The investment case for GDS Holdings rests on one proposition: China's AI infrastructure buildout is a demand shock for hyperscale colocation, and GDS is the only Nasdaq-listed pure-play developer of high-density data centers positioned to capture it with a debt-heavy, self-funded buildout.
The most important recent development is the second quarter 2026 interim report, filed in early September, which showed a sharp turn from the prior year's net loss. The swing is not an operating miracle. It is driven by a one-time dilution gain from the completion of DayOne's Series C convertible preferred share financing, a transaction that reduced GDS's stake in its deconsolidated international arm to 19.9 percent. Underlying adjusted EBITDA grew only in the low single digits, and the consequence is that the China colocation engine is decelerating even as the investment portfolio is being monetized.
The central tension is that the company is simultaneously shrinking its asset base in China, impairing long-lived assets where fixed-lease data centers are underperforming, and loading the balance sheet with RMB46.2 billion of total debt to fund new construction. The equity stack sits behind a growing tower of debt, preferred dividends, and convertible overhang. The February 2026 dollar300 million placement to Huatai Capital is the latest addition to that tower.
The near-term catalyst is the conversion of the committed area pipeline into billable power, at the pace implied by management's record sales commitment target. That is a question the third quarter print in late August and the April 2027 annual filing should answer.
GDS Holdings is a Cayman Islands holding company that develops and operates high-performance data centers in mainland China, a structure that requires variable interest entity agreements to consolidate PRC operating assets for United States reporting purposes. The company's revenue comes almost entirely from colocation services, where it leases racks, power, and cooling to cloud service providers and large internet customers under long-term agreements, with managed services and IT equipment sales providing a small supplementary layer. Two customers each accounted for roughly a quarter or more of 2025 net revenue, and the same pair represented nearly half of total committed area at year-end, a concentration that makes the entire revenue base hostage to a handful of hyperscaler renewal decisions.
The strategic pivot of the past eighteen months has been the separation of GDS's international business. In December 2024, the company deconsolidated DigitalLand Holdings, renamed DayOne Data Centers Limited, which operates data centers in Singapore and elsewhere outside the mainland. GDS chose not to participate in DayOne's Series A and Series B preferred share financings, a combined dollar1.9 billion of private capital that diluted its stake from a controlling interest to 35.6 percent. The mechanism matters because it pushed the business into equity method accounting and converted an operating segment, which had posted annual losses of RMB359.4 million in 2023, into an investment whose value is now determined by external financing rounds rather than GDS's own capital allocation.
The consequence for shareholders is a company that is now a pure-play China colocation operator with a minority equity stake in a separately financed international platform. The benefit is a cleaner operating story and access to DayOne's valuation multiples through private placements. The cost is that GDS no longer controls the asset, has limited visibility into DayOne's operations, and depends on DayOne's willingness to repurchase GDS shares or to take an IPO for monetization. The company retained the right to initiate a DayOne IPO, but that right is subject to qualifications in the shareholders agreement, and the timing is not in GDS's hands. The practical upshot is that GDS has traded an operating segment for a mark-to-model investment.
A second structural change came in March 2026, when a special shareholder resolution raised the voting power attached to Chairman William Wei Huang's Class B ordinary shares. The move takes the attached votes from 20 per share to 50. Huang beneficially owns all of the 43.6 million Class B shares. That share of the capital, which also carries the right to nominate five directors, is a standing discount factor for minority shareholders and a governance event that the market is pricing in.
The product is a data center. GDS builds and leases facilities characterized by large net floor area, high power density, and multiple redundancy across critical systems, with a focus on the high-performance segment that cloud providers and AI workloads demand. The company's proprietary data center operation management platform provides real-time monitoring of operating performance, and its sourcing process relies on customer relationships to identify the size, timing, and location of future demand before land and power are committed. The moat is not technology in the software sense; it is the combination of secured power allocations, long-term customer commitments, and the capital intensity that makes the asset base difficult to replicate quickly.
The power constraint is the binding input. In mainland China, new data center projects require energy quota approvals under the energy conservation review process, and GDS's development pipeline is gated on those approvals as much as on customer demand. This gives existing operators with secured power a structural advantage over new entrants, but it also means that the buildout pace is capped by regulatory supply rather than by the company's own capital. The area under construction at the end of June 2026 was 170,355 square meters. That figure, up nearly 29 percent year over year, signals an acceleration in the buildout but also an increase in execution risk before any of that capacity generates revenue.
The customer contract structure creates a second form of moat. Cloud and internet customer agreements are typically variable consideration, meaning revenue is recognized on actual power consumption with a minimum billable amount that steps up over the move-in period. Once a customer passes the move-in window, the agreement converts to a fixed charge for committed capacity plus a usage-based power charge. Financial institution and enterprise customers sign fixed consideration agreements with straight-line revenue recognition. The blend means that revenue visibility is high but not uniform, and the gap between committed area and utilized area, currently 79.2 percent utilization against a 93.0 percent commitment rate, represents revenue that is contracted but not yet billed.
The asset monetization program adds a third strategic layer. In March 2025, GDS entered definitive agreements to sell a 70 percent equity interest in several data centers through an asset-backed securities transaction led by China Life Insurance Company Limited. In July 2025, it completed the sale of two stabilized data centers to a newly established C-REIT listed on the Shanghai Stock Exchange, subscribing for 20 percent of the units. Both vehicles are listed, which means the company is converting owned real estate into listed securities and receiving a mix of cash, contingent consideration tied to asset performance, and a residual equity stake. The mechanism recycles capital without giving up the operating contracts, but it introduces performance-linked contingent consideration and a standing obligation to operate the assets under the original customer agreements.
Full year 2025 revenue was RMB11,432.3 million, up double digits from the prior year. Gross margin expanded to 22.6 percent. Adjusted EBITDA was RMB5,403.5 million, an improvement over the prior year. The GAAP income statement, however, was dominated by a one-time impairment charge on long-lived assets. The charge, driven by lower sales prices and slower move-in at certain data centers with fixed lease terms, pushed operating margin below zero. Net income landed positive only because of a dilution gain from the DayOne Series C financing. The effective tax rate was a product of the impairment-related deferred tax valuation allowance and the intra-group transfer tax in preparation for the potential second C-REIT asset injection.
The first half of 2026 shows the deceleration in its purest form. Adjusted EBITDA was RMB3,354.6 million, up double digits year over year, but that growth is almost entirely a base effect from the DayOne deconsolidation, since the prior year half included DayOne's operating losses within consolidated figures. The second quarter adjusted EBITDA grew only in the low single digits against the year-earlier quarter. The margin compressed to 45.5 percent on higher utility costs as a percentage of revenue. Gross margin fell by more than two points, a direct reflection of new data centers coming into service with fixed utility costs before utilization catches up. That is a dynamic the company itself flags as a recurring pattern during ramp periods.
The balance sheet is the most important statement. Total consolidated indebtedness at year-end 2025 was RMB46,210.4 million. A substantial portion of that is convertible bonds, with the remainder split between borrowings and finance lease obligations. Cash and cash equivalents stood at RMB14,306.0 million. Short-term debt of RMB3,648.9 million and roughly RMB3.6 billion of unused credit sit against that position, which leaves a liquidity cushion that is adequate but not generous for a company that spent heavily on capital expenditures. The February 2026 dollar300 million Series B convertible preferred share placement to Huatai Capital carries a minimum dividend that steps up after six years. It adds a permanent fixed charge and a conversion option that implies meaningful dilution if exercised in full. The company also issued RMB4,907.3 million of new debt financing in the second quarter alone, a pace that keeps the leverage ratio climbing even as operating cash flow improves.
The earnings quality issue runs deeper than the headline net income. The second quarter 2026 net income of RMB837.6 million excludes the DayOne dilution gain, which is an accounting artifact of selling a portion of an investment at a price above its carrying value, not a cash inflow to GDS shareholders. The first quarter 2026 net income was even more heavily influenced by the timing of the Series C closing, a quarter in which consolidated equity method income swung from a year-earlier loss to a large gain. Stripping out these non-recurring items, the underlying earnings power is closer to the adjusted EBITDA line, which is growing in the low single digits at the margin.
The forward variable that matters most is the conversion of committed area into billable power. Total committed and pre-committed area stood at 784,802 square meters at the end of June. The number is up 18.2 percent year over year and 8.2 percent quarter over quarter. Utilized area grew 13.2 percent to 542,236 square meters in the same period. The 79.2 percent utilization rate is improving and up from 77.5 percent a year earlier. It still leaves roughly 143,000 square meters of in-service capacity unbilled. Management has stated that it is on track to achieving a record sales commitment for the year, well above the original target. The gap between commitment and utilization is where margin risk accumulates, since fixed utility costs and depreciation begin when the asset enters service, not when the customer's servers do.
The AI demand narrative is real but unevenly distributed. The company's chairman has identified AI workloads as the primary driver of future demand in China, and the buildout of 170,355 square meters under construction, up 43.9 percent quarter over quarter, reflects that conviction. But the 2025 impairment charge at fixed-lease data centers is a warning that not all committed capacity is created equal, and that the mismatch between contracted pricing and actual demand can produce asset-level losses even within a portfolio whose aggregate utilization is improving. The execution risk is concentrated in the new-build cohort, where GDS has committed capital before full customer absorption and where the energy quota and construction schedule create a narrow window to hit the targeted utilization ramp.
The DayOne investment adds a second forward variable that is outside GDS's control. The 19.9 percent stake is a substantial portion of enterprise value in a business that GDS no longer consolidates. At the Series C offering price as of April 2026, it is worth over dollar2.2 billion. The value is real but illiquid, dependent on DayOne's own financing cadence, the willingness of its board to take an IPO, and the absence of further dilution. The January 2026 repurchase of 11 million shares at dollar35 per share for dollar385 million was a partial monetization, but it also reduced GDS's exposure at a price that was below the implied Series C valuation, a trade-off that favored near-term cash over long-term upside.
The capital allocation path is clear in direction but uncertain in execution. The company has signaled that expansion funding continues through debt, and the convertible bond stack, the preferred share placements, and the C-REIT and ABS asset sales are all tools in the same toolkit. The question for shareholders is whether the debt-financed buildout produces assets that clear their own hurdle rates before the refinancing maturities arrive, starting with the dollar620 million 2029 convertible notes.
The most immediate downside risk is a utilization shortfall on the new-build cohort. If the 170,355 square meters under construction fails to reach targeted utilization within the move-in period, the company faces the same dynamic that produced the 2025 impairment, a write-down of fixed-lease assets whose cash flows no longer support their carrying value. The mechanism is straightforward: depreciation and fixed utility costs are locked in at the moment the data center enters service, while revenue is variable and contingent on customer move-in timing. A second impairment cycle would be a direct hit to net income and a signal that the buildout has outrun the demand.
The leverage risk is the second major downside. Total debt of RMB46.2 billion against an adjusted EBITDA run rate of roughly RMB5.5 billion implies leverage in the mid-8x range. The interest expense of RMB746.7 million in the first half of 2026 is the largest single line item below the operating profit. It is down 11.8 percent year over year on lower rates. A rise in the loan prime rate or a failure to refinance the maturing convertible notes at acceptable terms would compress the equity value directly. The convertible structure provides a partial hedge, since the notes can be settled in shares rather than cash, but the conversion is only an option for the holder when the stock price is above the conversion price, and the company does not have the same discretion.
The customer concentration risk is structural and has been widening. The top customer's share of committed area rose to 37.9 percent at year-end 2025, and the loss or non-renewal of that customer's contracts would represent a revenue cliff that no new bookings could absorb in the near term. The company's churn rate of 0.9 percent in 2025 is low, but it is measured on utilized area, not committed area, and the committed base can contract through non-renewal of pre-committed space that has not yet moved in.
The governance and structural risk is a standing discount. The VIE structure that consolidates the PRC operating assets is a legal fiction under Chinese foreign investment law, and the March 2026 doubling of the founder's voting power to 50 votes per share reinforces the reality that minority shareholders have limited recourse in a governance dispute. The C-REIT and ABS transactions add a layer of regulatory exposure, since the company has made undertakings to regulators and investors that its returns on the sold assets are contingent on those assets meeting performance targets, and a failure to meet them would reduce contingent consideration and potentially trigger financial obligations.
The valuation framework for GDS has to be a sum-of-the-parts exercise because the company is no longer a single operating business. The China colocation segment, the DayOne equity stake, and the residual balance sheet each require a separate lens, and the sum is what the Nasdaq listing prices. For the China colocation segment, the appropriate multiple is enterprise value to adjusted EBITDA, applied to a forward run rate, and the Q2 print sets the denominator. The bear case assumes a further utilization shortfall triggers another impairment cycle, the DayOne stake is marked down to the repurchase price, and the leverage ratio rises as debt grows faster than EBITDA. In that scenario the China segment equity value approaches zero after net debt. The DayOne stake is worth RMB12,000 million, and against a share count of approximately 1.603 billion ordinary shares, that implies a price in the low single digits per ADS. The multiple is enterprise value to adjusted EBITDA, applied to a forward run rate. At a Q2 2026 adjusted EBITDA of RMB1,406.0 million, the company is running at that level. The annualized run rate is approximately RMB5,600 million. Applying a multiple in the range of 6x to 10x, which reflects the company's leverage, customer concentration, and buildout execution risk, produces an enterprise value for the China segment of roughly RMB34,000 million to RMB56,000 million. Subtracting net debt of approximately RMB32,000 million at the mid-point of the range leaves an equity value for the China business that is modest, which is why the DayOne stake is load-bearing for the overall valuation.
The DayOne stake is a private-market mark with no public trading discount. The company values it at over dollar2.2 billion, or approximately RMB16,000 million. That mark is based on the Series C offering price and the 19.9 percent holding. This figure is a private-market valuation with no public trading discount, and a reasonable haircut for illiquidity and the absence of control rights would put the attributable value in the range of RMB12,000 million to RMB16,000 million. The repurchase at dollar35 per share, which implied a lower valuation for the shares sold, suggests that the full Series C price may be optimistic, and the base case should assume the lower end.
The base case assumes utilization continues to improve to the low 80s, adjusted EBITDA grows in the low teens, and the DayOne stake holds at RMB14,000 million. The China segment at 7x forward EBITDA of RMB6,200 million produces an enterprise value of RMB43,400 million. After net debt of RMB34,000 million, the equity value is RMB9,400 million. At 10x EBITDA, the China enterprise value is RMB70,000 million. The spread between bear and bull is driven almost entirely by the utilization assumption and the DayOne mark, which are the two variables that the next two quarterly reports test.
Adding the DayOne stake gives a total of RMB23,400 million, or approximately RMB14.6 per ordinary share. That is RMB117 per ADS, which is below the current trading level and reflects the market's skepticism about the buildout. The bull case assumes the AI demand narrative materializes as a demand shock, utilization exceeds 85 percent, adjusted EBITDA reaches RMB7,000 million, and the DayOne stake re-rates at a premium to the Series C price. After net debt of RMB30,000 million, the equity value is RMB40,000 million. Adding the DayOne stake at RMB18,000 million gives a total of RMB58,000 million, or approximately RMB36 per ordinary share. That works out to RMB288 per ADS.
GDS is a company in the middle of a difficult transition, and the honest read is that the current price is not expensive relative to the bear case but is not cheap relative to the base case. The market is paying for the AI demand narrative and the DayOne stake, and it is discounting for the leverage, the customer concentration, and the governance structure. All three of those discounts are justified on the evidence in the filings, and none of them has a clear catalyst for reversal within the next two quarters.
The counterargument to the bear case is the one that the filings support most strongly: the committed area pipeline is growing faster than the utilization base, and the 18.2 percent year-over-year increase in committed area against a 13.2 percent increase in utilized area means that the revenue base is expanding ahead of the asset base. If the new-build cohort absorbs within the move-in period, the 2025 impairment was a one-time event and the margin compression of the past two quarters was a temporary ramp effect. That is the bull thesis in its cleanest form, and it is not irrational.
The judgment that matters for the investment is whether the debt-financed buildout is being sized to demand that is already contracted, or to demand that is expected. The filings point to the former, with two customers holding nearly half the committed area and a utilization rate that is rising, but the 2025 impairment at fixed-lease assets is a reminder that contracted demand can still underdeliver on price or timing. The company's own language about the record sales commitment target is the most honest forward signal, and it is a commitment that the third quarter print, due in late August, and the full year 2026 filing, due in April 2027, puts to the test.
For a shareholder, the position is one of holding an operating asset that is growing but not growing fast enough to justify the leverage, wrapped in an investment portfolio that is valuable but illiquid, governed by a founder whose voting power was just doubled. The risk-reward is acceptable for an investor who believes in the China AI buildout as a multi-year demand shock and who is comfortable with a balance sheet that is doing most of the work. The risk-reward is not acceptable for an investor who requires visible earnings growth at the equity level, because the fixed charge stack is large enough that the earnings growth has to be significant before it reaches the bottom line.