; customer concentration has dropped from 92% to 49%, leading in full-stack AIDC plus liquid cooling technology; projected 2025–2028E revenue/EBITDA CAGR of 40%/47%, warranting a Buy rating with a target price of 117 yuan. - Shanghai Athub: Boasts 444 MW capacity, with 401 MW already operational; relatively homogeneous customer base, slower transition to GPUaaS compared to Range, earning a Neutral rating with a target price of 34 yuan. - GDS: Solid leadership position, with 3.9 GW reserves primarily located in Inner Mongolia/Ningxia, projecting an MSR CAGR of -8%, SOTP target price $55. - VNET: Transitioning from retail to wholesale, with 2025–2028E revenue/EBITDA CAGR of 37–38%, target price $15.5. - SUNeVision: Hong Kong's largest operator, with clear plans to double capacity, target price HK$7.7.

Analysis framework

The report employs a three-dimensional 'Capacity-Demand-Profitability' framework: 1. Demand Side: Comparing US-China cloud provider capex, tracking AI inference token growth rates, and assessing price elasticity to confirm strong GPUaaS demand. 2. Supply Side: Using power quotas as a hard constraint, evaluating each company's capacity reserves and approval progress to forecast deployment pace over the next three years. 3. Profitability Model: Building a unit economics model for GPUaaS, conducting sensitivity analyses on rental pricing, utilization rates, and residual value, then factoring in western cost advantages to calculate IRR and payback periods. Finally, a scoring card synthesizes current operational metrics and forward-looking indicators to derive target valuations and ratings.

Methodology notes

• Valuation MethodSOTP Segment Valuation

  Applying SOTP valuation to GDS—separately valuing traditional IDC and AI subsidiary DayOne at 14.5x and 23x 2027E EV/EBITDA, respectively, then summing them and deducting a controlling stake discount.

  When a company operates across diverse segments with markedly different growth trajectories, segment-based valuation helps avoid distortions caused by single multiples, more accurately reflecting the premium attached to high-growth AI segments.

• Industry/Supply Chain Analysis FrameworkSupply-demand framework

  Treating power quotas as a supply-side constraint, combining AI capital expenditure and inference demand to gauge supply-demand gaps and price trends.

  In the data center industry, short-term dynamics hinge on power permits, while long-term growth depends on surging demand; mismatches between supply and demand directly influence rents and profitability.

• Company Fundamentals and Financial FrameworkFree cash flow analysis

  Using unit economics models to estimate IRR and payback periods for GPUaaS projects, assessing cash flow returns under high-utilization scenarios.

  For heavy-asset projects, first calculating the IRR of individual projects, then evaluating the company's overall funding needs and financing capabilities to determine the sustainability of expansion.

Asset mapping & comparison

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

• Range Intelligent (300442.SZ)

  Buy rating, key beneficiary of 6 GW capacity reserve plus GPUaaS transformation

  Strengths

  Largest capacity reserve in the industry, sharply reduced customer concentration, leading full-stack AIDC and liquid cooling technologies, diversified low-cost financing channels

  Weaknesses

  High capex of 21 billion yuan over the next three years, requiring ongoing financing

  Comparison

  Outperforms peers in both current and forward-looking metrics, exhibiting superior growth and profitability elasticity

  Risks

  Slower-than-expected financing progress, GPUaaS utilization below 80%, rental prices falling

• Shanghai Athub (603881.SS)

  Neutral rating, slower transition to GPUaaS compared to Range

  Strengths

  444 MW capacity, with 401 MW already operational, stable cash flow

  Weaknesses

  Relatively homogeneous customer base, limited capacity reserves

  Comparison

  Scores lower than Range, GDS, and VNET on the scoring card

  Risks

  Customer concentration risk, limited access to new capacity

• GDS Holdings (GDS/9698.HK)

  Maintains Buy rating, traditional IDC leader

  Strengths

  3.9 GW capacity reserve, leading resource allocation, disciplined financial management

  Weaknesses

  Traditional wholesale IDC pricing under pressure, MSR CAGR projected at -8%

  Comparison

  Strong scale and brand advantages, but growth lags behind Range

  Risks

  Continued price declines, volatility in DayOne valuation

• VNET Group (VNET)

  Maintains Buy rating, transitioning from retail to wholesale

  Strengths

  Wholesale IDC revenue share rapidly increasing, 2025–2028E revenue/EBITDA CAGR of 37–38%

  Weaknesses

  Potential shareholder dilution and financing pressures

  Comparison

  High flexibility in transformation, valuation with room for reevaluation

  Risks

  Shareholder dilution, elevated leverage

• SUNeVision Holdings (1686.HK)

  Maintains Buy rating, Hong Kong's leading operator

  Strengths

  Hong Kong's largest data center operator, clear capacity expansion plan

  Weaknesses

  Lower score than mainland peers, limited market capacity

  Comparison

  Regional monopoly plus high dividend yield, strong defensive attributes

  Risks

  Market capacity bottleneck in Hong Kong, rising electricity costs

Key data

• China's Data Center Demand CAGR 2025–2028E20%Driven primarily by high-density AI demand
• GPUaaS Rental Prices YTD IncreaseApproximately 30%Rising synchronously across US and China markets
• Western Data Centers Have Lower Unit Capex Than First-Tier Cities20%Advantageous land and construction costs
• Range Intelligent's Capacity Reserve6 GWLargest in the industry, with 5.4 GW located in mainland China
• Range Intelligent's 2025–2028E Revenue/EBITDA CAGR40%/47%Highest among covered companies
• IRR for H100 Cluster at 80% Utilization Over Five YearsDouble-digitAssuming a monthly rent of 60,000 yuan

Impact & implications

The report argues that AI infrastructure is currently experiencing a confluence of surging demand and favorable policy incentives: on the demand side, cloud providers continue ramping up AI investments; on the supply side, tightening power quota regulations are spurring industry consolidation, with leading firms boasting robust capacity reserves, diversified customer bases, and strong financing capabilities poised to benefit significantly. Western low-cost clusters will become the primary battleground for the next round of expansion, with companies that secure power resources early gaining outsized returns. Investors should closely monitor each company's progress in obtaining power quotas, the duration of GPUaaS contract lock-ins, and actual utilization rates.

Risks

• Fluctuations in GPUaaS rental prices and utilization rates leading to IRR below expectations
• Delays in approving new data center power quotas
• Tightening financing conditions amid high leverage
• Limited access to advanced GPUs or changes in regulatory policies

What to watch

• Progress in securing power quotas for each company
• Tracking long-term GPUaaS contract signings and utilization rates
• Verifying actual costs and demand after western data centers go live
• M&A opportunities emerging during industry consolidation