China’s data-center shift could cut emissions 27.8% by 2060

A coordinated redistribution of China’s rapidly expanding data-center capacity could reduce the sector’s operational carbon dioxide emissions by 27.8% in 2060 compared with continued expansion along the existing geographic pattern, according to a province-level modeling study of the country’s Eastern Data and Western Computing initiative. The projection is conditional rather than a forecast: its outcome depends on western power grids decarbonizing faster than eastern grids, on receiving regions being able to absorb infrastructure quickly, and on policies that reduce the concentration of computing activity in established eastern hubs. Under alternative assumptions in which regional grids decarbonize at the same pace, or western grids improve 50% more slowly than the reference trajectory, the modeled benefit disappears and emissions become slightly higher than under inertial expansion.

The researchers reconstructed operational data-center rack stocks and electricity-related emissions for all 31 provincial-level regions of mainland China from 2016 through 2023. National rack capacity increased from 1.24 million to 8.10 million over those years, while associated operational emissions rose from 31.9 million to 139.3 million metric tons of carbon dioxide. Average emissions per rack-year fell from 25.7 to 17.2 tons, indicating that efficiency gains and gradual grid decarbonization partly offset rapid capacity growth without reversing the rise in absolute emissions. Because provincial rack counts for 2020–2023 were estimated by distributing national totals according to each province’s earlier average share, the historical reconstruction is not an assessment of the realized effects of the policy launched after 2022.

Spatial concentration remained pronounced. Six leading eastern provinces—Guangdong, Jiangsu, Hebei, Shanghai, Zhejiang and Beijing—held 51.6% of national rack stock in the reconstructed 2023 distribution and generated 51.2% of the associated emissions. The eastern region as a whole accounted for 61.2% of both capacity and emissions, while the west held 24.0% of racks and produced 23.6% of emissions. The four designated western computing hubs of Inner Mongolia, Guizhou, Gansu and Ningxia together contained only 12.1% of national rack stock. The historical pattern therefore showed that China’s computing infrastructure remained far more concentrated near users, digital industries and network infrastructure than the country’s distribution of land, renewable resources and natural cooling advantages might suggest.

To distinguish the conditions for operating data centers from the advantages of existing digital hubs, the study constructed two composite indices. The Physical Hosting Index incorporated 19 indicators covering climatic cooling, air quality, geological safety, water supply, energy efficiency, clean-energy capacity and carbon control. A separate Digital Readiness/Lock-in Index used 14 indicators reflecting network infrastructure, digital industrialization and the penetration of digital technologies. The two measures were only weakly correlated in 2023, with a Pearson correlation of 0.30, showing that a province with favorable physical conditions for hosting servers need not have the networks, industries and technical ecosystems needed to sustain large-scale computing activity. Some eastern provinces scored highly on both dimensions, while provinces such as Inner Mongolia, Gansu, Tibet and Qinghai combined strong physical conditions with relatively low digital readiness.

The indices also exposed a tension within the designated hub network. Guizhou and Ningxia both fell below the study’s physical-hosting benchmark, illustrating that official hub status does not guarantee that every province has equally favorable measured conditions for expansion. The study therefore tested four nested governance pathways through 2060 while holding national computing-demand growth, operating parameters and reference grid-decarbonization trajectories constant. Inertial expansion continued the historical spatial pattern. Policy-directive alignment prioritized designated hubs, resource screening additionally directed growth toward provinces with stronger physical hosting conditions, and coordinated implementation combined policy designation, physical suitability, relief from digital lock-in, a bounded 2025–2030 migration window and dynamic updating of the inherited allocation pattern.

The simpler pathways had limited effects. By 2060, policy-directive alignment reduced modeled emissions by 3.7% relative to inertial expansion, while resource screening reduced them by only 0.7%. Resource screening briefly increased emissions by 0.3% in 2030 because it also favored high-PHI eastern provinces such as Beijing and Zhejiang rather than producing a substantial westward shift. Coordinied implementation was markedly different: modeled emissions peaked at 239.9 million tons in 2040 and declined to 133.1 million tons in 2060, compared with 184.3 million tons under inertial expansion. An ablation analysis attributed the 27.8% reference-case reduction to the combined package rather than any single rule: adding digital-lock-in adjustment increased the reduction to 8.54%, the migration window raised it to 14.89%, and dynamic reference-share updating brought it to 27.78%.

Under coordinated implementation, the west’s share of national rack stock rose to 54.5% in 2030 and 59.6% in 2060. Yet the region generated only 38.1% of national emissions in 2060, while the east held 25.3% of capacity and produced 43.1%. This inversion did not emerge immediately: in 2030, the west still accounted for 51.8% of emissions. It developed as capacity moved toward western provinces while their power systems followed the assumed long-term decarbonization paths. The modeled benefit therefore did not come simply from a large existing carbon advantage in the west. Estimated 2023 operational emissions per rack-year were similar across the east, central region and west, at 17.2, 17.7 and 16.9 tons, respectively, making the relative evolution of regional grids a decisive assumption.

Sensitivity analysis demonstrated that the direction of the effect could change. With a uniform annual decline in grid-emission intensity, coordinated reallocation produced 0.37% more emissions in 2060 than inertial expansion. If western decarbonization proceeded 50% more slowly than in the reference specification, the coordinated scenario produced 2.41% more emissions. By contrast, accelerating eastern decarbonization by 25% preserved a 20.73% reduction. These results are not causal estimates of the policy’s observed effect, nor are they predictions of actual emissions through 2060. They compare hypothetical allocation pathways under shared assumptions about demand, operations and regional power systems, and they cover only electricity-related operational emissions rather than embodied emissions from manufacturing, construction or decommissioning.

Implementation limits further narrowed the range of achievable benefits. A 27-case sensitivity grid varied migration friction in sending regions, minimum physical-hosting thresholds for receivers and annual growth caps on new capacity. The annual growth cap had the largest marginal influence, producing a 10.9-percentage-point swing in the 2060 outcome, compared with 8.1 points for the hosting threshold and 1.7 points for migration friction. Reductions ranged from 4.9% to 28.7%, with a median of 20.1%. Tighter intake limits were especially important because many prospective western receivers began with small rack bases and needed time to secure grid connections, cooling systems, land approvals and other infrastructure. The highest result excluded Guizhou and Ningxia from the eligible receiver pool, making it a different policy trade-off rather than a stronger realization of the original designated-hub scenario. Overall, the study identifies receiving-region absorptive capacity—not eastern reluctance to migrate—as the dominant implementation bottleneck in the modeled pathways.

Subject of Research: Scenario modeling of how spatial governance could reduce operational emissions from China’s expanding data-center sector

Article Title: Spatial governance of China’s data-center expansion as a climate-action lever under resource and digital lock-in

Article References: Chen, S., Wang, Z., Yang, K., Li, C., Peng, S., Zhu, Y., Wei, H., Wang, Q., & Pan, X. (2026). Spatial governance of China’s data-center expansion as a climate-action lever under resource and digital lock-in. Frontiers in Environmental Science, 14, 1990264. https://doi.org/10.3389/fenvs.2026.1990264

Image Credits: AI Generated

DOI: 10.3389/fenvs.2026.1990264

Keywords: data centers, China, carbon emissions, spatial governance, power grids, artificial intelligence, digital infrastructure, climate mitigation, Eastern Data and Western Computing, Spatial, governance, data-center