GDS Holdings

Phase A — Understand the business

Company Overview

Business model. GDS develops, owns/leases, and operates carrier- and cloud-neutral high-performance data centers in mainland China, selling colocation (space, power, cooling, racks) and, to a much smaller degree, managed hosting/cloud services. It is a wholesale + retail hybrid: hyperscale cloud/large-internet anchors take multi-year (3–10yr) pre-committed capacity; financial-institution and enterprise customers take shorter (1–5yr) contracts. Revenue is ~100% service revenue, recurring (IT-equipment resale is a rounding error — RMB4.2M of RMB11,432M in 2025).

Scale (as of 2025-12-31). 668,283 sqm net floor area in service (93.0% committed, 75.5% utilized); 73,994 sqm under construction (66.1% pre-committed); 98 self-developed DCs in service + under construction; ~3.7 GW of developable resources held for future development, mostly in new growth markets. 989 customers served.

Contract structure / payment terms. Two pricing regimes: unbundled (hyperscalers — priced per sqm/kW + metered power pass-through; variable consideration) and bundled (FI/enterprise — fixed per rack; straight-line revenue). A move-in period (6–24 months for anchors) with a stepping-up minimum billable floor — this is the structural reason utilization (75.5%) lags commitment (93.0%): committed capacity ramps to revenue over years. Churn is tiny: ~0.9% in 2025 (1.2% in 2024). Backlog (committed-but-not-yet-utilized) 165,263 sqm at YE2025. NPS 92.5%, client-sat 9.76/10 (company-run survey).

Key customers / suppliers / competitors. Customers: the largest PRC + global public clouds hosted in-facility (Alibaba Cloud, Tencent, etc. — the 20-F anonymizes them as "Customer 1–5," all cloud/large-internet). Suppliers: heavily concentrated — top-5 suppliers were >70% but <80% of opex purchases in 2025, and one single supplier was 60–70% of annual purchases (almost certainly a power utility / construction-EPC counterpart). Competitors: domestic carrier-neutral operators (VNET, Chindata/legacy, Sinnet/21Vianet-adjacent), the state telcos (China Telecom/Mobile/Unicom) as both competitors and intermediary contracting parties, and international neutrals to a lesser extent.

Supply Chain

Commercial-layer files for datacenters are missing (kb/datacenters/wiki/*.md not present) — this lens is built from the 20-F + industry knowledge; where inferred, labeled.

Upstream → GDS → end customer:

• Land & power quota (the true chokepoint): In China, land is secured via land-use rights (up to 50yr owned; 15–20yr leases), and — critically — energy quota under the local "energy conservation review opinion" gates every build. GDS's moat starts here: 3.7 GW of secured developable resources + ~4 GW land bank with "synchronized power quota allocation". Power is sourced from provincial grids; the October 2021 NDRC shift from fixed to market-rate coal power trading pushed power cost partly onto GDS for fixed-price contracts.
• Construction / equipment: In-house design + modular/pre-fab construction; third-party EPC contractors, generator/UPS/cooling/switchgear OEMs. Supplier concentration is extreme (one counterpart 60–70% of purchases) — plausibly a grid utility + a lead construction partner. GDS runs bulk-purchasing with strategic partners/customers to cut equipment cost.
• Chips (the new upstream variable): GDS doesn't buy GPUs, but its AI-capacity demand is gated by its customers' access to domestic accelerators. Management now frames the build plan as relying "entirely on domestic sources" (Huawei Ascend et al.), with imported silicon as "upside". This makes US export-control policy an indirect but material driver of GDS demand.
• GDS → end customer: Sold direct, or through the major PRC telcos as intermediate contracting parties (GDS bills the telco, collects from the telco) — a China-specific channel wrinkle that muddies "who is the customer."
• Financing counterparties (a supply chain of capital): China Life (led the ABS), the Shanghai Stock Exchange (C-REIT venue), Ping An/Huatai (pref), GIC/STT Garnet/Temasek (equity + converts), sovereign-wealth + PE convert buyers. For a business this capital-intensive, capital is the binding input — see Lens 5/11.

Single-source dependencies: (i) provincial power quota — no quota, no build; (ii) the one 60–70% supplier; (iii) two customers = 41% of revenue (Lens 4/13); (iv) the offshore/onshore cash wall (RMB26.2B restricted net assets — Lens 10).

Competitive Advantages (moats)

1. Land-and-power land bank in tier-1 China (the real moat). ~3.7 GW developable + ~4 GW land bank, much of it pre-secured with power quota in supply-constrained established markets (Beijing/North, Shanghai/East, Shenzhen-Guangzhou/South). New entrants can't conjure quota in Beijing; GDS spent 15 years accumulating it. This is a local-scale / regulatory-access moat, not a brand moat.
2. Switching costs / stickiness. Anchor contracts run 3–10yr with early-termination damages up to 12 months' fees; churn ~0.9%; move-in ramps lock customers in for years. Migrating a live hyperscale cluster is painful. Genuine, but standard for wholesale DC.
3. Operating track record + certifications. 25-yr history, 29 Uptime M&O-certified sites, PUE ~1.23 on post-2020 builds — matters for demanding hyperscalers. A quality/reliability moat at the margin.
4. Strategic-shareholder ecosystem. STT GDC/Temasek (Bruno Lopez) + GIC on the register and board — access to global know-how, customers, and (decisively) capital markets. This is why GDS can run the ABS/C-REIT/convert machine that keeps it funded.

Bargaining power — honest read: Weak over customers, mixed over suppliers. Two customers are 41% of revenue and 50%+ of committed area; pricing has trended DOWN for years (the 20-F concedes "the trend in pricing has been downward... due to capacity, competition and customer expectations"). Hyperscalers hold the whip. Over suppliers GDS has scale-buying leverage but is exposed to one dominant counterparty and to grid power tariffs. The moat protects occupancy and cash flow, not price.

Segments

GDS reports one operating segment (China DC services) and does not disaggregate revenue by product line or granular geography in the 20-F P&L — so a full segment EBIT bridge is n/a — not disclosed. What is sourced:

By service type (revenue):

Colocation dominates service revenue; managed hosting/cloud + consulting are the minority (the 20-F doesn't split the exact %). Trend: revenue +5.5% (2024) → +10.8% (2025) — reaccelerating, driven by AI-workload demand and utilization creeping up (73.8%→75.5%).

By geography (area in service, sqm, YE2025):

North + East = 84% of footprint — concentrated in the two highest-value, most quota-constrained metros. Owned vs leased: 43.3% of in-service area is owned buildings, 56.7% leased (long 15–20yr terms).

The DayOne "segment" that isn't in the P&L. The international business is now an equity-method line, not revenue. In 2025 GDS booked RMB717.2M (US$102.6M) share of DayOne results — but that's a net of a RMB1,681.0M dilution gain minus a RMB963.8M share of DayOne's operating loss. Translation: DayOne is still loss-making at the operating line; GDS's reported "equity income" is mostly a paper markup from DayOne raising money at ever-higher valuations, not cash earnings.

---

Phase B — Measure performance

Earnings Result

FY2025 (the print on the shelf).

*"(Loss)/income from continuing operations" here = the operating line before other income/expenses (interest, deconsol gain, tax, equity income).

The single most important thing about the 2025 "first-ever profit": It is manufactured below the operating line. Operating result was a RMB(55.8M) loss (−0.5% margin). The RMB959.4M net profit is built from RMB2,364.1M non-cash deconsolidation gain + RMB715.9M DayOne equity income (itself mostly a RMB1,681M dilution markup) — offset by RMB1,788.9M interest and a RMB1,561.2M asset impairment and RMB469.7M tax. Strip the two one-offs (deconsol gain + dilution gain ≈ RMB4,045M) and the underlying result is a loss. The cash engine (Adj EBITDA RMB5,403.5M, +10.8%) is real and improving; GAAP net income is an artifact of financial engineering. Be precise about which number you're quoting.

Margins & why. Gross margin +110bps YoY to 22.6% on operating leverage (revenue +10.8% vs cost of revenue +9.2%) as utilization rose. But D&A (30.3% of revenue) exceeds gross profit — the classic young-DC-asset drag. Adjusted gross margin (ex-D&A) is a healthier 51.7%.

Balance-sheet flags (YE2025):

• Cash & equivalents: RMB14,306.0M (US$2,045.7M)
• Short-term debt: RMB3,648.9M (US$521.8M)
• Long-term debt: RMB42,561.6M (US$6,086.2M) — incl. convertible bonds RMB12,144.4M
• Total debt ≈ RMB46,210.5M (US$6,608M); net debt ≈ RMB31,904M (US$4,562M)
• Net debt / Adj EBITDA ≈ 5.9x; gross debt/EBITDA ~8.6x. CFO explicitly cites net-debt/EBITDA 6.8x (YE2024) → 4.7x (Q1 2026) — deleveraging via asset recycling.
• CFO (continuing) RMB3,365.3M (US$481.2M) vs capexCapital expenditureMoney spent on long-lived things — buildings, machines, servers — rather than on running costs. RMB4,610.6M (US$659.3M) → still FCF-negative ~RMB1,245M. GDS does not self-fund growth; it funds it with debt + equity + asset sales.

2025 impairment (a recurring red flag): RMB1,561.2M (US$223.3M) long-lived-asset impairment — the second big write-down in three years (RMB3,013.4M in 2023). Management uses a DCF/sub-lease income approach; repeated impairments say some capacity was built ahead of profitable demand.

Q1 2026 (freshest print, off-shelf but material): Net revenue RMB3,367.1M (US$488.1M), +23.6% headline / +7.9% ex one-time; gross profit RMB1,131.5M, +75.5%; net income jumped to RMB2.65B on a further DayOne gain; ~200MW net new bookings in Q1 (340MW YTD).

FY2026 guidance (reaffirmed at Q1): Revenue RMB12,400–12,900M (+8.5–12.8%); Adj EBITDA RMB5,750–6,000M; capex ~RMB9,000M (≈2x 2025); ≥500MW new bookings (500–800MW/yr over the 3-yr plan).

Market reaction / what's priced: Stock ~$29.83, mkt cap ~US$5.98B, near the low of its 52-wk range ($26.97–$48.61). The market is not pricing the turnaround as durable — it's pricing China risk, dilution, and leverage. See Lens 7/12.

Earnings Calls (sentiment trend)

No transcripts on the shelf (transcripts/ empty). Sentiment read from the Q1 2026 call + 2025 commentary.

Q1 2026 tone: assertively bullish, a clear regime-shift from the defensive 2023–24 posture. Recurring language: "well-positioned," "strong balance sheet," "disciplined," "multi-year growth story," "unprecedented" demand. CEO William Huang: "strong demand... driven by the growing needs of AI deployments," customers "planning gigawatt-scale deployments in single clusters."

What management started saying (vs 2023–24 "survive the downturn / deleverage" tone):

• AI as the driver, framed around domestic chips as a feature not a risk — projections "rely entirely on domestic sources," imports = "upside." This is the biggest narrative shift.
• Aggressive re-investment: RMB30–50B over 3 years — the largest program in company history — a 180° turn from 2023's balance-sheet-repair messaging.
• Deleveraging scoreboard: CFO Newman quantifying 6.8x→4.7x net-debt/EBITDA as proof the asset-monetization strategy works, while pre-committing to re-lever to 5–6x.

What they downplay: pricing pressure (called "stable," aggressive bids dismissed as "not normal / not the whole market") and power constraints (framed as solved via the 4GW quota'd land bank). A skeptic notes management needs the AI-demand narrative to justify doubling capex — treat the optimism as directional, not neutral. Trend: tone inflected decisively positive across 2025 into Q1 2026, tracking the AI-capex cycle.

Comps

Peer multiples are **** with source/date; where not cleanly sourced, n/a. GDS's own EV blends China opco + the DayOne stake, which distorts a naive EV/EBITDA — flagged below.

Read: GDS at ~15x EV/EBITDA trades at roughly HALF the US wholesale peers (DLR ~23x, EQIX ~30x) and near where Chinese physical DC assets clear in the domestic REIT market (~13–17x). That is the "China discount" — VIE/HFCAA/geopolitics + leverage — not an operational-quality gap. The kicker: GDS's headline EV/EBITDA understates the discount because ~US$2.2B of the US$5.98B market capMarket capitalisationThe share price multiplied by the number of shares. What the market says the equity is worth. is the DayOne stake (which contributes zero to consolidated EBITDA). Back out DayOne and the core China opco is being valued at a materially lower multiple than 15x on its own EBITDA. Analyst PTs ($50, ~16.5x 2026 EBITDA) essentially bet the discount narrows. ·

5-yr avg ROE / dividend yield: ROE has been negative/erratic (accumulated deficit RMB5,094.7M at YE2025; only just turned GAAP-positive via one-offs) → a meaningful 5-yr avg ROE is n/a — not meaningful (loss-making history). Dividend yield 0% — never paid, none planned.

Stock-Price Catalysts (moves >5% over ~5 years)

Mostly; the pattern is the point.

• 2018 — Blue Orca short report: ADS fell sharply on allegations of inflated occupancy + related-party acquisition prices; GDS + tenant GZIDC rebutted, stock recovered.
• 2021 — China ADR de-rating: GDS collapsed with the broader China-tech/VIE/ HFCAA selloff (regulatory crackdown, Didi, delisting fears) — the dominant multi-year drawdown.
• Mar 2024 (−double digits) & Nov 2024 (−16%): Growth-scare / China-macro prints; the stock traded on China-macro sentiment and rate/leverage fear more than on the operating beat.
• 2025 rally (+~30% on HK line): Driven by (i) the DeepSeek/AI-capex narrative lifting all China AI-infra names, (ii) the ABS + C-REIT + DayOne monetizations validating the deleveraging story, (iii) the first-ever GAAP profit. Then a ~7% pullback as AI-hype cooled.
• Jan 2026 — DayOne Series C / $20B IPO headlines: Bloomberg/Mingtiandi reports of DayOne targeting a ~$20B US IPO + GDS's $385M share sale at ~6.5x MoM — a value-crystallization catalyst.

What the market actually reacts to: in order — (1) China-macro / ADR-delisting sentiment, (2) AI-capex-cycle narrative, (3) balance-sheet/monetization news (DayOne + REIT), (4) the operating print last. This is a sentiment-and-balance-sheet stock, not an earnings-surprise stock. That's why it can beat and still sit at 52-wk lows.

---

Phase C — Judge people & books

Management

CEO/Chairman — William Wei Huang (58), founder since 2001, CEO since 2002.

• Track record: Built China's #1 independent carrier-neutral DC platform from an IT-DR services startup; IPO'd Nasdaq 2016, HK secondary 2020; engineered the DayOne carve-out that turned an over-levered international build into a ~$20B IPO candidate while recycling ~95% of GDS's principal at ~6.5x MoM. Genuinely impressive capital-structure statesmanship.
• Skin in the game / control: Owns only ~2.8% economically but controls ~57.9% of the vote via Class B shares carrying 50 votes each (dual-class terminates only if his stake falls below 2–2.75%). Founder-controlled, entrenched. He was also DayOne chairman until April 2026.
• Capital allocation: A financial-engineer's playbook — fund growth with converts (2029/2030/2032 notes), pref (Ping An, Huatai), ABS (China Life), a C-REIT (16.9x), and DayOne dilution; then use proceeds to deleverage. It has kept a 6–9x-levered balance sheet solvent through a brutal China downturn — the ABS/C-REIT/DayOne recycling is the reason GDS survived. The flip side: relentless dilution + non-cash "gains" flattering GAAP earnings. ROIC has been poor (repeated impairments; only-now-positive net income). Verdict: excellent liability manager, unproven value-per-share compounder.

CFO — Daniel Newman (CFO since 2011), ex-BofA Merrill Lynch / Deutsche / Salomon investment banker. The right CFO for a company whose core competency is capital-markets access; his fingerprints are all over the ABS/C-REIT/convert architecture.

Board quality — a genuine differentiator for a China ADR:

• STT GDC / Temasek deeply embedded — Bruno Lopez (STT GDC CEO), Jonathan King, Sio Tat Hiang; STT Garnet (Temasek) holds 29.2% / GIC (Singapore sovereign) ~11.9%. Sovereign/strategic anchor money.
• Gary Wojtaszek — ex-CyrusOne CEO who built/spun/IPO'd a top-performing US DC REIT; a real DC capital-markets heavyweight on the board (CyrusOne made a US$100M equity investment in GDS in 2017).
• May 2026: appointed Hua (Kathy) Chen, a US CPA, as Audit Committee chair — a deliberate governance/HFCAA-credibility move.

Red flags (Lens 9 scope): dual-class entrenchment; DayOne-related-party web (GDS charged DayOne procurement/management/sales-commission fees, guaranteed ~US$213M of DayOne bank facilities — mostly now terminated; provided long-tenor lease/customer guarantees); founder was on both sides (GDS + DayOne chair) during the value transfer. All disclosed and mostly unwound by 2026, but the structure invites scrutiny (see Lens 13).

Archetype: founder-controlled financial engineer. Fit for a capital-intensive, distressed-then-recovering platform; the risk is that per-share value leaks to converts/pref/dilution even as enterprise valueEnterprise valueWhat it would cost to buy the whole company: its market value plus its debt, minus the cash you would get with it. Often very different from market cap. grows.

Forensic Red Flags

Ground: 20-F + regulatory file. Every figure unless noted.

1. Earnings quality — GAAP profit is non-cash-gain-driven. RMB2,364.1M deconsolidation gain (20.7% of revenue) + RMB1,681.0M DayOne dilution gain (inside equity income) = ~RMB4,045M of "profit" that is paper, not cash. Underlying operations lost money at the operating line (−0.5% margin). The number to trust is Adjusted EBITDA + CFO, not net income. High-priority interpretive flag (not fraud — fully disclosed).
2. Recurring impairments. RMB1,561.2M (2025) + RMB3,013.4M (2023) — capacity built ahead of profitable demand; a check on the "high-quality assets" narrative.
3. Leverage + FCF deficit. Gross debt ~US$6.6B, net debt/EBITDA ~5.9x, FCF-negative, RMB1.79B annual interest. Solvency depends on continued access to capital markets + asset sales — a financing-risk red flag, not an accounting one.
4. Cash-flow vs earnings divergence: actually favorable here — CFO (RMB3,365M) exceeds GAAP-quality net income adjusted for the paper gains, i.e. the cash engine is more real than the P&L. Good sign.
5. Restricted cash / offshore-onshore wall. RMB26,248.4M (US$3,753.5M) restricted net assets; substantially all revenue in RMB; SAFE controls on remitting capital offshore. GDS has never paid a common dividend and can't freely upstream onshore cash to service offshore debt/ADR holders — a structural claim-on-cash risk for the ADR.
6. VIE structure. Operates via VIEs + contractual arrangements (VIEs contributed 97.5% of 2025 net revenue) — enforceability "not tested in a court of law"; the standard, non-trivial China-ADR ownership risk.
7. Customer concentration (two = 41% revenue / 50%+ area) — a receivables/revenue concentration flag (Lens 4/13).
8. SBC modest and falling (RMB283.4M, 2.5% of revenue, down YoY) — not a flatter-the-non-GAAP concern. Good.

Regulatory findings (required sub-section).

• SEC Litigation Releases: None naming GDS Holdings (EDGAR EFTS LR, 2021–2026).
• SEC AAERs: None naming GDS Holdings (EDGAR EFTS AAER, 2021–2026).
• 10-K/20-F Item — Legal Proceedings: the 20-F states legal proceedings are covered under Item 4.B and discloses no material pending litigation with a quantified adverse exposure; "no significant changes" since the audited statements.
• Non-SEC / web: The material adversarial events are short-seller reports, not regulators — Blue Orca (2018) (alleged inflated occupancy via tenant GZIDC + related-party acquisition mark-ups; GDS + GZIDC rebutted, stock recovered) and J Capital Research (separate bear thesis). No FTC/DOJ/consent-decree hits found.
• HFCAA / delisting: GDS was placed on the SEC Commission-Identified Issuers (provisional delisting) list in 2023; the PCAOB's 2022 China access paused the HFCAA timer; GDS added a US-CPA audit-committee chair (May 2026) and holds a HK dual listing (9698) as a delisting hedge. Live tail risk, currently dormant.
• Bottom line: No SEC/AAER enforcement and no material undisclosed litigation as of 2026-07-06 (verified via SEC EDGAR EFTS + 20-F Item 4.B + web). The real "red flags" are structural — leverage, VIE, offshore-cash wall, and the paper-gain earnings — all fully disclosed, plus a live-but-dormant HFCAA tail.

---

Phase D — Project & stress-test

Forward Projection

GDS is a loss-to-breakeven GAAP-EPS story where EBITDA, leverage, and the DayOne mark matter more than EPS — so this lens projects Adjusted EBITDA + a sum-of-the-parts equity value, and treats GAAP EPS as secondary (it's distorted by one-offs). All outputs with arithmetic; inputs labeled.

Anchor (FY2025 actuals): Revenue RMB11,432.3M; Adj EBITDA RMB5,403.5M (47.3% margin); net debt ~US$4,562M; DayOne stake ~US$2.2B; ~195M ADS-equivalent (excl. converts).

Adjusted EBITDA path (base / bull / bear), FY2026–FY2028:

• Base: FY26 RMB5,875M (guidance midpoint ); FY27 RMB6,700M (+14%, utilization ramp on the RMB9B capex vintage); FY28 RMB7,700M (+15%). Margin holds ~47–48% (pricing-down offset by scale).
• Bull: AI bookings run 700–800MW/yr, utilization jumps → FY28 Adj EBITDA ~RMB8,600M; margin to ~49%.
• Bear: China-macro/pricing/oversupply bites, capex builds ahead of demand (repeat-impairment scenario) → FY28 Adj EBITDA ~RMB6,600M; a new write-down; margin slips to ~45%.

GAAP EPS — deliberately not a headline number. With RMB1.6–1.8B annual interest + heavy D&A, core GAAP EPS stays near breakeven even as EBITDA grows; reported EPS will swing on DayOne marks + any further deconsolidation/monetization gains, which are non-modelable. Honest call: base-case FY26 core (ex-one-off) GAAP EPS ≈ ~breakeven to slightly positive; reported EPS likely positive on DayOne gains. Precise multi-year GAAP EPS = n/a — not credibly forecastable given one-off dependence.

Sum-of-the-parts (the number that matters), rough:

• China opco EV at ~14x FY26 Adj EBITDA (RMB5,875M ≈ US$840M) ≈ US$11.8B EV → less net debt ~US$4.56B → ~US$7.2B equity.
• • DayOne stake ~US$2.2B (Series C mark; upside to ~US$4B if the $20B IPO clears and the stake isn't fully sold down).
• SOTP equity ≈ US$9.4B vs US$5.98B market cap → US$48/ADS, roughly the analyst consensus ($50). The gap = the China/leverage/dilution discount the bulls expect to close.

Brier forecast: per --watchlist rules, not logging a our model entry in the unattended loop. (If promoted to a thesis, the scoreable base call would be: "GDS FY2026 Adjusted EBITDA ≥ RMB5,750M," p≈0.80, resolves 2027-04-30 — essentially betting the guidance floor holds.)

Bull vs Bear

Bull case. GDS is the deepest-discounted way to own China's AI-datacenter buildout with a free option on a $20B IPO. The land-and-power land bank (4GW, quota-secured) is unreproducible in tier-1 China; AI bookings are inflecting (200MW/quarter, 1.8GW total, domestic-chip-fed); the asset-monetization machine (ABS/C-REIT/DayOne) cut net leverage from 6.8x to 4.7x while funding growth; Adj EBITDA compounds low-teens; and the DayOne stake alone ($2.2B, headed to a ~$20B IPO) is ~37% of the entire market cap yet contributes nothing to the EV/EBITDA the bears anchor on. At ~15x EV/EBITDA vs DLR ~23x / EQIX 30x, any normalization of the China discount + DayOne crystallization is worth 50–70% — which is exactly what the Strong-Buy consensus ($50 PT) prices.

Bear case (2–3 permanent-impairment risks).

1. The offshore ADR is a claim on trapped cash. RMB26.2B restricted net assets, no dividend ever, SAFE controls, VIE enforceability untested, live HFCAA tail. If US-China decoupling hard-forces delisting or Beijing tightens capital controls, the ADR can de-rate to the HK line or worse regardless of operations. This is a structural cap on the multiple, not a cyclical one.
2. Re-levering into a price-deflationary market. Doubling capex to RMB9B/yr into an industry where GDS itself admits pricing "trends downward" is how you get the third impairment. If AI demand disappoints even 20%, utilization stalls, leverage re-climbs past 6x, and the monetization spigot (which depends on buoyant markets) closes at the worst time.
3. The "profit" is paper. Once the deconsolidation/dilution gains lap, GDS is back to GAAP-marginal on a US$6.6B debt load — the market may re-anchor on that.

Pre-mortem (18 months out, thesis broke): A China-macro/property-adjacent credit wobble + softer-than-hoped domestic-chip supply caps AI bookings; GDS's RMB9B capex lands ahead of demand → a fresh impairment + net-debt/EBITDA back above 6x; a fresh equity/convert raise dilutes at a low price; and DayOne's IPO slips or prices below the $20B whisper, erasing the SOTP kicker. Stock re-tests/breaks the $27 low.

Are multiples too high? No — if anything the core is cheap. The risk isn't the multiple; it's that the denominator (EBITDA) and the DayOne mark are both financing-cycle-dependent, and the discount is structural.

Contrarian view (what the market refuses to see): The tape treats GDS as a levered China-macro proxy and ignores that ~37% of its market cap is a tier-1-crossover-validated (Coatue/SoftBank/Baupost/Citadel-backed) pre-IPO asset marching to a ~$20B listing. A DayOne IPO that lets GDS sell down its stake at IPO prices could retire a huge slug of GDS's offshore debt and reset the whole equity story — a catalyst that is real, dated (confidential filing reportedly near), and under-priced.

Devil's Advocate (short-seller)

You are the skeptic dismantling the bull case.

• Revenue concentration is a loaded gun. Two customers = 41% of revenue, 50%+ of committed area (Customer 1 alone = 37.9% of committed area). If Alibaba/Tencent-scale anchors in-source (they're all building own DCs) or renegotiate, a single decision craters occupancy and covenant headroom. Hyperscaler self-build is the existential competitor bulls under-weight.
• The moat protects occupancy, not price — and price is falling. GDS concedes the multi-year downward pricing trend. A land-bank moat that yields ever-lower unit economics is a melting moat in a capital-intensive, fixed-cost business.
• The whole equity story rests on perpetual capital-markets access. GDS is FCF-negative with US$6.6B debt; it survives by serially issuing converts/pref/equity and selling assets. That works in a bull market. In a China-credit or ADR-delisting shock, the spigot closes and the leverage becomes the story. This is a balance-sheet, not an earnings, short.
• Earnings are engineered. The first-ever "profit" is non-cash deconsolidation + dilution gains; strip them and it's a loss. A skeptic calls the GAAP turnaround cosmetic.
• Related-party lattice around DayOne. Founder chaired both sides; GDS fed DayOne fees + guarantees during the value transfer. Even fully disclosed, it's the kind of structure short-sellers (Blue Orca 2018, J Capital) have targeted before — and could again as DayOne IPO scrutiny rises.
• What must hold for today's price: continued AI bookings at ≥500MW/yr, no China-ADR delisting event, open capital markets to fund RMB9B capex, and a DayOne IPO that clears near $20B. Break any one and the thesis wobbles.
• If growth disappoints 20–30%: utilization stalls → a third impairment → leverage back above 6x → a dilutive raise at a low price → the $50 SOTP collapses toward the China-opco-only value ($25–30). Downside is to the current price or below; that asymmetry is thinner than the Strong-Buy consensus implies.
• Single scenario that permanently impairs: a forced ADR delisting under a reinstated/obstructed HFCAA regime while leverage is high — GDS holders get pushed to the HK line at a structurally lower multiple and lose the DayOne-IPO optionality if it's caught in the same decoupling. Plausibility: low-but-non-trivial and entirely exogenous.

Management Questions (ranked by information value)

1. Post-DayOne, what is your hard net-debt/EBITDA ceiling through the RMB30–50B capex program, and what specifically triggers a pause — a bookings number, a leverage line, or a pricing floor?
2. Walk us through the DayOne monetization endgame: at IPO, how much of your ~19.9% do you sell, and does the cash go to retiring offshore debt or funding China capex? Quantify the deleveraging.
3. Your two largest customers are ~41% of revenue and 50%+ of committed area. What is the renewal schedule on those contracts, and how do you model hyperscaler self-build eroding demand over five years?
4. The 2025 GAAP profit is largely deconsolidation + dilution gains. When those lap, what is your path to cash-based net income — which year, and on what utilization?
5. You've impaired assets twice in three years. What has changed in your build-ahead-of-demand discipline so RMB9B/yr of 2026 capex doesn't produce a third write-down?
6. Pricing "trends downward." At what unit price / power tariff does new AI capacity stop clearing your ~10–11% gross-yield / ~20% ROE hurdle?
7. How much of your AI-capacity demand is contractually contingent on domestic-chip availability, and what's the downside case if Ascend-class supply slips a year?
8. On HFCAA: what is your concrete contingency if the PCAOB timer restarts — full HK primary conversion, and what would ADR holders receive?
9. RMB26.2B of net assets are restricted onshore. How do you service offshore debt + ADR obligations without upstreaming onshore cash, and for how long can the current structure last?
10. The dual-class structure gives the founder 57.9% of the vote on 2.8% economics. What minority-protection commitments exist on related-party deals (DayOne, ABS, C-REIT) beyond audit-committee review?
11. What is the stabilized ROIC on a mature GDS data center today vs your cost of capital, and how has that spread moved over five years?
12. C-REIT/ABS recycling priced at ~16.9x EV/EBITDA — how much more of the portfolio can you monetize before you're selling the crown jewels, and what's the cap-rate sensitivity?
13. How do you defend occupancy if a state-telco intermediary or a hyperscaler shifts volume to a competitor or its own build — what's your real switching-cost lever?
14. What's the capital plan if the equity/convert window shuts for 12–18 months — can the business self-fund at a reduced capex, and what growth do you sacrifice?
15. Five years out, is GDS a China DC operator, a global platform (via DayOne re-consolidation?), or a capital-recycling REIT-like vehicle — which, and why should equity holders (not creditors) capture the upside?

---