GDS Holdings Limited (GDS)

Executive Summary / Key Takeaways

• China's AI Inferencing Inflection Creates Unprecedented Demand: The domestic tech industry has reached a critical juncture, shifting from remote AI training to latency-sensitive inferencing in Tier 1 markets. This represents a multiyear, gigawatt-scale opportunity that plays directly into GDS's 900-megawatt powered land bank in Beijing, Shanghai, Shenzhen, and Guangzhou.

• Capital Recycling Engine Becomes Unassailable Moat: GDS's pioneering C-REIT and ABS platform transforms capital intensity from a liability into a competitive advantage. The 5-6 year investment cycle generates low-teens unlevered IRRs and 20%+ levered returns, enabling aggressive expansion while deconsolidating debt and reducing net leverage from 6.8x to 6.0x in nine months.

• DayOne International: A Hidden Value Catalyst: With 750+ megawatts committed and a 1-gigawatt target within three years, the deconsolidated international business is ahead of schedule. A planned IPO within 18 months could distribute shares to GDS holders, unlocking an equity stake currently valued at approximately $1.3 billion ($7 per ADR) but likely worth significantly more given 265% EBITDA growth.

• Financial Discipline Meets Aggressive Expansion: Despite deconsolidating assets through monetization, GDS maintained 2025 guidance for 9.4-12.3% revenue growth and 6.4-10.5% EBITDA growth. Pro forma EBITDA actually accelerated 15.4% in the first nine months, demonstrating the underlying business strength.

• Critical Variables to Monitor: The thesis hinges on AI chip supply resolution in China, potential government regulation of data center expansion, and execution of the 152-megawatt hyperscale order requiring six-month delivery. MSR dilution from edge-of-town campuses will pressure per-unit metrics but expand total addressable market.

Setting the Scene: The AI Architecture Shift

GDS Holdings Limited, founded in 2001 and headquartered in Shanghai, operates at the intersection of China's digital infrastructure transformation and the global AI arms race. The company builds and operates high-performance data centers, providing colocation, managed hosting, and cloud services to hyperscalers, internet giants, financial institutions, and telecom carriers. This positioning matters because it places GDS at the nexus of the most capital-intensive and strategically critical layer of the AI stack: the physical infrastructure where inference workloads must reside to meet latency requirements.

The industry structure reveals why this moment is pivotal. China's data center market has long been dominated by state-owned telecom giants—China Telecom (CHA), China Mobile (CHL), and China Unicom (CHU)—who collectively control over 60% of capacity through integrated telecom-data center models. GDS leads the independent colocation segment with an estimated 10-15% share, but that framing misses the point. The real battleground isn't market share; it's architectural relevance. As AI demand explodes, the fundamental design pattern is shifting from centralized training in remote locations to distributed inferencing in Tier 1 markets. This shift is existential for incumbents whose assets are optimized for yesterday's topology.

GDS's strategic framework, established several years ago, anticipated this inflection. The company deliberately concentrated on Tier 1 markets, prioritized backlog delivery, and became highly selective about new opportunities. This discipline created a powered land bank of approximately 900 megawatts—strategic real estate that cannot be replicated because power quotas in these markets are effectively exhausted. AI inferencing is latency-sensitive and requires large sites distributed across economic hubs. GDS's customers are now discussing gigawatt-scale projects, similar to U.S. developments three years prior. The 900-megawatt bank, while substantial, will not be enough to meet demand, positioning GDS as the gatekeeper for the most valuable AI workloads.

Technology, Strategy, and Financial Innovation

The Powered Land Moat

GDS's competitive advantage begins with geography but extends far beyond location. The company's 900 megawatts of powered land in and around Tier 1 markets represents more than capacity—it represents optionality in an environment where new power quotas are "not easy" to acquire. This scarcity creates pricing power, but the real moat is temporal. GDS has shortened its book-to-build cycle from two years to approximately one year, as demonstrated by the 152-megawatt AI-driven order secured in Q1 2025. This order requires delivery within six months and full customer move-in within another six months, a velocity that state-owned competitors cannot match due to bureaucratic decision-making.

The technology differentiation lies in energy efficiency and carrier neutrality. GDS's facilities achieve significantly lower power usage effectiveness (PUE) than telecom incumbents' older assets, a critical advantage as AI workloads drive power density higher. Carrier neutrality matters because hyperscalers and financial institutions demand vendor diversity and direct connectivity to multiple networks—something integrated telcos resist. This combination enables GDS to command premium pricing while reducing operational costs, creating a structural margin advantage that compounds over time.

The Capital Recycling Flywheel

GDS's most underappreciated innovation is financial, not technical. The company completed China's first data center C-REIT IPO in July 2025, which began trading on August 8 at RMB 3.00 per unit and appreciated to RMB 4.375 by November 18—a 45.8% gain. The C-REIT trades at 24.6x projected 2026 EBITDA with a 3.6% implied dividend yield, validating the valuation premium for stabilized assets. This transforms the capital-intensive data center business into a capital-generating machine.

The mechanics are elegant: GDS develops new data centers, ramps them to stabilization, then monetizes them through the C-REIT at 16.9x-24.6x EBITDA multiples. The first post-IPO asset injection, planned for Q2 2026, targets RMB 4-6 billion in enterprise value, nearly double the RMB 2.4 billion IPO size. This 5-6 year cycle generates low-teens unlevered IRRs and 20%+ levered returns, creating a self-funding growth engine. Concurrently, the ABS transaction completed in March 2025 deconsolidated RMB 1.2 billion in debt at a 13.3x EBITDA multiple, further strengthening the balance sheet.

For investors, this solves the fundamental constraint that has historically limited data center REITs: capital recycling. While global peers like Digital Realty (DLR) and Equinix (EQIX) must rely on external funding for growth, GDS can now fund expansion internally while maintaining financial discipline. Net debt to EBITDA fell from 6.8x at year-end 2024 to 6.0x in Q3 2025, and the pro forma ratio adjusting for C-REIT reinvestment would be 5.7x. This deleveraging occurred while the company accelerated growth, a combination rarely achieved in capital-intensive industries.

DayOne: The International Option

DayOne, GDS's deconsolidated international business, represents a free call option on global AI infrastructure demand. After the Series B equity raise in December 2024, GDS's stake was diluted to 35.6%, transforming DayOne from a subsidiary into an equity investee. This accounting change obscures the underlying value. DayOne ended 2024 with 467 megawatts committed and secured 340 megawatts of new commitments during the year. By Q3 2025, total committed power exceeded 750 megawatts, with management confident of surpassing 250 megawatts in 2025 new commitments.

The growth trajectory is extraordinary. In Q2 2025 alone, DayOne added 246 megawatts, contributing to 244% revenue growth and 265% EBITDA growth. The EBITDA margin reached 31% in Q1 2025, "quite remarkable" for a business that began generating revenue only five quarters prior. The development yield is in the low-teens, higher than GDS's China returns, reflecting more favorable supply-demand dynamics in Southeast Asia and Europe.

The strategic significance extends beyond financial metrics. DayOne's expansion into Thailand and Finland—securing over 220 megawatts—demonstrates the ability to pioneer new markets where customers can scale efficiently with short lead times. This capability "truly sets DayOne apart" from competitors who lack the expertise to create markets rather than just serve them. The planned IPO within 18 months, with a Series C raise targeting over $1 billion at a $4-5 billion valuation, could distribute shares to GDS holders, unlocking substantial value. Based on the Series B benchmark, GDS's equity interest is worth approximately $1.3 billion, or $7 per ADR, but this likely understates current value given the accelerated growth.

Financial Performance as Evidence of Strategy

GDS's Q3 2025 results validate the strategic pivot. Revenue increased 10.2% year-over-year while adjusted EBITDA grew 11.4%, but these headline numbers understate the underlying momentum. Pro forma adjusted EBITDA for the first nine months grew 15.4% after adjusting for deconsolidated companies, demonstrating that the core business is accelerating. Gross move-ins reached 23,000 square meters, consistent with the 20,000+ quarterly pace established in Q1 and Q2.

The utilization rate dynamics tell a more nuanced story. Q1 2025 utilization hit 75.7%, Q2 reached 77.5%, and the company is tracking toward nearly 300 megawatts of new bookings for the full year, with 65% AI-related. This matters because it demonstrates demand acceleration at a time when the market was concerned about oversupply. The 152-megawatt hyperscale order—split across Langfang and Changshu—validates the strategy of building inventory in anticipation of AI demand. The six-month delivery requirement and subsequent six-month fill period create a predictable revenue ramp that de-risks the growth trajectory.

However, the monthly service revenue (MSR) per square meter declined 2.6% in Q1 and 1.7% in Q2 year-over-year. Management expects a 3-4% decline in 2026. This dilution reflects two factors: price resets on older contracts and a mix shift toward edge-of-town campuses for AI inferencing. While this pressures per-unit metrics, it expands the addressable market for AI workloads that require large, distributed sites. The unit economics remain "very solid," with stable new-build pricing and development costs near historical lows generating 11-12% cash-on-cash yields. The trade-off—slightly lower MSR for significantly higher total capacity—is accretive to long-term value.

The balance sheet transformation is equally significant. Organic CapEx for the first nine months was RMB 3.8 billion, with full-year guidance of RMB 4.8 billion. However, net of asset monetization proceeds, net CapEx will be approximately RMB 2.7 billion. Operating cash flow is expected to reach RMB 2.5 billion, making the China business almost self-funding. The effective interest rate dropped to 3.3%, reflecting China's favorable rate environment and GDS's improved credit profile. This financial engineering matters because it allows GDS to maintain aggressive expansion while improving leverage metrics, a combination that creates optionality for shareholders.

Outlook, Guidance, and Execution Risk

Management's guidance for 2025—revenue of RMB 11.29-11.59 billion and adjusted EBITDA of RMB 5.19-5.39 billion—appears conservative given Q3 performance. The company is "on track to achieve the midpoint of revenue guidance and at or above the top end of EBITDA guidance," despite deconsolidating assets through the C-REIT and ABS transactions. This implies a deliberate choice to maintain credibility while setting up for potential outperformance.

The 2026 outlook is more telling. Management expects adjusted EBITDA growth in the low-teens, before considering potential mega orders or additional asset monetization. This baseline assumption reflects confidence that the AI demand wave is sustainable, not cyclical. The key driver is the expectation that new bookings "in the coming years could be better" as AI chip supply uncertainties resolve. William Huang's observation that domestic GPU advancements could catch up within 12 months if international supply faces challenges suggests a pragmatic view of geopolitical risks.

Execution risk centers on three variables. First, the 152-megawatt order's six-month delivery timeline requires flawless supply chain management. While management notes that China's supply chain is "not an issue" and construction can complete in 8-12 months, the compressed timeline leaves little margin for error. Second, government regulation requiring approval for projects 7 megawatts or above could slow new capacity additions, though GDS believes its existing 900-megawatt land bank with obtained power quotas will not be impacted. Third, the AI chip supply uncertainty has created a "wait and see" approach for some customers, potentially delaying bookings.

The DayOne trajectory adds another layer of execution complexity. The business is "well ahead of schedule" to meet its 1-gigawatt target, with an internal KPI of at least 500 megawatts of new commitments annually. The Series C raise, driven by performance exceeding expectations, will be crucial for funding this expansion. The IPO timeline—within 18 months—creates a near-term catalyst, but also requires DayOne to maintain its growth momentum while building public-market readiness.

Risks and Asymmetries

The most material risk to the thesis is AI chip supply disruption in China. Management acknowledges that "AI demand was relatively quiet due to the uncertainty of chip supply," with customers evaluating domestic versus imported options based on "performance, technology, availability and other considerations." If domestic GPUs fail to meet performance requirements or if geopolitical tensions restrict access to advanced chips, the AI inferencing demand wave could stall. This risk is mitigated by the fact that GDS's 900-megawatt land bank is technology-agnostic—it's powered shell ready for any compute architecture—but a prolonged delay would impact growth trajectory.

Government regulation poses a second-order risk. Market talk suggests new policies controlling AI data center expansion, requiring approvals for projects 7 megawatts or above and potentially favoring state-owned enterprises. GDS's confidence that its existing land bank is grandfathered appears reasonable given its track record of securing power quotas, but future expansion could become more difficult. This matters because the 900-megawatt bank, while substantial, is insufficient for the multiyear demand outlook. The company's ability to secure additional power quotas will determine its long-term growth ceiling.

Customer concentration remains a persistent vulnerability. While specific percentages aren't disclosed, the 152-megawatt order from an existing hyperscale customer highlights the outsized impact of large deals. If hyperscalers accelerate proprietary data center builds or shift workloads to internal facilities, GDS's growth could decelerate. The risk is mitigated by the carrier-neutral model and the fact that even hyperscalers cannot replicate GDS's Tier 1 land bank, but revenue concentration creates earnings volatility.

MSR dilution is a measurable headwind. The 3-4% expected decline in 2026 will pressure revenue per square meter as edge-of-town campuses comprise a larger portion of new bookings. However, this is an acceptable trade-off if it enables capture of the AI inferencing market, which values capacity and power availability over location premium. The key is whether total revenue and EBITDA growth can outpace per-unit declines, which appears likely given the 300-megawatt booking target.

The DayOne IPO timeline creates both opportunity and risk. While a successful listing would unlock value and validate the international strategy, any delay or valuation disappointment could weigh on GDS's stock. The Series C valuation of $4-5 billion pre-money suggests high expectations that must be met.

Valuation Context

At $33.43 per share, GDS trades at 48.5x trailing earnings and 16.2x EV/EBITDA, a premium to domestic peer VNET (VNET) (12.95x EV/EBITDA) but justified by superior profitability and growth trajectory. VNET's negative net margin (-5.97%) and higher debt-to-equity (3.59x) contrast sharply with GDS's positive margins and improving leverage (1.75x debt-to-equity, down from higher levels). The valuation gap reflects GDS's unique capital recycling capability and AI positioning.

The C-REIT provides a crucial valuation benchmark. Trading at 24.6x projected 2026 EBITDA, it demonstrates that stabilized assets command a significant premium to the development stage. GDS's ability to develop at 13.3x EBITDA (ABS transaction) and monetize at 24.6x creates a value arbitrage that compounds over the 5-6 year cycle. If the first post-IPO injection achieves the RMB 4-6 billion target, it will validate the model and likely drive re-rating of the parent company.

DayOne's implied valuation adds another dimension. The Series B benchmark valued GDS's 35.6% stake at $1.3 billion, but DayOne's 265% EBITDA growth and 750+ megawatt commitment suggest significant appreciation. A successful IPO at the targeted $4-5 billion valuation would imply GDS's stake is worth $1.4-1.8 billion, or $7.50-9.50 per ADR—representing 22-28% of GDS's current market cap. This hidden value is not fully reflected in the stock price.

The balance sheet strength supports the valuation. With $6.4 billion in market cap, $11.16 billion enterprise value, and net debt-to-EBITDA of 6.0x (improving to 5.7x pro forma), GDS has the financial flexibility to fund growth while maintaining discipline. The 3.3% effective interest rate and strong cash generation (RMB 2.5 billion expected 2025 operating cash flow) provide downside protection.

Conclusion

GDS Holdings stands at the intersection of two powerful forces: China's AI inferencing infrastructure build-out and a financial engineering breakthrough that transforms capital intensity into competitive advantage. The company's 900-megawatt powered land bank in Tier 1 markets positions it to capture the multiyear, gigawatt-scale demand shift from remote training to latency-sensitive inferencing. Meanwhile, the C-REIT and ABS platform creates a repeatable, low-cost capital recycling engine that generates 20%+ levered IRRs while de-risking the balance sheet.

The investment thesis hinges on whether GDS can execute on its accelerated deployment timelines while navigating chip supply uncertainties and potential regulatory headwinds. The 152-megawatt hyperscale order, requiring six-month delivery, will be a critical test of operational capability. Success would validate the shortened book-to-build cycle and likely trigger additional mega-orders. Failure would expose execution risk and potentially slow the growth trajectory.

DayOne represents a free call option on global AI infrastructure demand, with 750+ megawatts committed and a 1-gigawatt target within three years. The planned IPO within 18 months could unlock substantial value, potentially adding $7-9 per ADR to GDS's valuation. This hidden asset provides upside asymmetry while the core China business delivers steady, AI-driven growth.

For investors, the key variables are chip supply resolution, government regulation of data center expansion, and execution on the accelerated deployment timeline. If GDS navigates these challenges successfully, the combination of AI demand tailwinds and capital recycling innovation could drive sustained outperformance. The stock's premium valuation reflects high expectations, but the company's unique positioning and financial engineering create a durable moat that justifies the premium. The story is no longer about surviving in a capital-intensive industry—it's about thriving through financial innovation while capturing the most valuable workloads of the AI era.