GDS HLDGS LTD

Business Model and Scope

GDS develops and operates carrier- and cloud-neutral data centers in mainland China. It supplies secured space, power, cooling, network access and continuous operations to hyperscale cloud providers, large internet companies, financial institutions and enterprises. Customers either contract directly or through intermediaries; the end user pays for reliable computing capacity near networks and other digital infrastructure.

The value chain begins with land or long leases, permits, grid connections, design and construction. Equipment vendors and utilities supply electrical and cooling systems and power. Telecommunications carriers provide connectivity. GDS then operates the facility, monitors performance and bills contracted capacity. Customers install servers or use cloud infrastructure housed there.

2025 net revenue was RMB11.432 billion, up 10.8%. Area in service was 668,283 square meters; 621,188 was committed and 504,843 utilized. Commitment and utilization rates were 93.0% and 75.5%. Another 73,994 square meters was under construction, 66.1% pre-committed. Commitments include some non-binding agreements or confirmations, so they are not identical to enforceable backlog.

DayOne, formerly GDS International, ceased to be a consolidated subsidiary at the end of 2024 and became an equity investment. By the filing date, further financing had diluted GDS's interest to about 19.9%. The operating business analyzed here is therefore mainly mainland China data centers plus an equity claim on DayOne, not the earlier consolidated international footprint.

The Cayman parent uses mainland subsidiaries and variable-interest-entity arrangements. Public ADS holders own shares in the offshore holding company, not direct title to every operating license or asset.

Customers and Purchasing Decisions

Hyperscale and internet customers can build their own data centers, lease from competing wholesale or retail operators, use another cloud region, optimize server utilization or delay deployment. Financial and enterprise customers compare colocation providers, private facilities and public cloud. They choose on power availability, location, latency, carrier access, uptime, security, expansion capacity, price and counterparty strength.

Switching a live deployment can require migration planning, new network connections, duplicate capacity and operational risk. Multi-year agreements—typically three to ten years for large cloud and internet customers—raise practical switching cost. Yet new workloads can be directed elsewhere without moving installed equipment, so renewal and expansion are more contestable than physical occupancy suggests.

Customer concentration is extreme. Two customers generated 29.0% and 12.0% of 2025 revenue and represented 37.9% and 11.8% of area committed. The top five committed shares totaled 70.4%. A large customer can negotiate price, phasing and specifications and can leave purpose-built capacity underutilized if deployment slows.

Carrier neutrality and dense customer clusters can create ecosystem value because customers want multiple network choices and proximity to cloud nodes. The economic proof is high utilization, renewal and returns after facility capital. A 93% commitment rate alongside only 75.5% utilization shows a timing gap: signed or indicated demand does not yet produce full revenue.

Profit Creation and Value Capture

Revenue is utilized capacity multiplied by contracted pricing plus ancillary services. 2025 utilized area rose from 453,094 to 504,843 square meters, driving revenue growth. Cost of revenue was RMB8.847 billion, or 77.4% of sales. Utilities were RMB3.995 billion and depreciation and amortization RMB3.212 billion. Power usage grows with customer load; depreciation, leases, staffing and minimum power obligations are substantially fixed.

This creates strong operating leverage. Filling an already powered hall adds revenue with limited new building depreciation, although variable electricity rises. Conversely, delayed move-in leaves capital and fixed costs producing little revenue. Unit economics must be assessed per megawatt or square meter through the full facility life: contract margin less energy, operations, rent, maintenance, taxes and the cost of land, construction and financing.

Net income was RMB959.4 million, but it is not a proxy for data-center operating cash. The result included a RMB2.364 billion gain from deconsolidating DayOne and RMB715.9 million of equity-method earnings, while interest expense was RMB1.789 billion. GDS also recognized RMB1.561 billion of data-center impairment after lowering pricing and slowing move-in assumptions. Those items expose both the non-operating source of reported profit and the sensitivity of property values to utilization forecasts. Cash from continuing operations was RMB3.365 billion. Continuing-operation capital expenditure was RMB4.611 billion; accounting operating cash therefore did not fully fund expansion. Property and equipment and right-of-use assets dominate asset quality and require impairment when utilization or pricing forecasts fall.

Working capital is less important than capital-in-progress and financing timing. Customers may prepay or provide deposits, while construction vendors and equipment suppliers extend payables. This can reduce near-term cash use without changing the economic cost of a facility. Incremental returns require pre-commitment to become utilization at pricing that covers both building and debt cost.

Industry Structure and Capital Cycle

Data-center entry requires sites, grid allocations, permits, capital, technical reliability and customer trust. In top markets, land and power are scarce. Nevertheless, large developers, telecom carriers, infrastructure funds and customers themselves can build capacity. Standardized servers make the computing workload mobile even when a specific installation is not.

Utilities and local authorities have high bargaining power through power price, connection timing, energy quotas and permits. Equipment suppliers gain power during capacity shortages. Hyperscale customers wield high buyer power because their deployments are large and they can sponsor competing facilities. Smaller enterprises have less leverage but can shift toward cloud.

The capital cycle is long: demand expectations prompt land and shell construction years before utilization. Easy infrastructure credit can create excess capacity, lower pricing and impair assets. Power scarcity or AI-driven demand can produce the opposite—pre-commitment and attractive incremental fill. GDS's area under construction fell while commitment remained high, which may improve discipline, but new competitors and customer self-build remain.

Exit is costly because a data center is location- and power-specific. Equipment and buildings can be repurposed only at a discount. Asset-backed securitization or REIT transactions can recycle capital, but also transfer future cash flows and add related-party or service obligations.

Sources and Durability of Competitive Advantage

Potential advantages are clusters in major Chinese markets, secured power, relationships with hyperscalers, operational history and carrier-neutral connectivity. The mechanism is lower deployment risk and faster expansion: customers can add capacity inside a known operational ecosystem, while GDS can share networks, staff and procurement across nearby facilities.

Scale can improve financing and purchasing but is not valuable if funded above facility returns. Two customers account for almost half of committed area, making relationship strength a concentration exposure as well as an advantage. A large installed base can lower renewal friction, yet customers can direct incremental workload to competitors.

Replication is possible with land, power and capital; substitution comes from self-build, public cloud architecture and better server efficiency. Technology can raise rack density, making old cooling and power designs obsolete or reducing square meters per unit of compute. Regulation can limit cross-border capital, VIE enforceability, data handling and energy use. New carrier routes can weaken established network density.

Disconfirming evidence would include commitments failing to become utilized area, falling revenue per utilized square meter without compensating cost reductions, loss of a top customer, impairment of recently built centers, power restrictions, or recurring capital expenditure above operating cash with no improvement in per-share cash generation.

Operating System and Strategic Trade-offs

GDS selects sites, secures land and power, designs and builds halls, obtains permits, contracts customers, installs mechanical and electrical equipment, commissions facilities and then operates them continuously. Sales phasing must match construction and customer deployment. Power procurement, preventive maintenance, physical security, network management and incident response determine uptime.

Building before commitment preserves delivery speed but risks idle capital; requiring pre-commitment lowers risk but may surrender scarce market opportunities. High rack density improves revenue per area while raising cooling and power-design requirements. Leasing sites reduces upfront land cost but creates long obligations and renewal dependence. Owning centers provides control but concentrates capital.

The China structure adds legal and cash-flow layers. Operating entities and VIEs earn cash and service debt; statutory reserves, debt covenants and foreign-exchange rules can restrict dividends to the Cayman parent. Parent convertible notes and preferred shares must be serviced from parent liquidity or upstream distributions, which may not be fully fungible with project cash.

DayOne's deconsolidation recycles capital and separates international funding, but leaves GDS with minority-value and dilution exposure rather than direct control of cash. Any REIT or asset-backed scheme similarly needs evaluation of sale price, retained service economics, guarantees and the quality of assets left behind.

Financial Resilience

Consolidated cash and cash equivalents were RMB14.306 billion, restricted cash RMB130.3 million and time deposits RMB667.7 million. Bank and other borrowings were RMB26.315 billion: RMB411.4 million short-term, RMB2.540 billion current portions of long-term debt and RMB23.363 billion noncurrent. Credit facilities totaled RMB28.329 billion, of which RMB3.563 billion was unused; drawdowns remain subject to lender approval and conditions.

The parent also had RMB12.144 billion of noncurrent convertible bonds. Outstanding U.S.-dollar notes comprised $620 million due 2029 at 0.25%, putable in March 2027; $580 million due 2030 at 4.50%, putable in January 2028; and $550 million due 2032 at 2.25%, putable in June 2029. The puts move the economic refinancing dates well ahead of legal maturity. Project loans generally run one to 15 years, are largely secured and carry varied bank rates; 2025 short-term borrowing averaged 2.20%. Rising Chinese or U.S. rates affect floating project debt and refinancing.

Cash exceeds the next year's bank current portion but not total bank, convertible and construction claims. Project collateral and entity-level covenants can prevent free transfer. A $300 million Series B preferred issuance completed in February 2026, disclosed in the annual filing, added liquidity but also preference, dividend and eventual conversion or redemption claims.

A severe scenario combines loss of the largest customer's expansion, utilization falling ten points, a 20% power-cost increase, delayed collections, closure of offshore convertible markets and tighter project-loan covenants. Fixed depreciation continues; operating cash could fall below capex; 2027 and 2028 note puts would approach. GDS could slow construction, draw conditional facilities, sell or securitize assets, monetize DayOne or issue equity. These actions protect liquidity but may crystallize discounts or dilution. Resilience is supported by cash and contracted occupancy but limited by secured project debt, parent puts and customer concentration.

Capital Allocation and Shareholder Outcomes

2025 continuing capex of RMB4.611 billion exceeded continuing operating cash of RMB3.365 billion. Reinvestment is rational only where contracted pricing and utilization produce returns above project and parent financing cost. Asset recycling and the DayOne stake can fund growth, but proceeds must be compared with cash flows surrendered.

The company issued $550 million of 2032 convertibles and ordinary shares in May 2025. Class A shares outstanding rose from 1.512 billion to 1.607 billion, while Class B shares remained 43.59 million. The 2025 basic weighted denominator was 1.521 billion and diluted was 1.645 billion; 99.2 million shares from the 2029 convertibles and 24.82 million restricted shares were dilutive. Share-based compensation was RMB283.4 million.

There was no substantive common-share repurchase or common dividend offset. Preferred dividends were paid, and the 2026 Series B issue added a senior equity claim. Issuing shares to a depositary for award settlement or convertible hedging may be treated specially in accounting, but future delivery still matters to each existing ordinary share.

Debt reduction can lower refinancing risk; new construction can create more value if pre-committed returns are superior. DayOne financing diluted GDS to about 19.9%, illustrating that enterprise growth and per-share ownership growth differ. The appropriate scorecard is stabilized cash flow and asset value per fully diluted ordinary share after project debt, parent converts and preferred claims.

Legal and Regulatory Exposure

PRC corporate structure, VIE and foreign-listing rules — medium probability, very high severity, potentially long duration and incompletely reversible. Authorities could change rules on contractual control, cybersecurity review, overseas listings or foreign ownership. The channel is loss of control or cash rights, forced restructuring, fines, delisting or capital trapped below the parent.

Data security, cyber and uptime — recurring medium probability, very high event severity, duration from hours to years and partly reversible. Outage, fire, cooling failure or breach can trigger credits, claims, lost customers and regulatory action. Redundancy can restore service; trust and corrupted data may not be recoverable.

Power, land, construction and environmental permits — high probability of continuing constraint, high severity, multi-year duration and partially reversible. Grid allocation, energy quotas, emissions rules or defective land and building rights can delay utilization or close capacity. Moving a data center is economically difficult.

Customer, financing and related-structure exposure — medium probability, high severity, multi-year and partly reversible. A hyperscaler can renegotiate or defer deployment; project lenders can accelerate after covenant or cross-default events; asset-recycling and DayOne transactions can create valuation or conflict questions. New customers and refinancing can repair future economics but not idle-period losses.

Conclusion, Uncertainties and Disconfirming Evidence

How is value created? GDS converts scarce sites, power and operating capability into reliable, utilized data-center capacity under multi-year contracts. The critical return occurs when committed halls fill and their cash exceeds power, operation, depreciation-equivalent replacement and financing cost.

Why can it retain value? Cluster density, carrier neutrality, secured power and customer familiarity reduce deployment risk. Customer concentration and replicable construction give large buyers substantial bargaining power.

How durable is it? Physical sites and network relationships are long lived, but technology, self-build, new capacity, energy rules and customer routing can weaken them. Commitments must translate into utilization to prove durability.

Is it financially resilient? RMB14.3 billion of cash and positive operating cash provide capacity, but bank debt, secured structures and $1.75 billion of convertible notes with 2027–2029 puts create meaningful refinancing needs. The balance is resilient to a short delay, not necessarily a prolonged utilization or capital-market shock.

Do common shareholders receive the benefit? The ordinary denominator increased, convertibles and awards dilute, preferred capital ranks ahead and DayOne ownership fell. Common holders benefit only if stabilized value grows faster than all these claims.

Invalidating evidence includes top-customer loss, commitments cancelled or never utilized, persistent capex above cash without rising stabilized returns, material facility impairments, VIE or data-rule intervention, failure to refinance note puts, or fully diluted per-share cash value falling while capacity grows. Business quality and valuation are separate; this analysis makes no claim about the appropriate ADS price.