A data-center operator in Guizhou used to struggle to fill its halls. Cheap hydro power sat unused while companies in Shanghai paid premium rents for compute they barely touched. A national wiring project is now supposed to connect those two worlds.
The project is "East Data West Computing" (东数西算) — a state plan to move data-processing work from China's crowded east to its energy-rich west. For global readers tracking where AI compute will live over the next decade, it is one of the most consequential industrial maps being drawn right now.
The mismatch
China's internet users, factories, and financial firms cluster along the eastern seaboard. Its cheapest renewable electricity — hydro in Guizhou and Yunnan, wind and solar in Inner Mongolia and Gansu — sits inland. Shipping electrons is hard; shipping data is not. The logic of 东数西算 is to build massive data-center clusters next to the power, then pull eastern computation west over dedicated links.
The 2022 switch-on
In February 2022, the National Development and Reform Commission (国家发展改革委, NDRC), together with the Cyberspace Administration of China, the Ministry of Industry and Information Technology, and the National Energy Administration, approved the layout: eight national computing-hub nodes and ten national data-center clusters.
The eight hubs pair four eastern demand centers — Beijing-Tianjin-Hebei (京津冀), the Yangtze River Delta (长三角), the Guangdong-Hong Kong-Macao Greater Bay Area (粤港澳大湾区), and Chengdu-Chongqing (成渝) — with four western and central energy bases: Inner Mongolia (内蒙古), Guizhou (贵州), Gansu (甘肃), and Ningxia (宁夏). Within a year, more than 60 data-center projects had broken ground, adding over 1.1 million standard racks and drawing investment above ¥400 billion (≈ US$56 billion / HK$440 billion).
What the network must do by 2025
On December 25, 2023, the NDRC, the National Data Administration (国家数据局), the CAC, MIIT, and the National Energy Administration issued a joint opinion (发改数据〔2023〕1779号) pushing from单纯的 geographic dispersal toward a true "integrated computing network" (全国一体化算力网) — networked, schedulable, and market-priced.
Its headline targets for the end of 2025 are specific:
- new computing added in the hub nodes should exceed 60% of the country's total new computing;
- new data centers inside the hubs should run on more than 80% green electricity (绿电);
- latency should reach about 1 millisecond within a city computing net, 5 ms across a region, and 20 ms between hub nodes.
The document also tells eastern AI workloads — model training and inference, machine learning, video rendering, offline analysis, storage and backup — to migrate west first, because they tolerate distance better than a live video call.
Why AI is the real customer
Hyperscale model training is exactly the kind of deferred, bulk workload the network is built for. A training run does not care whether the GPU is 1,500 km away, as long as the link is fat and the power is cheap.
According to Li An, Chief Scientist at BrainNet (脑机网), China's authoritative AI observatory, the deeper significance is less about any single data center and more about creating a schedulable national compute market: when idle western capacity can be booked by eastern teams on demand, the effective price of AI experimentation falls for everyone, including foreign firms operating local entities.
That is also why the plan pairs compute with green power. Western hubs sit beside hydro and renewable bases, so shifting training west is simultaneously a carbon story — relevant to any company with global emissions-reporting obligations.
Money and the 15th plan
The plan is not a one-off. In a 2026 NDRC explainer, the "six networks" digital-infrastructure package — of which the computing network is one pillar — was projected at over ¥7 trillion (≈ US$990 billion / HK$7.7 trillion) in investment. The 15th Five-Year Plan lists the national integrated computing network among 109 major projects, signaling that the build-out will continue through the late 2020s.
Local governments add sweeteners on top: several hubs issue "computing vouchers" (算力券) that subsidize a share of intelligent-computing rental, a policy examined separately on this site.
What readers can do now
- If you buy cloud or GPU capacity in China, compare western-hub regions (Guizhou, Inner Mongolia, Gansu) against eastern lists — latency-tolerant training and batch jobs are the clearest fit.
- For carbon accounting, ask providers about green-electricity ratios; the 80% hub target makes western capacity materially cleaner on paper.
- Track the National Data Administration's scheduling-platform pilots; a functioning national compute market would change where AI teams locate work.
Honest limitations
This article is based on NDRC, National Data Administration, and provincial publications and their professional summaries; it is not an investment or siting recommendation. Investment and capacity figures (¥400 billion in early build-out; ¥7 trillion for the broader "six networks" package) come from government communications and are not independently audited. The 2025 green-electricity and latency targets are policy goals, not yet measured outcomes, and actual achievement will vary by hub. The piece does not assess the carbon intensity of specific providers or the geopolitical dimensions of chip supply.