AI development is often imagined as a race for scale in which competitive edge is gained by harnessing enough energy to operate AI technology. Tech billionaires such as Elon Musk have stoked this narrative, suggesting that China’s energy abundance alone could hand it victory in the AI race. Such claims need to be scrutinised.

In fact, China’s data-centre expansion under the Eastern Data, Western Computing (EDWC) initiative reveals that abundant, cheap power alone is no reliable sign of a unified, operable computing network. China’s data-centre construction effort has fuelled inter-governmental competition, speculative overbuilding and fragmented infrastructure that has left many data centres idle.

Musk’s prediction that China’s electricity output would triple the United States’ by 2026 overlooked the fact that electricity consumption does not equate to generating capacity across regions nor to AI computing capacity. Data centres now account for only 1.68 percent of China’s total electricity consumption, increasing to 3 percent by 2030, which is still less than the US in both absolute amount and portion.

Energy costs and availability are seen as potential advantages for China, given its new data centres and state-backed power infrastructure. The reality is more complicated. Energy generation alone does not automatically translate into usable computing capacity, efficient utilisation or operational AI superiority.

China’s 2021-2023 Three-year Data Center Plan anticipated expanding capacity and optimising allocation of computing resources, setting goals of a 20 percent annual growth rate for server rack installation, 60 percent increase in utilisation rates and more than 200 eflops of computing power, equivalent to the capacity of the world’s fastest supercomputer, El Capitan. The EDWC initiative sought to create a unified, interoperable national computing network that used China’s abundant energy resources while coordinating deployments to support carbon peaking and carbon initiatives. By 2024, 633 hyperscale and large data centres had been constructed and made operational under the initiative, boosting China’s computing power to 268 eflops.

But soon after this rapid expansion, both domestic and international outlets began reporting overcapacity issues, followed by reports that China planned to sell its excess computing power. This mismatch exposed deep flaws in China’s attempt to merge AI ambitions with energy planning, illustrating how its rapid electricity-chasing expansion did not lead to tangible success.

Cool climate, cheap energy and local ambitions

Beijing’s data-centre policy focused on strengthening national computing capacity, harnessing excess renewable energy and increasing data-centre efficiency. The strategy was to relocate energy-intensive tasks to data centres in western regions while those in eastern regions would handle latency-sensitive tasks requiring instant response.

From an energy standpoint, the logic for western data centres made sense. Colder climates would reduce cooling costs, while the abundant renewable energy – particularly wind and solar – offered cheaper costs. Distributed solar projects in the west could settle at prices as low as 0.19 yuan per kilowatt-hour, compared with up to 0.43 yuan per kilowatt-hour in some eastern regions. Cheaper energy made data construction in provinces such as Guizhou, Gansu and Ningxia particularly attractive.

Local administrators in the economically underdeveloped hinterlands of the west saw the EDWC project as an opportunity to draw investment and enhance their political standing. Local governments began boosting the construction of data centres, sparking a development frenzy particularly in inland provinces. Even provinces that were not selected as clusters for the initiative but had cheap electricity, such as Shaanxi, rapidly expanded their data centre bases in a bid to become future EDWC hubs. In the past few years, Shaanxi has constructed 22 large-scale data centres and three big data industrial parks, backed by millions of yuan in subsidies. Such moves exemplify the powerful allure of the EDWC for western regions, which contributed to the unintentional outcomes of this national strategy.

Provincial competition and overinvestment

The overzealous ambitions of local officials exacerbated issues that later emerged from the EDWC. With an eye on increasing regional GDP, local governments began providing subsidies, rent reductions, tax benefits and housing support to encourage data centre development. Provinces such as Shandong and Guizhou already had provincial level plans for data-centre expansion, resulting in redundant investments and oversupplied infrastructure.

Both eastern and western provinces launched initiatives to attract customers, introducing compute vouchers in 2023 to subsidise AI development. These continue to be implemented nationwide at disproportionate measures as the scale and value vary across provinces, with Beijing, Shanghai and Chengdu offering more favourable incentives compared with western provinces such as Kunming. These policies have made eastern provinces more attractive for data centre expansion and applications.

As local governments proliferated the AI sector and overcrowded it with incentives, President Xi Jinping began to speak out against growing overinvestment. At a high-level Chinese Communist Party meeting, he rebuked local officials, saying ‘when it comes to projects, there are a few things – artificial intelligence, computing power, and new energy vehicles. Do all provinces in the country have to develop industries in these directions?’ He warned local government leaders not to become those officials who made reckless decisions and hasty investments but ran from their positions when debts and failures emerged.

The fallout of such overinvestment is seen most plainly in Guizhou, a province often heralded as a poster child for EDWC. In reality, Guizhou has not turned a profit on its data-centre expansion despite incentives and massive subsidies. The province now ranks a low 22nd in regional GDP due to high debts and faces corruption issues in its big data industry. This lack of economic success was the result of Guizhou failing to adequately address underlying logistical barriers.

Logistical downfalls of EDWC

Although the western provinces had abundant renewable energy and overwhelming support from local governments, they faced significant structural and technical barriers. Issues regarding latency and transmission concerns, intensive AI workloads, and proximity to talent and customers further limited the success of EDWC.

Many of the western data centres were built without adequate attention to the constraints of their locations. Remote regions lacked the fibre-optic cables necessary for moving massive volumes of data in real time, requiring operators to spend more on transmission capacity. The data centres also struggled to attract enough skilled workers and local customers.

AI workloads further complicated the situation. While AI training is generally not done in real time, AI inference (running trained models to generate outputs) requires low latency networks. The eastern regions of China have more high-priority data, known as ‘hot data’, and real-time users due to high population density and presence of economic hubs. In western regions, however, hot data cannot be processed without compromising response times and user experience. So eastern data ends up being processed in the east, while western data is processed in the west.

As a result, instead of the two regions working in tandem as the strategy intended, some western data centres sit empty, with utilisation rates as low as 20 to 30 percent, a far cry from the original policy goal of more than 60 percent. Meanwhile, data centres in Shanghai, Beijing and Shenzhen strain electricity grids that rely heavily on coal. Further exacerbating the energy issue, renewable-energy output reduction rates in the western region continue to reach more than 30 percent.

No unified, interoperable computing network

Despite abundant electricity supporting the infrastructure buildout, the project is constrained by internet transmission bottlenecks, AI workload demand and local governments’ inefficient use of existing capacity. The advantages of lower electricity prices are also being limited by the differing data transmission costs between eastern and western regions, along with business losses due to transmission delays.

In response to the failures of EDWC, several data-centre expansions have been halted. Beijing has moderated its ambitions, making large-scale efforts to consolidate scattered data-centre capacity. It restated that all data centres should have a utilisation rate of no less than 60 percent, and that no new large or super-large data centres should be built in cities where data centres already exist and operate at below 50 percent.

In November 2025, China began introducing electricity subsidies of up to 50 percent for data centres that used domestically produced semiconductors, following restrictions that prevented Chinese technology giants from purchasing Nvidia’s most advanced chips. While the subsidies are intended to alleviate the added energy cost of using less advanced chips, the discount also boosts utilisation. These subsidies are further evidence of weak demand that failed to meet the expectations of EDWC.

Conclusion

The EDWC initiative has fallen short of its goal to efficiently harness western China’s energy surplus. Instead of promoting coordinated optimisation, it has fuelled inter-governmental competition and speculative semiconductor stockpiling, diverting resources from priority sectors. Fragmentations in chip system design and standards across these over-built facilities, coupled with a shortage of skilled and reliable operators, have further undermined Beijing’s ambition to establish a unified, interoperable national computing network.

China’s persistent bottlenecks in network infrastructure and resource scheduling should not be overlooked when evaluating the benefits of its rapid data-centre expansion and low energy costs. Even if Beijing could solve issues around transmission, talent or scheduling over the next few years, there may still be a misalignment between provincial incentives, workload distribution and system coordination that continue to limit China’s operational AI advantage. Thus, many of China’s data centres are at risk of becoming stranded assets that are impressive on paper but devoid of actual productive force.