Sina Finance analysis of GDS Holdings reveals that order backlogs are outpacing recognized revenue, highlighting strong near-term utilization rates across its China-based AI-ready campuses.
The answer points to data center operators. They do not manufacture compute; they construct facilities, supply power, manage operations, and lease racks to cloud providers and internet companies. On August 13, GDS Holdings released its unaudited financial results for the second quarter of 2026. The market responded swiftly, with shares surging intraday to a high of HK$35.16, marking a peak gain of 13.3% the following trading day. At first glance, the financials present a compelling picture: first-half net profit reached approximately RMB 3.49 billion, turning profitable year-over-year, while second-quarter net profit stood at RMB 838 million, successfully reversing a prior-year loss of RMB 71 million.
Operational metrics are even more striking. First-half new contract signings totaled 470 megawatts, surpassing the entire 2025 full-year total within just six months. Management has consequently doubled the full-year sales target to 1 gigawatt. Amid the AI compute demand surge, GDS, acting as the industry’s shovel seller, is experiencing unprecedented historical tailwinds and is being re-priced by the market. Yet the numbers require closer examination. How much of the profit surge stems from core operations? When will the frenzied order book actually convert to cash? What is the true standing of this leading Chinese third-party data center operator in the AI era? Fundamentally, GDS operates as a digital-era landlord, leasing completed facilities, stable power, cooling, and network bandwidth by rack or power draw. Its clients are naturally compute-intensive: hyperscalers, major internet firms, and financial institutions requiring disaster recovery and core system hosting. Contracts typically span five to ten years, ensuring high client stickiness and highly predictable revenue.
This heavy-asset, long-cycle, high-leverage model was established when the company founded in 2000, but AI has fundamentally altered its dynamics. From a financial standpoint, overall revenue has stabilized. The headline profit figures are impressive—a 27.1% net margin—but sustainability warrants scrutiny. More importantly, cash flow is surging, signaling that previously accumulated contract backlogs are finally converting into tangible cash. This is the most solid positive signal in the report. The mechanics are straightforward: data center contracts are predominantly take-or-pay. Regardless of actual utilization, payments are due. Once a client provisions their equipment, the revenue becomes stable and recurring. This is the sector’s defining appeal: substantial upfront investment and long development cycles yield cash flow certainty and continuity that far exceed many other industries once the ramp-up phase concludes.
Viewed holistically, this revenue growth rate is hardly spectacular for an AI compute infrastructure play; it reflects steadiness rather than aggressive acceleration. After 26 years in operation, GDS has transitioned from rapid expansion to a steady climb, aligning with industry norms. This indicates that its core rack-rental business no longer possesses the early-stage momentum of 30%+ annual growth. Instead, it relies on a heavier, accumulation-dependent model where orders lead and cash follows. Meanwhile, gross margins contracted slightly. Second-quarter gross margin stood at 21.5%, down from 23.8% a year ago. Management attributes this to rising electricity costs and the front-loaded depreciation and operational expenses associated with newly commissioned facilities. It is a classic case of costs arriving ahead of revenue.
This dynamic underscores how AI has reshaped the IDC cost structure. While data centers have always been power-intensive, AI workloads have multiplied power density several times over—a single modern rack can replace what used to be an entire row. Consequently, electricity’s share of operating expenses continues to climb. According to the National Energy Administration, national compute center electricity consumption reached 170 billion kilowatt-hours in 2025, accounting for 1.6% of total societal electricity usage, a proportion that remains on an upward trajectory. As GDS expands its operational footprint, its utility bills will inevitably grow thicker. Furthermore, regulatory push for compute-power synergy mandates increasingly higher green power consumption ratios for new facilities. Since green power carries a premium over thermal generation, this introduces additional cost pressure.
Beyond cost structures, AI infrastructure demand has generated substantial new orders, with GDS expanding its client base beyond traditional internet and cloud giants into emerging sectors. If the income statement bears some polish, the operational data is unvarnished and reveals undeniable growth momentum. During the first half, GDS secured significant new orders from all three of its core hyperscale clients and forged partnerships with a cohort of emerging AI leaders. Currently, its hosted workload is split roughly 50% CPU and 50% GPU, with management projecting an even higher GPU allocation next year. These client profiles and demand patterns simply did not exist prior to 2023. In the first half of 2026, GDS signed 471 MW of new capacity, exceeding its 2025 full-year total. A single quarter—Q2—accounted for 260 MW of new signings, prompting management to raise the full-year target from 500 MW to 1 GW. Currently, GDS has locked in 600 MW of resource reservation intent and holds a binding backlog of 757 MW, pushing total committed capacity beyond 2 GW. Management anticipates reserved capacity will exceed 1 GW by the end of 2026.
Approximately half of GDS’s first-half new contracts originated from emerging markets such as Ulanqab, Horinger, and Shaoguan, balancing mature and emerging hubs roughly evenly. In June 2026, GDS signed a strategic cooperation agreement with the Ulanqab municipal government to invest over RMB 30 billion across five years, developing a gigawatt-scale green power direct-connected data center cluster. Through direct grid connections and green power trading, the project aims to achieve over 80% green power coverage. This strategy highlights GDS’s pursuit of scarce western power quotas and green energy credentials, as top-tier cloud vendors have already embedded renewable procurement mandates into their standards.
Translating signed contracts into recognized revenue involves a 12-to-24-month lag encompassing construction, delivery, client provisioning, debugging, and billing. Consequently, the second-quarter revenue of RMB 3.088 billion, representing a modest 6.5% year-over-year increase, primarily reflects projects signed in 2024 or earlier. The explosive 470 MW order intake from the first half will likely drive the true revenue realization peak in the second half of 2027. Today’s financial report is merely the eve; the main stage arrives in late 2027. Current market valuations are not pricing in present-day profitability but rather anticipating the revenue inflection point of 2027.
The traditional IDC business logic was straightforward: secure land, power, and energy consumption quotas in core metropolitan circles, construct facilities, lease them to clients, and capture rental and power arbitrage margins. The primary barrier to entry was locational resource access. However, the AI compute boom has rewritten the rules. Customers no longer seek basic server rooms; they require high-power-density clusters capable of running large language model training. Rack power density has skyrocketed from 4–6 kilowatts to over 20 kilowatts, liquid cooling has become standard, and green power supply is now a hard requirement. Facilities unable to meet these specifications hold little value, regardless of price.
Mature markets—nodes surrounding Beijing-Tianjin-Hebei, the Yangtze River Delta, and the Greater Bay Area—cater to low-latency demands. Emerging hubs like Ulanqab, Horinger, Shaoguan, and Zhongwei leverage abundant green power and lower land costs to host large-scale AI training workloads. The geographic logic is clear: proximity to tier-one cities supports low-latency inference and real-time computing, while western regions accommodate the stringent cost and energy demands of massive training clusters. Yet a critical question remains regarding capital efficiency. It is not a matter of whether capital is available, but whether each dollar deployed generates sufficient returns. The IDC sector typically exhibits suboptimal ROE and ROIC during expansion phases, as depreciation hits the income statement before assets reach full utilization. GDS currently navigates this exact juncture. Projects invested in previous years are progressively delivering and ramping, yet their profit contributions remain unrealized, while fresh order books compel continued capital expenditure.
Ultimately, GDS is betting on a singular thesis: AI compute infrastructure will shift from adequate to constrained. When that tipping point arrives, operators possessing ready-made high-power facilities, secured green power channels, and entrenched client relationships will capture pricing power premiums. The industry does not lack data centers; it lacks high-quality, AI-ready facilities. Inefficient supply is being cleared out, allowing market leaders to absorb incremental demand. Holding the number one market share position inherently constitutes a long-term moat.
Nevertheless, moats require continuous reinforcement. Liquid cooling architectures are rapidly evolving, power distribution systems are undergoing upgrades, and clients are imposing increasingly stringent PUE requirements. Sustaining a first-mover advantage hinges on precisely timing technological shifts within a fast-changing landscape. Falling behind by even one or two years could result in irreversible competitive displacement.