Goldman Sachs: China’s data-center computing expansion accelerates, while business models and valuation trends diverge

Why is computing-capacity expansion accelerating, and where are the bottlenecks?

MIIT data show that China added 407 EFLOPS of AI computing capacity (FP16) in the third quarter of 2026, 37% more than the average quarterly addition in the first and second quarters, and up 146% year over year. Goldman Sachs considers increased imports of advanced computing capacity to be one reason for the faster growth.
Domestic chip shipments are still ramping up under constraints: Cambricon’s third-quarter revenue was 34% below the Visible Alpha consensus. Goldman Sachs expects that, under chip-supply constraints, VNET’s newly added customer onboarding capacity in the third quarter will be 57 MW, broadly flat versus the second quarter; GDS’s will be 45 MW, with the latter also affected by lower orders in the previous year.
Meanwhile, extrapolating from management guidance, average quarterly capex at Range Intelligent and Guangdong HEC from the fourth quarter of 2026 through the first quarter of 2027 could exceed 60 billion yuan, close to the average level of about 65 billion yuan Goldman Sachs estimates for Alibaba and Tencent in the third quarter of 2026, and could add 150–300 MW of new IT capacity per quarter.

Goldman Sachs’ return framework shows that the traditional model, which mainly provides powered data-center facilities, generates a return on invested capital of about 5% on an after-tax operating-profit basis. An integrated AI data center with long-term GPU computing-service contracts can generate returns above 15%, provided it can obtain advanced computing capacity at attractive costs. The firm considers monetization to remain the key challenge for domestic computing architectures.
Range Intelligent’s third-quarter results were below Goldman Sachs’ expectations because of the timing of sales recognition, but its AI computing investment for the quarter was about 8 billion yuan, close to the roughly full-year 2025 total of 10 billion yuan. Goldman Sachs expects it to invest a further 40 billion yuan from the fourth quarter of 2026 through the first quarter of 2027, and raised its forecast for 2027 AI data-center revenue to 18.9 billion yuan.
Traditional business growth is slowing: Goldman Sachs forecasts Range’s China IDC revenue to grow 10% in 2026 and 5% in 2027, respectively. Capacity delivery in 2027 is expected to be 50 MW, below 130 MW in 2026.

Capital recycling helps support continued expansion. GDS plans to inject 4.8 billion yuan of assets into a publicly offered REIT; the National Development and Reform Commission has approved the plan and recommended it to the China Securities Regulatory Commission for final approval. VNET’s injection of 2.3 billion yuan of assets into its existing private REIT has been approved by the Shanghai Stock Exchange. Range expects the injection of its 78 MW A7 and A8 data centers into an existing publicly offered REIT to be completed in the first quarter of 2027.
Range also announced 50 billion yuan in new credit facilities and an equity investment of 6.8 billion yuan by the Export-Import Bank of China in its indirectly wholly owned subsidiary. Goldman Sachs considers access to low-cost domestic funding to be a competitive advantage for Chinese AI data-center operators.
Increased investment has not removed pressure on traditional businesses. Sinnet’s third-quarter revenue fell 1% year over year to 1.7 billion yuan, still 3% above Goldman Sachs’ forecast; net profit, however, fell 99.5% year over year to 100 thousand yuan, mainly due to a year-over-year gross-margin decline of 2.8 percentage points amid ongoing pricing pressure, as well as non-operating items such as interest expense and fair-value losses.

Greater China data-center stocks covered by Goldman Sachs fell an average of 25% over the previous three months, with VNET and GDS down 58% and 34%, respectively. The firm mainly attributed the share-price performance to differences in the cost of capital resulting from a widening China–US interest-rate spread, as well as market concerns about returns on AI capex and financing.
Goldman Sachs lowered its target prices for the US-listed ADRs it covers by an average of about 50%, and its target prices for A-shares and H-shares by about 15%, mainly because it adopted lower target valuation multiples. Nevertheless, the average valuation of A-shares and H-shares remains more than 20% above the trough of 2023–2024, reflecting continued optimism priced in for the AI upcycle.
Views still differ by stock: Goldman Sachs maintained Buy ratings on Range, GDS and VNET; it maintained Neutral ratings on Athub and SUNeVision, and a Sell rating on Sinnet.