GDS Holdings

GDS lifted its fiscal 2026 revenue growth guidance to 12.4% midpoint from 10.6% previously and raised adjusted EBITDA growth to 11% from 8.7% previously. Fiscal 2026 capital expenditure guidance has also been lifted to CNY 10 billion from CNY 9 billion previously.

Why it matters: Management believes it is at the beginning of a multiyear growth story, supported by the increasing availability of domestic chips, and indicated a willingness to commit CNY 30 billion-CNY 50 billion of capital expenditure over the next three years to meet artificial intelligence-driven demand. It expects a return on equity of 20%. GDS has strong sales momentum in 2026 with 263 megawatts of new commitments in the second quarter following 208 MW in the first quarter, and only 61 MW in second-quarter 2025. Management expects 2026 move-in of 235 MW, around 700 MW in 2027, and around 950 MW in 2028. Capex in the first half was only CNY 2 billion, so GDS has CNY 8 billion to spend in the second half. The company is planning another asset injection into its C-REIT for second-half 2026, which should help fund the increased 2026 capex. In 2025, it raised CNY 2.3 billion for asset monetization.

The bottom line: We increase both our capex and EBITDA forecasts, with our fair value increased to HKD 39 (USD 40) from HKD 37 (USD 38). The stock is currently 3-star-rated. Our no moat rating is also retained. GDS' current price/book ratio is around 1.6 times compared with 3.4 for VNET. GDS does not revalue its data centers each year, unlike a property REIT, and we estimate that its early data centers in the central business districts of major cities have increased in value.

Long view: With GDS leveraged to long-term growth themes such as the shift of IT to the cloud and AI, we see it as a decent long-term investment. However, the share price has been volatile due mainly to high financial leverage and variable demand, which is why it has a Very High Morningstar Uncertainty Rating.

Our fair value estimate for GDS stands at USD 40 per share. We assume GDS revenue grows at an average of 13% per year over the next five years, with operating income increasing 23% per year. In 2035, we assume a steady state in which its data centers are more mature, with portfolio-wide EBITDA margins of 66%. Australian data center company NextDC provides financials for two of its most mature data centers in Melbourne and Sydney, with utilization rates over 90%. The numbers show EBITDA margins of around 80% and returns on invested capital in the midteens, and we see these as illustrative of the types of returns we can expect from GDS once its data centers mature.

The company is also excited about its international expansion into Southeast Asia, but in raising funds for the growth of this business, named DayOne, it has sold down its stake to 23% and has now deconsolidated this business. Its most recent sale was 15% for USD 385 million, valuing the remaining interest at USD 2.2 billion or USD 11.18 per share. Management has previously indicated that an IPO of this business is a possibility.

We assign GDS a no-moat rating because we believe the bargaining power of its large hyperscale customers will offset the switching costs normally associated with the data center industry. Moats in data center businesses under Morningstar’s coverage can be derived from network effects and customer switching costs. We generally rate hyperscale data network businesses (those that lease large spaces to hyperscale tenants on lengthier contracts, such as GDS) as less moaty than retail colocation businesses. Most data center companies do both hyperscale and retail, with GDS focusing more on hyperscale. Around 90% of GDS's revenue is from hyperscale customers, with the remaining 10% from retail customers. In a capital-intensive industry in a high-growth phase, we do not expect returns on invested capital to exceed the cost of capital in the forecast period.

We estimate GDS’s capacity in service in downtown areas of major cities represents well under half of GDS’s existing data center capacity in service. While it is very hard to get data center capacity in these locations, so it will be hard for customers in these data centers in particular to switch to other providers, our moat ratings are forward-looking and we see the bulk of GDS’s new data center capacity being deployed in areas at the fringes of the major cities where large customers will likely have more options to switch to other data center suppliers when their contracts expire, or use the threat of moving to keep a lid on returns for GDS. We typically see switching costs as competitive advantages for data centers. If data center tenants leave, they need to take significant risks to move critical servers and networking equipment and incur expenses, which, according to Align Communications, could cost tens of millions of dollars for multi-megawatt customers.

GDS has generally reported churn rates of around 3-5%, respectively, which are low numbers in themselves, and the company indicated that this churn is from smaller retail clients, not the larger hyperscale customers from which it generates most of its revenue. GDS has not seen any churn from its largest customers. Agreements with the large cloud service providers typically have periods of three to 10 years, whereas other customers typically have contract periods of one to five years. Often the initial contract for data center capacity is a competitive tender, but then expansions are negotiated. Crucially, GDS doesn’t build new data centers unless it has enough committed customers—it does not build a data center and then try to fill it.

The data center market has been growing strongly, and we believe most of the recontracted data centers have been in inner city areas where demand has been high, and supply has been low. We expect that the large hyperscale customers will likely hold better bargaining power in future contracts that expire in outer city data centers. GDS is transparent with its large hyperscale customers about its costs, pricing on a “cost plus reasonable return” basis. Its returns for the hyperscale customers are therefore slightly lower than for the retail component of its business, but still above the cost of capital according to the company. We believe hyperscale data centers are less moaty than retail because, given the longer contracts wholesale customers sign, they are far more discerning and more likely to justify the costs of moving. We also know hyperscale customers can replace a presence at GDS with one in a self-owned data center. However, GDS’ largest hyperscale customers, such as Alibaba and Tencent, after favoring their own data centers in the earlier years, have moved away from self-built and owned data centers and now use capacity almost exclusively from independent providers. We also believe GDS is helped in the tenders by its experience and reputation for building quality data centers, but do not see this as strong enough to warrant an intangible asset moat.

The network effect for data centers is generated in major internet exchanges where key internet service providers are connected, particularly for applications where latency is important. In this situation, service providers attract customers to a data center, which then attracts more service providers. In China, the key areas where network effects are likely present are in the CBD areas of Tier 1 cities where new data center capacity is difficult to build, such as in Shanghai and Beijing. GDS has significant capacity in service in these areas. It has 100,000 sqm in service in downtown Shanghai and 107,000 sqm in downtown Beijing, which represents under half of its existing data center capacity in service. In addition, most of its area under construction and land held for future expansion is at the edge of these cities, where suitable sites are more plentiful, so we would expect the amount of capacity in prime downtown areas to fall below 50% within the next two years. In our view, networks are crucial to maintaining a network effect because they make data centers more valuable both to each other and to cloud providers, which then draw their own customers.

While GDS has a bigger network of data centers than some of its independent competitors, we are not convinced that scale provides enough advantage for a moat. GDS’s total data center capacity is around twice as large as other independent data center operators such as VNET, Chindata, and Sinnet. In any case, the three Chinese telecom companies have much bigger data center networks than GDS. We note that the Chinese telecom companies have a much greater percentage of their data centers located outside the main Tier 1 markets and, in many cases, are the main data center providers in Tier 2 and Tier 3 cities. To the extent that any scale-based cost advantages are available in this industry, we would expect the Chinese telecom operators to be the main beneficiaries.

GDS’ rapid expansion in the “land grab” stage of the industry should allow it to own data centers in sought-after areas such as downtown in Tier 1 cities and in data center clusters.

5G, industrial internet, cloud computing, artificial intelligence, and other innovations that increase the demand for data and connectivity leave us in the early stage of a data center renaissance.

As an early entrant into the data center business in China, GDS has built up expertise and a reputation that should allow it to remain competitive in data center construction and operations.

GDS' data centers are not significantly differentiated from those that others, including its biggest customers, could supply.

GDS needs its biggest customers more than they need it, and its highly concentrated customer base (in terms of revenue exposure) leaves the company at their mercy.

A Chinese property downturn could leave GDS exposed, given it owns around two-thirds of its developed and developable capacity of over 1 million square meters.