One Year, Two Official Numbers: 170 vs 196 Billion kWh of Compute Electricity in China
In one line: on the question "how much electricity did China's compute facilities use in 2025", two official sources give different numbers — the National Energy Administration: 170 bn kWh, 1.6% of national electricity; CAICT's basis: about 196 bn kWh, 1.9%, up 18.1% year on year. The gap is roughly 26 bn kWh (~15%). This piece places the two bases side by side with attribution and does not infer the cause. Meanwhile, policy moved two levels in 2026 — from a sector action plan to a 15th Five-Year Plan chapter plus an annual designation list.
1. The two bases, side by side
| Basis | Value | Share | Year-on-year | Source |
|---|---|---|---|---|
| National Energy Administration | 170 bn kWh | 1.6% | — | NEA |
| CAICT | ~196 bn kWh | 1.9% | +18.1% | China Academy of Information and Communications Technology |
The gap is about 26 bn kWh, roughly 15% in relative terms.
⚠️ This piece does not conclude on the cause of the gap. Statistical scope (whether self-built facilities are included, whether non-hub sites are counted, how racks are defined), the reporting body, and the collection method could all contribute — but no official explanation of the difference was found.
Handling rule: record both, attribute each, do not average and do not pick one. Any use of a "China compute electricity" figure should state both basis and source — giving a single number without declaring its basis is manufacturing precision.
2. Two readings of the same data
Reading A (stock share): 1.6% – 1.9%. → Conclusion: compute electricity is still a small share of national electricity.
Reading B (growth rate): +18.1%, more than 5× the growth of national electricity consumption. → Conclusion: small share, but growth several times the aggregate.
Both readings are correct and point in opposite directions. Reading A alone yields "not a problem"; Reading B alone yields "urgent". They describe different positions on the same curve.
Forward-looking basis (NEA): China's compute electricity is expected to reach 800 bn kWh by 2030, about 6% of national electricity.
3. The 2026 policy chain: two levels up in one year
| Date | Document | Key content |
|---|---|---|
| 2024-07 | Four-department Data Centre Green and Low-Carbon Development Action Plan | Six workstreams: layout optimisation · stricter entry for new projects · retrofitting existing stock · higher renewable share · resource efficiency · energy-saving technology deployment |
| 2026-04 | CPC Central Committee and State Council Opinion on Higher-Level, Higher-Quality Energy Conservation and Carbon Reduction | First time digital infrastructure is listed as one of five key sectors alongside industry, buildings, transport and public institutions; strict management of PUE and other efficiency indicators required |
| 2026-05 | NDRC/NEA Document No. 688 (2026) on orderly development of multi-user direct green power connection | Priority support for compute facilities, green hydrogen/ammonia/methanol and other emerging industries to connect directly to green power |
| 2026-07 | State Council 15th Five-Year Plan Carbon Peaking Action Plan (Guo Fa [2026] No. 22) | A dedicated section on green and low-carbon transformation of compute facilities: strengthen co-location with renewables, support direct green power supply, require new facilities to be mainly powered by non-fossil electricity, encourage zero-carbon compute facilities |
| 2026 (NEA) | Energy Sector Energy Conservation and Carbon Reduction Action Plan (2026–2028) | Promote co-planning of renewable bases and compute facilities; support higher green power consumption; tap the temporal-spatial flexibility of compute loads |
| 2026-09 | Five-department Notice on the 2026 National Green Compute Facility Designation | Applications close 2026-11-16; evaluation framework restructured (below) |
What "listed as one of five key sectors" means: digital infrastructure moves from "an industry that should save energy" to a constrained object at the same level as industry, buildings and transport. The verb shifts from "promote" to "manage".
4. The 2026 designation: the indicators changed
The new evaluation framework upgrades the 14th Five-Year Plan design and is built on four dimensions:
Energy use · Carbon emissions management · Water use · Compute resource utilisation
New indicators: energy efficiency per unit of compute · carbon utilisation efficiency · digitalised carbon management · product carbon footprint
Indicators with raised weighting: carbon efficiency · water use · green power share · rack occupancy rate
| Application threshold (partial) | Requirement |
|---|---|
| Continuous stable operation | 1 year |
| Rack scale (excluding public institutions, financial compute and intelligent-compute facilities) | at least 3,000 standard racks |
| Rack occupancy rate | no less than 60% |
| Location | New large and very large facilities should in principle be located in national integrated compute network hubs and compute-power coordination pilot regions |
| Efficiency | National standards level 2 or above for data centres and servers |
| Incident record | No major safety, environmental or compliance incidents in the past three years |
"Entry and exit": the list is dynamically managed; facilities failing on PUE, carbon efficiency, green power share or other key indicators will be removed promptly.
This matters more than the thresholds themselves. A one-off selection is an honour; dynamic management is a standing constraint. It separates "winning the plaque" from "keeping it", and the latter requires routine measurement and data capability — precisely the weakest link for most facilities today.
5. Baseline as of end-2025
| Indicator | Value |
|---|---|
| Standard racks in use | over 13.73 million |
| Intelligent compute at the eight hubs (ten clusters) | 1.388m PFlops, over 80% of the national total |
| Average PUE of very large compute facilities | 1.34 |
| Facilities rated 4A or above for green low-carbon performance | over 160 |
Price signal on the green side (green certificates):
| Basis | Value |
|---|---|
| August 2026 average, by generation year | 2024 vintage CNY 0.61 · 2025 vintage CNY 3.21 · 2026 vintage CNY 6.32 |
| Q1 2026 average traded price | CNY 5.13/certificate (+217% y/y) |
| China's 2025 green power traded volume | 328.5 bn kWh (+38.3%) |
Pricing certificates by generation year is the most notable feature of this market: it means the vintage attribute is now itself priced, with current-year certificates commanding the highest premium.
6. Boundaries and definitions
- Two official bases coexist; this piece does not pick a side and does not convert. Attribution is mandatory when citing. See section 1.
- The 800 bn kWh figure for 2030 is a projection (scenario basis), not an accomplished fact. NEA publishes other longer-horizon scenarios; projections must not be placed on the same level as measured snapshots.
- Green certificate prices are monthly/quarterly snapshots that move with vintage quotas and supply-demand; they are not long-run levels.
- No facility-level location information and no user-side electricity data appear here; all figures are national and regional aggregates.
- No commercial data vendors are used.
Data sources
Sourcing floor: all data here comes from public government bodies. No commercial data vendors.
| Data | Source | Grade | Confidence | Record ID |
|---|---|---|---|---|
| 2025 national compute centre electricity 170 bn kWh, 1.6% of national total | National Energy Administration | L1 | A | tbm-155 |
| 2025 compute centre electricity about 196 bn kWh (+18.1%) | China Academy of Information and Communications Technology | L2 | A | tbm-156 |
| Projected 2030 national compute electricity 800 bn kWh, about 6% | National Energy Administration | L1 | B | — |
| Q1 2026 internet data services electricity 22.9 bn kWh (+44.0%) | National Energy Administration | L1 | A | tbm-157 |
| Data Centre Green and Low-Carbon Development Action Plan | NDRC and three other departments | L1 | A | tbm-146 |
| CPC Central Committee / State Council Opinion — digital infrastructure as one of five key sectors | CPC Central Committee and State Council (via National Energy Conservation Centre meeting material) | L1 | B | — |
| NDRC/NEA Document No. 688 (2026): priority for compute facility direct green power connection | NDRC, NEA | L1 | A | tbm-147 |
| State Council 15th Five-Year Plan Carbon Peaking Action Plan (Guo Fa [2026] No. 22) | State Council | L1 | A | — |
| Energy Sector Energy Conservation and Carbon Reduction Action Plan (2026–2028) | National Energy Administration | L1 | A | — |
| Five-department 2026 National Green Compute Facility Designation: four dimensions, new indicators, deadline 2026-11-16, dynamic management | MIIT, NDRC, NFRA, National Government Offices Administration, NEA | L1 | A | — |
| Over 13.73 million standard racks in use (end-2025) | National Data Administration | L1 | A | tbm-148 |
| Eight hubs hold 1.388m PFlops of intelligent compute, over 80% of national total | National Data Administration | L1 | A | tbm-150 |
| Average PUE of very large compute facilities falls to 1.34 (end-2025) | National Data Administration | L1 | A | tbm-152 |
| August 2026 green certificate prices 0.61 / 3.21 / 6.32 yuan by vintage | National Energy Administration | L1 | A | tbm-162 |
| Q1 2026 average green certificate price 5.13 yuan (+217%) | National Energy Administration | L1 | A | tbm-160 |
| China's 2025 green power trading 328.5 bn kWh (+38.3%) | National Energy Administration | L1 | A | tbm-163 |
Grade key: L1 = government / international organisation / intergovernmental body / customs; L2 = international standards bodies / regulated exchanges / state think tanks; L3 = industry associations / company self-disclosure / official media relay (grey — original source must be cited); L4 = commercial data vendors / self-media (prohibited in this project).
Entries marked "—" are published here for the first time and are not yet in the on-site record index.
Data as of 2026-10-11. Individual records can be retrieved by record ID via the on-site data index or the MCP endpoint.