China Datacenter Model
Every datacenter building in mainland China that matters, tracked one by one: 1,000+ facilities across 60+ players, with annual / quarterly capacity tracking from 2017 to 2032. Built to the same bottom-up standard as the global model, for the 2nd largest datacenter market in the world.
- Buildings
- 1,000+
- Operators
- 60+
- Coverage
- 2017–2032
- Updates
- Quarterly
The second largest market, building by building.
The SemiAnalysis China Datacenter Model extends our datacenter coverage to mainland China, the second largest datacenter market in the world and the one the global model does not reach. It is built to the same standard, bottom-up and building by building. Each row is one development phase, carrying its operator, its tenant where attributable, its capacity, its location, and its construction and operations dates.
While APAC excl. China has been ramping up fast to reach 15GW capacity as of 2026YE, China alone is on track for a 24GW+ fleet by end-2026, second only to the United States. Its largest hyperscaler and biggest spender discloses no financials, and the two US-listed colocation operators that do disclose represent less than 15% of the market. Headline numbers are eye-catching and often misleading. The model replaces them with operating, under-construction, and dated planned megawatts, quarter by quarter.
Who uses it, and what it decides.
The buyers of this model, and the calls they make with it.
- Hyperscalers & AI labs
- A reference point for how fast China builds AI infrastructure: how quickly ByteDance, Alibaba, Tencent, Baidu, and Huawei add capacity, how much of it is self-built versus leased, and where it sits. District-level tracking shows how much capacity sits in the western hubs favored for training versus sites around tier-1 metros suited to inference, how provinces compete to win that compute, and how that may influence overall AI competitive offerings.
- Hedge funds & public market investors
- GDS and VNET, plus the A-share landlords few models cover, and the private operators that file nothing yet hold most of ByteDance’s leases. Who is winning hyperscaler orders, how fast they build and deliver, and how concentrated the customer book is.
- Equipment & power vendors
- Where and when the next gigawatts land, by province, hub, operator, and quarter, with an adjustable capex-per-megawatt envelope to size the addressable spend.
Five sections, one supply ledger.
Zoom in to the supply side of China datacenter market, at district and cluster level. The dashboard’s nine tabs are grouped into five sections here; every one reads from the same building-level rows, so capacity, tenants, capex, and load stay consistent with each other.
Supply analytics, 2017 to 2032
Delivered capacity with an annual ramp from 2017 to 2023 and quarterly from 2024 through 2032, as absolute megawatts or net additions. Every chart can be cut by:
- EDWC hub
- Market
- Province
- Type
- Status
- Owner
- Tenant
- Grid
Only dated, confirmed pipeline is charted: planned capacity with no operations date and no ramp is held out as speculative megawatts, so the later years read as a floor, not a slowdown forecast.
Companies, hyperscalers, and AI labs
Who owns each building and who fills it. Owner snapshots for over sixty operators, hyperscaler views that combine self-build sites with leased capacity, and chip-implied megawatts for the AI labs:
- HyperscalersSelf-built and leased facilities of ByteDance, Alibaba, Tencent, Baidu, Huawei, Kuaishou, JD, Meituan and moreLease + self-build capacity
- Colocation operatorsA comprehensive list of facilities owned by the country’s leading colocation players, with key tenants behind major campusesOperating, under construction, and planned capacity
- Telecom carriersMajor facilities owned by China Mobile, China Telecom, and China Unicom, who used to operate over 60% of China’s datacenter projectsCarrier owned facilities
- AI labsSelected major AI labs with capacity implied from their installed chip countsChip-implied MW
Leasing: signing and delivery
Two trackers for the leased half of the market. The delivery tracker follows leased megawatts as they come live, while the signing tracker dates the lease as disclosed or estimated, so commitments show up when they were made:
Delivery tracker
Leased capacity by landlord and tenant as it energizes, quarter by quarter
Signing and full commitment tracker
Leases signed and self-build starts put together each year, viewed against capex trends
Capex and load growth
The money and the power implied by delivered megawatts. Capex runs at a unit cost, split into power, cooling, shell, and other, and timed three ways: as scheduled, at commercial operation, and as cash. Load growth applies a PUE ratio, a load factor, and a utilization ramp to the same rows.
A reported-capex overlay sets each hyperscaler’s disclosed or estimated spend against its datacenter trend, so the money can be read next to the megawatts.
Map, explorer, and model builder
Three ways to explore in detail besides the charts:
Map
Filter by type, status, region, and operator, with province and district heatmaps, a distance measure, and expansion over time
Explorer
Every building as a row: company, province, city, campus, status, tenant, and megawatts by status, with its 2017 to 2032 ramp
Excel model builder
The full model as a workbook, cut, extract, or filter the way you need it
Eastern data, western compute.
The top-level design blueprint of China datacenter industry set eight national computing hubs and ten datacenter clusters. Every building in the model is tagged to its hub, so the buildout can be read against the plan, including where it departs from it.
West hub · power-sensitive training
Hohhot–Ulanqab
The fastest-growing hub, the Johor of China. Growth is heading northwest and most of it lands here. ByteDance is starting to ramp self-build sites in the region, and Huawei Cloud has run Ulanqab since 2016 with Horinger added in 2025.
- ByteDance
- Alibaba
- Kuaishou
- Huawei Cloud
- Tencent
- GDS
- VNET
East hub · inference near population
Zhangjiakou
The capital region’s cluster, close to Beijing’s demand. Capacity today is still concentrated around Tier 1 cities and increasingly inside eastern hubs like this one.
- ByteDance
- Alibaba
- Tencent
- Hoyinn
- Chindata
- VNET
- ZDATA
Outside all eight hubs
Shanxi
The exception. Shanxi sits outside every hub yet hosts one of China’s largest live fleets, anchored almost entirely by ByteDance. The model records it as it is, not as the plan drew it.
- Chindata
- ZDATA
- ByteDance
West hub · power-sensitive training
Zhongwei
A western hub built on cheap power and land. Training workloads are power-price sensitive, which is why they concentrate in the north and west.
- Alibaba
- Meituan
- AWS
- Telecoms
- Centrin Data
- Chindata
- ZDATA
West hub · power-sensitive training
Qingyang
The westernmost of the eight hubs, small today, yet a rising city of token factories.
- VNET
- Chindata
East hub · inference and cloud near population
Wuhu
Shanghai’s next-door capacity, one hour inland. Hyperscaler self-builds and carrier parks make it the delta’s largest pipeline.
- Huawei Cloud
- ByteDance
- Tencent
- Telecoms
East hub · inference and cloud near population
YRD Demo Zone
Shanghai’s inference zone, next to the population it serves. Quota-constrained and largely built out. The delta’s growth now flows to Wuhu.
- Alibaba
- Tencent
- Telecoms
East hub · inference and cloud near population
Tianfu (Chengdu) and Chongqing
The largest cluster in middle China, serving a large population base in Chengdu, Chongqing and nearby cities, with abundant hydropower.
- Alibaba
- Tencent
- Telecoms
West hub · cold storage
Gui’an
The cold storage cluster of China and Huawei Cloud’s first national hub, live since 2021. Huawei self-builds nearly all of its capacity and delivers only a fraction of announced, due to positioning.
- Huawei Cloud
- Tencent
- Apple
- Telecoms
East hub · inference near population
Shaoguan
The Greater Bay Area’s cluster, two hours north of Guangzhou and Shenzhen, and the only place in Guangdong still issuing large scale datacenter energy quota.
- Tencent
- Chindata
- GDS
- Telecoms
Find it in the model
- Supply analyticsby EDWC hub
- Maphub filter, province and district heatmaps
- ExplorerEDWC zone on every row
Bubble sizes on the model’s own map are weighted by live and dated pipeline capacity; this schematic only places the clusters.
How the model is built.
Every megawatt starts as a public footprint and ends as a dated quarterly ramp.
-
Read every building
Announcements, filings, construction news, and operator disclosures give each development phase its owner, its tenant where attributable, its location and hub, and its construction and operations dates.
-
Convert to IT megawatts
Reported IT capacity where a source states it. Otherwise a fixed order of preference: rack counts at the quoted or estimated density, investment size at unit capex, and sometimes net floor area at a conversion ratio.
-
Classify, ramp, allocate
Each phase is Operating, Under Construction, or Planned. Speculative megawatts, planned with no operations date, stay out of every chart. Delivered capacity follows a quarterly ramp; major tenants are identified building by building or allocated per colocation customer mix; lease signings are dated by disclosure or estimated from the construction timeline.
Common questions.
Anything not covered here, ask the team directly through the form below.
What does the China Datacenter Model cover?
More than a thousand datacenter buildings in mainland China across over sixty players, with an annual capacity ramp from 2017 to 2023 and a quarterly ramp from 2024 to 2032. Colocation, telecom, hyperscaler self-build, and financial-institution capacity are all in scope. Each row is one development phase, carrying its owner, tenant where attributable, location, EDWC hub, and construction and operations dates.
How is it different from the Datacenter Industry Model?
Same bottom-up approach, and the same core tabs, metrics, and views. Key differences: 1) this model is supply-side focused; 2) it does not include some of the new functions we introduced specifically for the global Datacenter Industry Model (the neocloud tabs, the HPC Lease Tracker, Datacenter Developers, Cryptominers, Datacenter Backstops, the Vision Model, or the grid view on the map); 3) it adds a country-specific EDWC hub analysis tab, and a hyperscaler commitment tracker that compares leasing and self-build start trends with reported capex; 4) locations are given at the district level, and there is no satellite view.
What is likely under-captured?
Small projects under 30 megawatts, captive government and enterprise datacenters, edge nodes, and very early announcements. Financial-institution self-builds are not covered comprehensively. Announced phases with no construction visibility are recorded but labelled speculative and kept out of every chart.
How often is it updated?
A major update every quarter. As with our global Datacenter Industry Model, we will keep adding layers with every update. Subscribers get the map today and ongoing tracking of China’s AI buildout.
How is it delivered?
We present a live dashboard with supply analytics, company and hyperscaler views, leasing trackers, EDWC hub analysis, capex and load growth, an interactive map, and an explorer table, plus a full Excel model builder and the methodology notes.
Is it part of the SemiAnalysis newsletter subscription?
No. Industry models are separate institutional offerings and are not included with the annual newsletter membership.
Models that pair with this one.
The global fleet this model completes, and the chips and economics that sit inside the buildings it tracks.
Datacenter Industry Model
Global capacityThe same bottom-up standard applied to more than five thousand facilities outside China, with power supply and demand reconciled on top.
View model
Accelerator & HBM Model
Demand forecastThe chips that gate the buildout. China builds fast on a project basis, and growth is still constrained by chip supply.
View model
AI Cloud TCO Model
Cost modelThe rental economics and cost of ownership of the clusters deployed into this capacity.
View modelGet the China Datacenter Model.
Start with the sales team. They come back with scoping, licensing, and pricing for your mandate.
- Scoped to your use case
- A major update every quarter
- Custom research engagements available
