Slower 2Q26 deployment weighs on near-term results, but the 12-month investment thesis for China data centers is unchanged

Overview

This report is a 2Q26 preview for the China data center industry and related companies. The core view is that, due to tight supply of domestic and imported chips, industry deployment progress in 1H26 was weaker than previously expected, which may weigh on 2Q26 quarterly results; however, AI training and inference demand, progress in domestic AI hardware, demand taken on by third-party data centers, and order and capacity reserves at key companies keep the 12-month investment thesis intact.

Core views

Goldman Sachs maintains a constructive view on the China data center industry. In the short term, 2Q26 results are mainly affected by slower deployment pace, with VNET revenue expectations lowered and GDS target price slightly reduced. Over the medium to long term, the report still expects China data center demand to achieve 20% CAGR over 2025-28E, with wholesale IDC, AI data centers, and GPUaaS-related opportunities remaining the main drivers. At the stock level, the report continues to recommend GDS, VNET, and Range Intelligent, while watching GDS order updates, VNET's strategic cooperation with CATL and overseas expansion, and Range Intelligent's capacity reserves and GPUaaS earnings contribution.

Analysis framework

The report analyzes industry supply and demand, incremental intelligent computing power, chip availability, domestic hardware performance, data center deployment demand, company backlog, capacity commissioning pace, revenue/EBITDA forecasts, and EV/EBITDA valuation. GDS uses SOTP valuation, VNET uses a 12-month forward EV/EBITDA multiple, and Range Intelligent uses a valuation method based on discounting 2030E EV/EBITDA back to 2026E.

Methodology notes

• Valuation methodSOTP sum-of-the-parts valuation

  Value GDS China and DayOne separately, then combine them and apply a holding company discount.

  GDS target price is based on SOTP: GDS China uses 13.5x 2027E EV/EBITDA, DayOne uses 23x 2027E EV/EBITDA adjusted by GDS's 19.9% stake, plus a 10% holding company discount.

• Valuation methodEV/EBITDA multiple valuation

  Use the target EV/EBITDA multiple multiplied by forecast EBITDA to derive enterprise value and target price.

  VNET target price of US$16 is based on 12x 2027E EV/EBITDA; Range Intelligent target price of Rmb117 is based on 18x 2030E EV/EBITDA discounted back to 2026E at a 10% cost of equity.

• Industry supply-demand analysisDeployment demand and intelligent computing power increment framework

  Use new intelligent computing power, order volume, deployed MW, and capacity utilization to measure the pace of data center demand realization.

  Based on MIIT-disclosed 1H26新增595 EFLOPS of intelligent computing power and Goldman Sachs's estimated AI-related deployment volume of 0.7GW, the report lowers 2026E deployment demand but maintains its 2025-28E medium- to long-term growth view.

• Factor and database toolsGS Factor Profile and Quantum

  GS Factor Profile compares stocks across growth, financial returns, valuation multiples, and composite factors; Quantum provides financial history, forecasts, and ratio databases.

  The report states that Goldman Sachs uses these internal frameworks to provide investment context, cross-company comparison, and support for financial analysis.

Asset mapping & comparison

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

• GDS HOLDINGS LTD / GDS / 9698.HK

  A core beneficiary, China carrier-neutral data center platform; the report maintains Buy.

  Strengths

  Leading position in China's wholesale data center market, with developable capacity in key computing clusters; balance sheet improving through capital discipline and capital recycling; DayOne provides overseas growth and earnings leverage.

  Weaknesses

  Deployment pace already slowed in 1Q26, and 2Q26E is expected to be slightly weaker; China MSR is expected to decline 7%-9% YoY in 2026-28E on an Rmb/kW basis.

  Comparison

  Compared with VNET, GDS places more emphasis on the large-scale wholesale platform and the segment value of GDS China and DayOne; the target price was lowered due to fine-tuning of 2027E forecasts.

  Risks

  Improvement in deployment demand and utilization below expectations, slower-than-expected ramp-up in overseas revenue and profitability, lower-than-expected pricing in Chinese and overseas markets, customer churn, and slower deleveraging progress.

• VNET GROUP INC / VNET

  A beneficiary of the wholesale IDC transition; the report maintains Buy.

  Strengths

  Transforming from traditional retail IDC to fast-growing wholesale IDC, driven by strengthened AI investment; wholesale IDC revenue and EBITDA are expected to achieve 37%-38% CAGR over 2025-28E.

  Weaknesses

  2Q26E revenue was lowered due to slower deployment, and growth in traditional cloud computing and VAS businesses is weaker.

  Comparison

  Compared with GDS, VNET's valuation framework is more directly tied to 12x 2027E EV/EBITDA; the market is focused on the rising wholesale IDC mix, cooperation with CATL, and overseas expansion.

  Risks

  Insufficient financing capability, weaker-than-expected order acquisition and execution, AI-related geopolitical risks, further decline in traditional business, and weaker-than-expected or unexpectedly changing AI model training demand due to technological changes.

• Range Intelligent / 300442.SZ

  The preferred Buy name among A-share data center operators.

  Strengths

  Abundant capacity reserves, full-stack AIDC capabilities, strong customer relationships, and diversified low-cost financing capability; the report mentions China's first 200MW, 100,000-chip-scale data center building has been in operation since 1H26.

  Weaknesses

  High growth depends on order acquisition, utilization ramp-up, financing, and overseas execution.

  Comparison

  Superior to Athub and Sinnet among A-share data center operators, and the report maintains it as the top pick.

  Risks

  Lower-than-expected orders in a competitive environment, slower-than-expected utilization improvement, greater-than-expected pricing pressure, weaker-than-expected overseas expansion execution, changes in chip availability due to regulatory changes or delays in domestic capacity ramp-up, and financing difficulties.

• Sharetronic Data / 300857.SZ and Lettall Electronic / 603629.SS

  A-share companies related to GPUaaS/GPU leasing; the report notes them as not covered but mentions strong preliminary results.

  Strengths

  Rising GPUaaS/GPU leasing spot prices may bring earnings leverage.

  Weaknesses

  The report does not provide formal ratings or target prices, and the evidence for the investment conclusion is limited.

  Comparison

  More exposed to AI hardware leasing price elasticity rather than the capacity realization logic of traditional data center operators.

  Risks

  Spot price volatility, changes in chip supply, long-term contract price increases lagging spot prices, and limited information on uncovered companies.

• Shanghai Athub / 603881.SS and Beijing Sinnet Technology / 300383.SZ

  Comparable A-share data center peers.

  Strengths

  Part of the China data center coverage universe.

  Weaknesses

  The report is Neutral on Athub and Sell on Sinnet, indicating lower relative preference.

  Comparison

  The report explicitly lists Range Intelligent as the top pick among A-share data center operators.

  Risks

  Industry competition, insufficient orders and utilization, pricing pressure, and financing constraints.

Key data

• China's new intelligent computing power in 1H26595 EFLOPS (FP16)Below the 802 EFLOPS added in 2H25 and close to the 1H25 level, reflecting a slowdown in deployment pace under tight chip supply.
• China data center deployment demand in 2026Eabout 3GWAI and non-AI combined, lowered from previous expectations; the report estimates that more than 50% comes from intelligent computing demand and more than 80% is undertaken by third-party data centers.
• 2025-28E CAGR of China data center demand20%The report expects demand to rise from 16GW in 2025 to 27GW in 2028E.
• GDS 2Q26E revenueRmb3.07bn, YoY +6%2% below Visible Alpha consensus; 2Q26E adjusted EBITDA is Rmb1.34bn, YoY -2%, 2% above consensus.
• GDS target priceUS$46 / HK$45Lowered from US$49 / HK$47, based on SOTP valuation.
• VNET 2Q26E revenueRmb2.77bn, YoY +14%Lowered by 3% due to slower deployment, but 2Q26E adjusted EBITDA is broadly maintained at Rmb927mn, YoY +26%.
• VNET target priceUS$16Unchanged, based on 12x 2027E EV/EBITDA.
• VNET wholesale IDC revenue mix57% in 2028E vs. 35% in 2025The report believes the increasing contribution from wholesale IDC will drive valuation re-rating and value compounding.
• Range Intelligent target priceRmb117Based on 18x 2030E EV/EBITDA discounted back to 2026E; the company had 750MW of operating capacity as of 2025 and recently about 6GW of reserve capacity.
• GPUaaS/GPU leasing spot priceMonthly spot price for Nvidia H100 servers rose from about Rmb50k to about Rmb100kTight chip supply combined with strong AI demand has pushed up spot leasing prices; long-term contract prices may not rise as much as spot prices.

Impact & implications

Near-term earnings pressure mainly comes from deployment pace rather than disappearing demand, so market focus should shift from single-quarter revenue volatility to order conversion, capacity commissioning, improvements in chip supply, and the sustainability of AI demand. If new intelligent computing power additions re-accelerate in 2H26, the wholesale IDC growth of GDS and VNET, and Range Intelligent's capacity reserves and GPUaaS positioning, may receive stronger validation; conversely, if chip availability, financing, or order execution fall short of expectations, upside to valuations may be limited.

Risks

• Continued tight chip supply, including insufficient availability of both domestic and imported chips.
• Improvement in deployment demand and capacity utilization below expectations.
• AI training or inference demand falls short of expectations due to changes in technology paths, efficiency improvements, or geopolitical factors.
• Greater-than-expected pricing pressure in China and overseas data centers, with MSR decline affecting revenue quality.
• Customer churn or weaker-than-expected new order acquisition.
• Slower-than-expected ramp-up in overseas expansion revenue and profitability.
• Financing difficulties or slower-than-expected deleveraging progress, limiting capacity expansion.
• A decline in GPUaaS/GPU leasing spot prices, reducing related earnings leverage.

What to watch

• GDS order updates in July-August, especially whether the potential 500MW of orders in 1H26 previously mentioned by management materializes.
• Further comments on VNET's earnings call regarding strategic cooperation with CATL, overseas expansion, and large new orders.
• Whether China's new intelligent computing power additions re-accelerate in 2H26.
• Delivery and actual performance of domestic AI hardware systems such as Huawei Atlas 950 SuperPoD and Alibaba T-head Lingjun Zhenwu M890 SuperPoD.
• The pace of increase in wholesale IDC revenue mix, and the margin performance of VNET's transition from retail IDC to wholesale IDC.
• The extent of MSR decline, renewal area ratio, and utilization changes at GDS China.
• Whether DayOne's revenue, adjusted EBITDA, and margin ramp-up meet high-growth expectations.
• Whether the gap between GPUaaS/GPU leasing spot prices and long-term contract prices widens or converges.