As AI transforms "computing power" into the oil of the new era, data centers are no longer just a capital-intensive business but have begun to become the core infrastructure re-priced by the capital market—GDS Holdings' latest financial report stands right at the critical point of this cycle shift. For a long time in the past, data centers were viewed as an extension of real estate logic: leasing land, building facilities, installing racks, and collecting rent. This model, constrained by high depreciation and interest expenses, often fell into the dilemma of "revenue growth without corresponding profit growth." However, given the explosive growth of generative AI, the scarcity of computing power has been redefined, and the physical space hosting computing power has also surged in value. GDS Holdings, a leading neutral data center operator in China, saw its latest financial performance not only reflect the recovery of its own business cycle but also mirror the value reconstruction of the entire digital infrastructure industry under the AI wave. The capital market is observing whether this veteran operator can shed its historical burdens and truly become the "computing power landlord" in the AI era.
What Does This Financial Report Really Show: "Structural Recovery" Behind Steady Growth
If looking only at the surface, this is a typical "steady-growth financial report." In 2025, GDS Holdings' revenue grew by 10.8% year-on-year to RMB 11.4 billion, with adjusted EBITDA reaching RMB 5.4 billion, maintaining a profit margin of over 47%, and achieving a net profit of nearly RMB 1 billion. This set of data indicates that the company has emerged from the profitability pressure of its previous high-investment phase and entered a cycle of "gradual cash flow realization." For the data center industry, improvements in the income statement often lag behind the balance sheet, as the massive capital expenditures in the early stages require a long time to be absorbed through depreciation. GDS Holdings' ability to turn net profit positive demonstrates that the operational efficiency of its existing assets has covered fixed costs, which is a crucial inflection point signal.
More attention should be paid to the improvement in operational metrics. The billed area increased by 11.4% year-on-year, and the billing rate rose to 75.5%, indicating that new capacity is being absorbed more effectively and the capacity utilization rate is steadily advancing. For a typical capital-intensive industry like data centers, this is a key variable for valuation recovery—because what the market truly cares about is not "how much has been built," but "whether it can be leased out." Over the past few years, due to macroeconomic fluctuations and the slowdown in the internet industry's growth, some data centers faced the risk of insufficient rack deployment rates, leading to high asset idle costs. The recovery of the billing rate to over three-quarters shows that downstream customer demand is warming up, and the company's sales strategy and customer structure optimization have yielded tangible results.
Furthermore, the company has strengthened its financial capabilities through asset-backed securities, public REITs, and equity financing, essentially addressing the core constraint in the data center industry: the balance between capital expenditures and financing capabilities. Data centers are a capital-intensive industry with long construction cycles and slow returns, making them extremely sensitive to financing costs. GDS Holdings is actively exploring the path of public infrastructure REITs, attempting to package mature data center assets for listing, thereby recycling funds for new project development. If this closed-loop model of "investment, financing, management, and exit" can be successfully executed, it will greatly reduce the company's debt pressure and improve its return on net assets.
However, the problems are equally obvious. The projected capital expenditure for 2026 remains as high as RMB 9 billion, meaning the company is still in a "high-investment, high-debt" expansion phase. Against the backdrop of the interest rate environment and capital market volatility, this model inherently carries a valuation discount. Although operating cash flow has improved, the massive capital expenditures mean that free cash flow remains tight. What investors need to weigh is whether these newly invested capacities can be rapidly converted into revenue within the next two years, or whether they will become a new depreciation burden. Especially under the advancement of the "East Data and West Computing" project, data centers in some remote areas face the risk of insufficient demand, while energy consumption quotas around core cities are extremely scarce. Where GDS Holdings directs its capital expenditures is key to determining its future earnings quality. From an investment perspective, the essence of this financial report is: the fundamental recovery is clear, but it has not yet entered the "high-elasticity growth" stage.
New Opportunities in the AI Era: Data Centers Evolving from "Supporting Facilities" to "Core Assets"
What truly changes the valuation logic of GDS Holdings is not the current performance, but the demand reconstruction brought by AI. In the traditional cloud computing era, data centers were "supporting facilities" for internet companies, with relatively stable demand and growth highly correlated with the internet traffic cycle. At that time, the rack power density was usually between 4 and 6 kilowatts, mainly to meet general computing and storage needs. But in the AI era, the situation is undergoing a fundamental change—computing power demand is showing exponential growth.
Computing power vendors represented by NVIDIA are driving the continuous expansion of GPU cluster scales, and this high-density computing power poses higher requirements for data centers: higher power density, more complex cooling systems, and more stable energy supply. The rack power density for AI training clusters is often as high as 20 kilowatts or even over 50 kilowatts, posing severe challenges to power supply stability and cooling technology. Traditional air cooling solutions can no longer meet the demand, and liquid cooling technology is becoming the standard for newly built AI data centers. This means that data centers are no longer simply "rack leasing" but are upgrading to "computing power infrastructure platforms." Data centers equipped with high power density delivery capabilities, liquid cooling retrofit capabilities, and sufficient power quotas will become scarce resources.
For GDS Holdings, this brings two layers of opportunities. First, enhanced demand certainty. The long-term demand for AI training and inference makes large tech companies and cloud providers more willing to sign long-term contracts, thereby improving order visibility. The continuous growth of GDS Holdings' signed and pre-signed area in 2025 is a reflection of this trend. In particular, cooperation with leading cloud providers and AI large model companies is often accompanied by long-term agreements of five or even ten years, providing a safety cushion for revenue. Second, the increase in unit value. The unit investment and rental levels of AI data centers are much higher than those of traditional data halls, which will drive up the overall ARPU in the industry, thereby bringing about an improvement in revenue quality. A high-power rack capable of hosting AI clusters can generate rental revenue three to five times that of a traditional rack.
In other words, AI is transforming data centers from "capital-intensive burdens" into "scarce resources." This scarcity is reflected not only in physical space but also in energy consumption quotas. Under China's dual carbon goals, the approval of energy consumption quotas for data centers in core cities is extremely strict. Its substantial existing assets deployed in core economic zones such as the Yangtze River Delta and the Greater Bay Area, with their legal energy consumption quotas and stable power supply, possess attributes similar to "prime commercial real estate." As AI computing power demand extends to the edge, the value of core nodes close to the user side will be further highlighted. This resource barrier is difficult for new competitors to replicate in a short period.
Hidden Concerns and Divergence: A Game of "Cyclical Stock vs. Core Asset"
Despite the grand narrative of AI, market divergence regarding GDS Holdings still exists. This divergence concerns not only the company itself but also the pricing logic of the entire infrastructure sector. Some investors view it as a typical cyclical stock—highly dependent on capital expenditures, with long return cycles and sensitivity to interest rates. In recent years, such assets have generally been subject to valuation compression. They worry that once the AI boom fades, or if the technological route undergoes a revolution (such as the emergence of quantum computing or more efficient algorithms), the massive data center assets may face impairment risks. Furthermore, competition in the data center industry is becoming increasingly fierce. The three major telecom operators, leveraging their network and financing cost advantages, are increasing their investments, which may squeeze the market space for neutral operators.
On the other hand, another group of funds has begun to view it as a "core infrastructure asset" in the AI era, similar to data center and optical module companies in the early cloud computing cycle. Once demand explodes, it will usher in a valuation reconstruction. They believe that computing power is the "water, electricity, and coal" of the AI era. No matter which large model company wins, it needs to consume computing power and lease data centers. This "pick-and-shovel" logic endows data centers with the ability to traverse cycles. As long as the general direction of AI development remains unchanged, the long-term demand for data centers is guaranteed.
The key to this divergence lies in one question: Can AI demand remain sustainable and strong enough to cover its high capital expenditures? If the answer is yes, GDS Holdings will transform from a "capital-intensive company" to a "toll booth in the computing power era"; but if AI demand falls short of expectations or fluctuates, its high-leverage expansion model will once again become a source of risk. Additionally, geopolitical factors cannot be ignored. Supply restrictions on high-end computing chips may affect the expansion speed of downstream customers, which in turn will transmit to data center demand. Meanwhile, compliance requirements for cross-border data flows may also affect the layout of multinational enterprises' data centers in China.
From the perspective of valuation models, the market is attempting to shift from traditional P/E valuation to EV/EBITDA valuation to more accurately reflect its pre-depreciation cash flow capabilities. But no matter how it switches, the interest rate environment is always the Sword of Damocles hanging overhead. Data centers operate with high debt, and even minor fluctuations in interest rates will significantly affect financial expenses. Against the complex backdrop of China's interest rate cut cycle and high U.S. interest rates, GDS Holdings' ability to manage financing costs will be a key variable determining the speed of its net profit release.
Conclusion: Finding Balance in the Tension Between Assets and the Future
The current value of GDS Holdings lies not in what it has already achieved, but in where it stands: on one end is the capital-intensive cycle that has not yet been fully digested, and on the other end is the just-unleashed AI computing power boom. What the capital market is truly pricing is the distance between the two.
The essence of this game is a contest between the depreciation of the old era and the premium of the new era. If GDS Holdings can utilize the high-gross-margin orders brought by AI to quickly cover the depreciation costs of old assets and complete the optimization of its financing structure, it is expected to break free from the fate of cyclical stocks and become the core foundation of the digital economy. Conversely, if capital expenditures get out of control or demand is falsified, capital-intensive assets will once again become a drag. For investors, the focus should not merely stay on revenue growth, but should closely monitor the efficiency of capital expenditures, the proportion of high-power racks, and changes in financing costs. In an era where computing power is king, data centers are not just buildings for storing servers, but containers hosting the intelligent future. Whether GDS Holdings can fill this container depends on its steadiness and execution amid cyclical fluctuations.