// DATA CENTER POWER & COOLING VALUE CHAIN

Data Center Power & Cooling Value Chain map

This chain covers only the power and cooling of data center physical infrastructure: from the property boundary with the public grid or a direct green-power line to the power and heat interfaces of IT equipment, plus the operator's efficiency and waste-heat work.

Content comes only from the fixed evidence package; each statement is marked as a sourced fact, a structural inference or a gap, and facts carry a source ID and location.

Nodes list reviewed source records only: 4 companies in total (a company in several nodes counts once). Companies that merely match by industry or name are not listed.

Company IDs, names and codes come from frozen source records; they do not mean the current legal entity has been verified.

Targets for end-2025 set by the 2024 plan (by scope)

  • FactNational average data center PUE below 1.5, with an overall rack utilisation of at least 60%.[S01 · p1,一、主要目标]
  • FactPUE within 1.25 for new and expanded large and hyperscale data centers.[S01 · p2,二(二)]
  • FactPUE no higher than 1.2 for data center projects in national hub nodes.[S01 · p2,二(二)]
  • FactGreen power above 80% for new data centers in national hub nodes.[S01 · p3,二(四)]
  • GapThese are plan targets and scope-specific requirements, not a uniform limit for every project. Whether they were met, and which constraints apply after 2025, has not been verified here.

Nodes at a glance

Nodes are ordered by stage and do not imply an electrical series connection; interfaces not confirmed by a source are marked as gaps.

N1

Grid connection and green power

Boundary

  • InferenceFrom the public grid or a direct green-power line to the campus high/medium-voltage incoming feeders.

Interfaces and neighbours

  • FactUpstream can be the grid company or a renewable project supplying a single user over a direct line; for grid-connected direct-supply projects, safety responsibility is split at the property boundary.[S03 · 条目一、条目十]
  • InferenceDownstream connects to N2, medium-voltage distribution and transformation.

Who pays

  • FactDirect green-power projects must still pay transmission and distribution charges, system operation charges, policy cross-subsidies and government funds and surcharges.[S03 · 条目十三]
  • GapS03 does not mention data centers at all; it is cited only for charges and responsibility boundaries at grid connection, not as a data-center rule.

Revenue and cost

  • GapThe share of electricity in data center operating cost, grid-connection approval time, and dual-feed or high-reliability supply charges: no primary source obtained.

Risks and constraints

  • FactThe 2024 plan targets green power above 80% by end-2025 for new data centers in national hub nodes (applies only to new projects in hub nodes).[S01 · p3,二(四)]
  • FactFor existing data centers below the energy-saving level and those under construction or planned that miss the advanced level, the 2024 plan uses the wording "study" tiered electricity prices based on the efficiency gap.[S01 · p4,三(一)]
  • GapThat wording does not establish that tiered pricing has been implemented; its current status has not been verified here.

Verified source-record associations

No verified source-record association yet. This does not mean the stage has no companies, only that none is verified on this site.

N2

Medium-voltage distribution and transformation

Boundary

  • InferenceIncoming feeder → medium-voltage switchgear → transformer → low-voltage distribution.

Interfaces and neighbours

  • FactVertiv's products include low/medium-voltage switchgear and busbar.[S06 · Item 1. Business]
  • InferenceDownstream connects to N3; an integrated 10kV-to-DC power module folds part of this node into one unit (see N3).

Who pays

  • InferenceMedium-voltage distribution and transformer equipment may be bought directly by the data center builder or through a system integrator or general contractor; the actual purchasing path for this node is unverified.

Revenue and cost

  • Fact金盘科技 (688676) discloses that its transformers are used in more than one hundred data center projects.[S11 · (7)新型基础设施建设应用领域]
  • GapDomestic switchgear makers' data center business, lead times and material costs: no primary source obtained.

Risks and constraints

  • FactThe plan encourages upgrading power distribution systems such as high-efficiency transformers (an encouraged direction, not a mandate).[S01 · p4,二(六)]

Verified source-record associations

N3

Uninterruptible power (UPS / HVDC / power modules)

Boundary

  • InferenceThe boundary depends on the architecture: low-voltage AC UPS and 240V/336V HVDC sit between low-voltage distribution and IT distribution; an integrated 10kV-to-DC power module spans N2, N3 and N5. The three architectures are listed separately below.

Interfaces and neighbours

  • GapOnly the HVDC design described in S09 has a primary description of the battery interface; battery interfaces of the other architectures and the generator transfer point are unverified, so no series topology is drawn here.

Who pays

  • InferenceThe data center builder or IDC operator.

Revenue and cost

  • Fact科华数据 (002335) "data center products" line: 2024 revenue RMB 1.921 billion, gross margin 32.25% (one company's product line, not industry data for this node).[S08 · p20–21,营业收入构成]

Risks and constraints

  • FactThe plan encourages DC power, power modules and modular power supplies (an encouraged direction; it does not require any project to adopt a given architecture).[S01 · p4,二(六)]
  • GapMarket share of each architecture and primary evidence for other domestic makers such as 科士达 and 易事特: not obtained.

Three uninterruptible power architectures (kept apart)

Low-voltage AC UPS
  • Fact科华数据 (002335) data center UPS range covers 300kVA–1200kVA, in tower and modular types.[S08 · p13]
  • GapThe source does not describe how batteries or generators are connected.
240V/336V high-voltage DC (HVDC)
  • FactConverts 380V AC mains to 240V (or 336V) DC; batteries hang directly on the DC output busbar (the design described by 中恒电气).[S09 · p2]
  • GapThis battery arrangement holds only for the S09 design and is not generalised to other HVDC products.
Integrated 10kV-to-DC power module
  • FactIntegrates 10kV distribution, transformer, uninterruptible power and output distribution units; one set supports 2.4MW of IT load (中恒电气 Panama power module).[S09 · p3]
  • GapSpans N2, N3 and N5; the source does not say where batteries connect.

Verified source-record associations

N4

Backup power

Boundary

  • InferenceTwo branches: generator sets for long-duration backup (fuel type unverified), and batteries or lithium backup systems for short ride-through.

Interfaces and neighbours

  • GapWhere generators connect and transfer in the power path is not documented here by any standard or vendor wiring document, so no series topology is drawn; for batteries see the N3 architectures.

Who pays

  • InferenceThe data center builder.

Revenue and cost

  • Fact南都电源 (300068) reports winning second place in Alibaba's annual tender and the sole award of a GDS high-voltage lithium project during the period.[S10 · p5]
  • FactCummins states that customer applications of its standby and prime generator sets include data centers (this only shows a generator maker lists data centers as an application).[S12 · Item 1, Power Systems segment]

Risks and constraints

  • FactIn Uptime's explanation of tiers, Tier I includes an engine generator, and Tier II redundant components include UPS modules, chillers or pumps, and engine generators (an explanatory article, not the standard itself).[S04 · Tier I / Tier II]
  • GapPrimary sources for domestic generator makers, the fuel mix of data center generators, and GB 50174 backup-power requirements: all unverified.

Verified source-record associations

N5

Rack power distribution

Boundary

  • InferenceIT distribution → row distribution, busway → rack PDU → server power supplies.

Interfaces and neighbours

Who pays

  • InferenceThe data center builder; with fully integrated racks it may come with the server vendor.

Revenue and cost

  • Fact科华数据 (002335) cites a "largest busway application project".[S08 · p14]
  • GapPDU makers and integrated-rack power specifications (such as OCP Open Rack v3): no primary source obtained.

Risks and constraints

  • InferenceHigh-density and fully integrated racks may change the distribution form and who buys it; not yet sourced.

Verified source-record associations

N6

Cooling (air / evaporative / liquid)

Boundary

  • InferenceCooling source → room-level or in-row air units; or cooling source → CDU → manifold → quick connector → cold plate.

Interfaces and neighbours

  • Fact英维克 (002837) liquid cooling products cover cold plates, quick connectors, manifolds, CDUs, racks, coolant, piping and cooling sources.[S07 · p2]
  • FactNVIDIA's GB200 NVL72 is a rack-scale, liquid-cooled design.[S05 · 产品页 / product page]
  • InferenceThe choice of server platform may shape the facility-side cooling interface.

Who pays

  • Fact英维克 (002837) discloses that its products are supplied directly or through system integrators to data center owners, IDC operators, large internet companies and telecom operators (one company's sample, not generalised to all cooling products).[S07 · p2]

Revenue and cost

  • Fact英维克 (002837) cumulative liquid-cooling deliveries reached 1.2GW by March 2025; 2024 revenue RMB 4.589 billion (including non-data-center business).[S07 · p2][S07 · p3,主要会计数据]
  • GapSeparate liquid-cooling revenue, immersion cooling, and other liquid-cooling makers: no primary source obtained.

Risks and constraints

  • FactThe plan encourages efficient cooling such as liquid cooling, evaporative cooling, heat pipes and pumped refrigerant, adapted to local conditions (an encouraged direction, not a mandate).[S01 · p4,二(六)]
  • InferenceCooling and power-distribution losses together affect PUE; S01 sets PUE targets only and does not describe the breakdown.

Verified source-record associations

N7

Owners and operators

Boundary

  • InferenceData center owners, IDC operators, cloud providers, telecom operators, and enterprises running their own data centers.

Interfaces and neighbours

  • InferenceBuys N2–N6; either rents rack space or compute services to outside tenants, or uses the capacity for its own business.

Who pays

  • InferenceThe final buyer of the equipment in this chain. Outside-tenant payment and own use are two different value relationships; see the value flow below.

Revenue and cost

  • Fact科华数据 (002335) "IDC service revenue": RMB 1.231 billion in 2024 at 23.32% gross margin, based on its 10 self-built data centers (one company's data, not extended to cloud providers, operators or in-house data centers).[S08 · p20–21,营业收入构成][S08 · p15]
  • GapIDC operators' cost structure (electricity, depreciation) and how in-house data centers allocate cost: no primary source obtained.

Risks and constraints

  • FactThe plan states that where data centers in operation for over a year in a city have overall utilisation below 50%, new clusters and large or hyperscale projects should in principle not be planned there.[S01 · p2,二(一)]

Verified source-record associations

N8

Efficiency and waste heat

Boundary

  • InferencePUE management, demand response and waste-heat recovery, spanning N1–N7.

Interfaces and neighbours

  • FactThe plan encourages capable data centers to take part in demand-side management, and encourages heat-recovery systems for campus heating, district heating and facility agriculture.[S01 · p3,二(五)]

Who pays

  • InferenceThe operator.

Revenue and cost

  • GapNo primary case for how waste-heat recovery earns revenue.

Risks and constraints

  • FactFor PUE targets see "Targets for end-2025 set by the 2024 plan" at the top; they apply separately to the national average, new and expanded large data centers, and national hub nodes.[S01 · p1–2]
  • GapResearch note: company samples for this chain are retrieved by the power grid equipment, energy storage and energy services industries; an industry match does not prove supply to data centers and must be checked against primary sources each round. Verified associations on this site are only source records 34, 31, 35 and 93.

Verified source-record associations

No verified source-record association yet. This does not mean the stage has no companies, only that none is verified on this site.

Value flow: who pays whom

  • InferenceBuying power: data center owner or operator → power supplier, paying for electricity; the contracts and who collects each charge are unverified.
  • FactDirect green-power projects must pay transmission and distribution charges, system operation charges, policy cross-subsidies and government funds and surcharges under national rules (a general direct-supply rule, not specific to data centers).[S03 · 条目十三]
  • InferencePurchasing: data center owner or operator → equipment suppliers, possibly directly or through system integrators or general contractors; the actual path for each equipment type is unverified.
  • FactVerified sample: 英维克 (002837) discloses that its thermal-management products reach customers directly or through system integrators, and that it acted as MEP general contractor delivering a turnkey service for the 广州新一代液冷算力中心 project (one company and one project, not generalised to the whole chain).[S07 · p2]
  • FactVerified IDC service revenue sample: 科华数据 (002335) IDC service revenue of RMB 1.231 billion in 2024 (one company's data; the actual charging model is unverified).[S08 · p20–21]
  • InferenceRenting out: outside tenants → IDC operator, paying for rack space or compute services; the actual charging model is unverified.
  • InferenceOwn use: the enterprise's own business uses the data center with no outside charge; value shows up as internal cost allocation or support for its own revenue.

Adjacent maps

Sources

read-verbatim: quoted word for word from the original PDF, hash recorded; tool-extracted: web page extracted by a tool, check the original before quoting. Page numbers are physical PDF pages.

  1. S01 Special Action Plan for Green and Low-Carbon Development of Data Centers (NDRC Huanzi [2024] No. 970), Chinese original · NDRC and other departments · 2024-07 · read-verbatim · sha256 30d310ed765d…
  2. S02 NDRC expert commentary on the data center action plan, Chinese original · NDRC · 2024-07-30 · tool-extracted
  3. S03 Notice on orderly development of direct green-power supply (NDRC Energy [2025] No. 650), Chinese original · NDRC and National Energy Administration · 2025-05 · tool-extracted
  4. S04 Explaining the Uptime Institute's Tier Classification System (April 2021 Update) · Uptime Institute Journal · 2021-04 update · tool-extracted
  5. S05 NVIDIA GB200 NVL72 product page · NVIDIA · accessed 2026-10-15 · tool-extracted
  6. S06 Vertiv Holdings Co, Form 10-K FY2024 · Vertiv / SEC EDGAR · FY2024 · tool-extracted
  7. S07 英维克 (002837) 2024 annual report summary, Chinese original · 深圳市英维克科技股份有限公司 (issuer filing) · FY2024 · read-verbatim · sha256 8b52034d5dae…
  8. S08 科华数据 (002335) 2024 annual report, full text, Chinese original · 科华数据股份有限公司 (issuer filing) · FY2024 · read-verbatim · sha256 557f9c970e9d…
  9. S09 中恒电气 (002364) 2024 annual report summary, Chinese original · 杭州中恒电气股份有限公司 (issuer filing) · FY2024 · read-verbatim · sha256 4c93ada0359d…
  10. S10 南都电源 (300068) 2024 annual report summary, Chinese original · 浙江南都电源动力股份有限公司 (issuer filing) · FY2024 · read-verbatim · sha256 40348fca57fe…
  11. S11 金盘科技 (688676) 2024 annual report summary as printed in 中国证券报, 2025-04-03, p. B066, Chinese original · 海南金盘智能科技股份有限公司 (statutory disclosure printed in a newspaper) · FY2024 · read-verbatim · sha256 b578fdca4132…
  12. S12 Cummins Inc., Form 10-K FY2024 · Cummins / SEC EDGAR · FY2024 · tool-extracted

Related research

Lists only public reports whose tags contain both 产业链专题 (chain topic) and the full chain name.

No related research yet.