GDS Holdings Limited (GDS) | The Buildout — AI Infrastructure
The Verdict
GDS Holdings is a pure-play provider of physical data center infrastructure for China's hyperscale AI platforms. It acquires land, secures government power quotas, and builds massive, high-density facilities purpose-built for AI training and inference clusters. The company does not supply chips or software; its role is to deliver the space, power, and cooling that AI compute requires at gigawatt scale. This makes GDS a foundational layer in China's AI supply chain — without its campuses, the country's largest tech firms couldn't deploy the GPU clusters that drive their AI models.
| Market Cap | — |
| Revenue (TTM) | $1.7B |
| Revenue Growth | +10.4% |
| EBITDA Margin (TTM) | 46.6% |
| Net Debt | $4.7B |
| Earnings Beats | 7 of 7 |
| P/E (TTM) | — |
| EV/EBITDA (TTM) | — |
What We Like
- Bookings target raised to 500–800 MW per year, with 340 MW already secured in early 2026, signaling multi-year demand.
- Landbank of nearly 4 GW — more than double current total bookings — creates a resource-based moat that is hard to replicate.
- Development cost per kW has declined 15% over three years on a like-for-like basis, improving unit economics even as scale balloons.
- Balance sheet transformed: net leverage reduced to 4.7x from 6.8x, with over RMB 19 billion in cash, enabling the RMB 30–50 billion investment plan.
- Revenue inflection expected in H2 2027 as move-in doubles from 2026 levels, driven by a 600 MW backlog becoming billable.
What We’re Watching
- Whether the >660 MW of reservations convert to firm bookings within 1–2 years, as management expects.
- Power quota conversions for the 4 GW landbank — any delays by grid operators would stall the growth plan.
- Domestic chip supply must meet the forecast; the 2026 plan relies entirely on domestic GPUs, making GDS vulnerable to production setbacks.
- Customer concentration: two customers account for 49.7% of committed area; loss or slowdown by the top one would severely damage the thesis.
The thesis is strengthening: bookings are tracking well above the minimum target, the landbank has quadrupled, and the balance sheet is now in fortress shape, all of which support the company's ambitious three-year buildout. The revenue payoff is still a year away, and customer concentration remains the core structural risk. The open question is whether GDS can convert its overwhelming demand pipeline into on-time, on-budget capacity without a misstep, especially given the silent C-REIT plan and the governance flag of the audit chair's sudden resignation.
Earnings Beat
GDS reported Q1 FY2026 revenue of $488.2 million, up 7.9% year-over-year on a reported basis and 12–13% pro forma after adjusting for deconsolidated assets. Gross margin jumped to 33.6% from 23.7% a year earlier, and EBITDA rose to $252.2 million with a 51.7% margin. The quarter included a large net income gain of $384 million, driven by the partial sale of DayOne equity.
| Metric | Q1 FY2026 | Q4 FY2025 | Q1 FY2025 | YoY |
|---|---|---|---|---|
| Revenue | $488M | $418M | $375M | +30.1% |
| Gross margin | 33.6% | 21.0% | 23.7% | +990bps |
| EBITDA | $252M | $198M | $168M | +49.8% |
| EPS | $1.51 | $-0.34 | $0.47 | +222.9% |
We believe this is the beginning of a multiyear growth story, supported by increasing availability of domestic chips.— William Huang, CEO, May 20, 2026
Management tone: Management struck a notably confident tone, reframing the opportunity as a multi-year, gigawatt-scale buildout while repeatedly stressing financial discipline and order selectivity. They were candid about domestic chip reliance and transparent on unit cost improvements, but sidestepped questions on C-REIT progress and the FY2025 bookings outcome.
Management Guidance
Management reaffirmed full-year 2026 guidance and set a new bookings target of at least 500 MW for the year, with the year-to-date figure at 340 MW already. They unveiled a three-year plan for 500–800 MW of annual bookings and RMB 30–50 billion in total investment. For move-in, they guided to somewhat over 70,000 sqm in 2026 and a roughly double amount in 2027, concentrated in the second half. Net leverage is expected to rise to 5–6x as investment ramps. The 2026 plan is based entirely on domestic chip supply; any import access would be upside.
Trajectory
Reported revenue rose 7.9% year-over-year in Q1 FY2026, held back by the deconsolidation of monetized assets; on a pro forma basis, growth was 12–13%. Gross margin expanded sharply to 33.6% from 23.7% a year earlier, while EBITDA margin reached 51.7%. The revenue trajectory is expected to remain modest through 2026 before accelerating in H2 2027 as a doubling of move-in kicks in. Management attributed the future step-up to a 600 MW backlog and a robust new bookings pipeline.
The Model
The model projects FY+1 revenue of $2,047M and EBITDA of $1,069M (52.2% margin). For FY+2, revenue climbs to $2,487M with EBITDA of $1,331M (53.5% margin). The near-term is anchored by the 600 MW backlog and the 2026 bookings target, while FY+2 reflects the full-year effect of the 2027 move-in doubling and sustained annual bookings of 500–800 MW.
| Metric | FY2025 | Next FY (E) | Following FY (E) |
|---|---|---|---|
| Revenue | $1.6B | $2.0B | $2.5B |
| YoY Growth | — | +27.7% | +21.5% |
| EBITDA | $716M | $1.1B | $1.3B |
| EBITDA Margin | 44.7% | 52.2% | 53.5% |
Projections are the median of 5 independent model runs. The model’s revenue sits 19.7% above analyst consensus.
Management reaffirmed full-year 2026 guidance and set a new bookings target of at least 500 MW for the year, with the year-to-date figure at 340 MW already. They unveiled a three-year plan for 500–800 MW of annual bookings and RMB 30–50 billion in total investment. For move-in, they guided to somewhat over 70,000 sqm in 2026 and a roughly double amount in 2027, concentrated in the second half. Net leverage is expected to rise to 5–6x as investment ramps. The 2026 plan is based entirely on domestic chip supply; any import access would be upside.
What Could Go Right — and Wrong
- Reservations convert to bookings faster than expected, pushing annual bookings above 800 MW and extending the growth horizon.
- Import chip access reopens, allowing hyperscale customers to accelerate GPU deployments, pulling forward move-in timelines.
- C-REIT or a major asset sale is executed, validating a scalable capital recycling model and reducing reliance on debt.
- New customer verticals, such as neocloud providers, sign material contracts, diversifying the client base.
- Power quota conversions for the landbank proceed without delay, enabling seamless capacity expansion.
- Domestic chip supply encounters a bottleneck, delaying customer move-ins and stranding newly built capacity.
- The top customer shifts to self-build or selects a competitor, collapsing the committed area base and triggering asset underutilization.
- Grid operators restrict power quotas for the 4 GW landbank, stalling the expansion plan and impairing the moat.
- VNET's CATL partnership and first-mover REIT advantage allow it to win a disproportionate share of mega-deals, capping GDS's growth at the lower end of guidance.
- Execution miscues during the RMB 30–50B buildout lead to cost overruns and delayed delivery, eroding returns and customer confidence.
Looking Ahead
Over the next twelve months, GDS's primary narrative will pivot on the conversion of its extraordinary demand pipeline into contracted backlog and actual move-in. The FY2026 new bookings outcome is the nearest catalyst, with the full-year figure likely to exceed the 500 MW minimum. Management's credibility will be tested on the C-REIT front and on their ability to manage the rapid construction ramp without cost overruns. Any developments on domestic chip supply or import policy will directly affect customer deployment schedules, and the resolution of the governance overhang — the audit chair vacancy and the law-firm investigations — will shape sentiment.
- FY2026 (Q4 2026 call)FY2026 bookings outcome — Whether new bookings exceed the 500 MW minimum; YTD pace suggests upside.
- 2H 2026Asset monetisation update — Any move to revive C-REIT or sell additional DayOne equity.
- 2026Audit chair appointment — Permanent replacement for Lim Ah Doo; signals governance stability.
- Mid-2027Backlog conversion — Majority of 600 MW backlog expected to become billable.
- H2 2027Move-in step-up — Revenue inflection expected as move-in doubles from 2026 level.
- 2027–2028Reservation conversion — >660 MW of reservations expected to convert to firm contracts.
Financials
Annual Summary
| Metric | FY2025 | TTM |
|---|---|---|
| Revenue | $1.6B | $1.7B |
| Gross Margin | 22.6% | 25.5% |
| EBITDA | $716M | $1.4B |
| EBITDA Margin | 44.7% | 46.6% |
| Net Income | $130M | $409M |
| Free Cash Flow | −$186M | −$385M |
| Net Cash | — | — |
Key Ratios (Trailing)
- P/E TTM—
- EV/EBITDA TTM—
- EV/Revenue TTM—
- Price/FCF TTM—
- Gross Margin (TTM)25.5%
- EBITDA Margin (TTM)46.6%
- Net Margin (TTM)23.8%
- ROIC2.5%
- FCF Conversion-25.1%
- SBC / Revenue2.5%
The Company
GDS Holdings is a leading developer and operator of high-performance data centers in China, providing colocation services and, to a lesser extent, managed services and IT equipment sales. Its facilities are the physical foundation for the GPU clusters that power AI training and inference at the country's largest technology platforms.
The business is asset-heavy: GDS acquires land, secures power quotas from state monopoly grids, designs and builds large-scale facilities, and leases capacity to hyperscale customers. It operates across multiple regions through subsidiaries and VIEs, and it holds an equity interest in an international vehicle called DayOne, which it uses for asset monetization.
Business Segments
Competitive Landscape
GDS operates in a highly competitive landscape where it competes with state-owned telecom carriers that also build data centers, and with direct carrier-neutral peers VNET and Chindata. The company's principal advantages are its secured landbank, government relationships, and operational scale. However, VNET is matching GDS on build cost and has gained a potential edge through a CATL energy partnership and first-mover REIT monetization.
- China TelecomState-owned carrier that supplies connectivity but also competes in colocation; named as a competitor in the 20-F.
- China UnicomState-owned carrier that supplies connectivity but also competes in colocation; named as a competitor in the 20-F.
- China MobileState-owned carrier that supplies connectivity but also competes in colocation; named as a competitor in the 20-F.
- Most direct listed peer; wholesale revenue surpassed retail; won a 510 MW mega-order; build cost identical to GDS at ~RMB 20k/kW; backed by planned CATL strategic investment.
- ChindataCarrier-neutral peer; smaller scale; competes in the hyperscale data center market.
Supply Chain
GDS sits at the center of a chain that begins with land and power grids and ends with China's largest tech platforms deploying AI. It depends on monopoly grids for electricity, on carriers for connectivity, and on a global network of MEP equipment suppliers.