Professor Shipeng Yan and 王筱晴
1 July 2026
In recent years, a wave of artificial intelligence (AI) has swept the globe, and the popularization of generative AI has led to a sharp increase in demand for compute resources. Behind the impressive models and products, the infrastructure underpinning this technological revolution is an array of data centres located across the world. These seemingly unremarkable facilities have in fact become not only a key pillar of the digital economy, but also a strategic asset in great-power rivalry.
At the same time, data centres’ enormous consumption of energy and resources has placed far-reaching pressure on the environment and society. As countries around the world race to invest heavily in AI infrastructure, its negative effects and developmental bottlenecks are also gradually emerging, warranting detailed analysis.
The lifeblood of the digital era
The core function of data centres is to provide society as a whole with data storage and compute services. Cloud-based work, online shopping, video streaming, or today’s much-discussed generative AI all depend on the compute resources provided by data centres. Clearly, without data centres, the digital economy could scarcely function.
It is precisely for this reason that global capital is flowing into this sector on an unprecedented scale. Investments by technology giants are now often measured in hundreds of billions of US dollars. In January 2025, OpenAI launched its Stargate project, with plans to invest US$500 billion over four years in the large-scale construction of AI infrastructure, in partnership with companies including Oracle, Microsoft, Nvidia, and SoftBank (see Note 1). Meanwhile, China also plans to invest approximately RMB2 trillion over the next five years to build a nationwide, interconnected network of data centres (see Note 2). Whether undertaken by private enterprises or the public sector, the scale of this “compute arms race” is formidable.
For both China and the US, the importance of data centres and compute capacity has risen to the level of national strategy. Compute capacity is regarded as a strategic resource on a par with electricity and oil, with a direct bearing on competitive dominance in the AI era. In recent years, China has vigorously promoted its “Eastern Data and Western Computing” project, while the US has made concerted efforts to support domestic AI infrastructure. Data centres have effectively become the main battleground in a new round of technological competition.
Negative externalities on the rise
However, the rapid expansion of data centres is generating a series of negative externalities that cannot be ignored, foremost among them energy consumption. Data from the International Energy Agency show that global data-centre electricity consumption was approximately 460 terawatt-hours (TWh) in 2022 and may exceed 1,000 TWh by 2026, almost equivalent to Japan’s total annual electricity consumption (see Note 3). In Ireland, where data centres are densely concentrated, their electricity consumption even accounts for more than 30% of the national total. Such enormous electricity demand has, to a certain extent, crowded out residential power use and driven up electricity prices, leaving the public to bear the cost for the expansion of data centres.
Water resources are also coming under tremendous pressure, as servers in data centres generate substantial amounts of heat during operation and require continuous cooling. Traditional evaporative cooling systems consume approximately 1,500 to 2,500 gallons of water per hour for every megawatt of heat removed (see Note 4).
Moreover, the air handling units, server rooms, and cooling fans in data centres can generate noise levels as high as 90 decibels, sufficient to cause hearing damage (see Note 5). The servers and cooling equipment inside data centres also create a “heat island effect”, increasing surrounding surface temperatures (see Note 6).
The health risks arising from air pollution are particularly concerning. Elon Musk’s xAI data centre in Memphis, Tennessee, USA, installed 35 gas turbines to address power shortages and is estimated to emit 1,200 to 2,000 tonnes of nitrogen oxides (NOx) each year. Such smog-forming pollutants have serious effects on local air quality and residents’ health (see Note 7).
The mounting burden of additional corporate costs
These negative externalities will ultimately translate into business operating costs and risks. When the expansion of data centres affects the vital interests of local communities, community friction can become a huge hidden cost. In various parts of the US, residents have filed class-action lawsuits against data centres over noise and pollution, while local opposition and protests have delayed the site selection and construction of some projects. In June 2026, Texas governor Greg Abbott called for stricter restrictions and regulatory controls on the rapidly expanding data-centre industry, along with the removal of certain tax incentives (see Note 8).
In fact, data-centre development itself faces multiple constraints. The first is a compute-capacity bottleneck. High-end graphics processing units (GPU) are central to AI training, but their production capacity is concentrated among a small number of manufacturers, making the undersupply problem difficult to reverse in the short term. The second is a power-supply bottleneck. The power density of AI data centres far exceeds that of traditional server rooms, while the capacity and transmission and distribution capabilities of existing power grids are clearly insufficient. This has left many projects with equipment in place but no power supply. The third is a cooling bottleneck. The greater the compute capacity, the more heat is generated; excessive temperatures can trigger chip throttling and limit performance. The industry is therefore swiftly shifting from traditional air-cooling technologies to next-generation cooling solutions such as liquid cooling and even immersion cooling, placing greater demands on data-centre infrastructure.
Operational strategies must adapt to the times
For AI innovation companies that build their own compute infrastructure and develop large models, including OpenAI, Google, and ByteDance, data centres themselves are core assets and competitive moats, and these companies directly face the dual pressures of legitimacy and cost. Such enterprises should elevate green compute capacity to the highest level of corporate governance. On the one hand, they should proactively secure stable and clean long-term power sources such as nuclear energy, while seeking to sign long-term power purchase agreements with renewable-energy producers in order to hedge against future electricity-price volatility and carbon regulation risks. On the other hand, they should increase R&D investment in technologies such as compute-capacity optimization and advanced cooling to safeguard their technological advantages. More importantly, companies must actively manage their relationships with local communities and regulators, securing legitimacy through transparent environmental disclosures and substantive compensation mechanisms.
The current explorations of AI innovation companies have already extended to the deep sea and outer space. The Shanghai Lingang undersea data centre is powered by offshore wind energy and uses deep seawater for natural cooling, thereby addressing both energy conservation and emissions reduction (see Note 9). SpaceX in the US is exploring the use of Starship to send data-centre equipment into orbit, powered by solar energy, with data transmitted via Starlink. China has also launched space computing satellites to test in-orbit processing capabilities (see Note 10). Although these explorations are still at an early stage, they constitute notable points of reference in corporate strategic planning.
For AI application companies built on existing models and focused on vertical use cases, compute is a cost rather than an asset. Their strategic focus should shift from owning compute capacity to using it efficiently. First, they should pursue efficiency in model selection and engineering design, avoiding paying for redundant compute through compliant model distillation, quantization, and precise invocation. Second, they should incorporate indicators including compute providers’ renewable-energy share and power usage effectiveness into procurement decisions, thereby compelling data centres to reduce emissions. Finally, they should closely monitor bottlenecks in compute supply and geopolitical risks, and reduce reliance on any single supplier through multi-cloud services.
Data centres are the cornerstone of digital civilization, and their importance is beyond doubt. However, the rapid growth of compute capacity is by no means cost-free. Negative externalities relating to energy use, water resources, noise, and air pollution are continuously accumulating as operational costs for data centres and hidden burdens for society. Only by taking environmental and community costs into account, and by actively exploring various green development pathways while enhancing compute resources, can data centres truly become a new force that supports the digital future rather than mortgaging it.
Note 1: https://openai.com/index/announcing-the-stargate-project/
Note 4: https://www.ampacwatersystems.com/ai-data-centers-water-consumption-crisis-2026/
Note 5: https://www.environmentalhealthproject.org/post/the-dangers-of-data-centers
Note 6: https://edition.cnn.com/2026/03/30/climate/data-centers-are-having-an-underrported
Note 7: https://time.com/7308925/elon-musk-memphis-ai-data-center/
Note 8: https://www.nytimes.com/2026/06/10/us/texas-abbott-data-centers-regulation.html
Note 9: https://www.news.cn/fortune/20260529/8de9d068e5fd46ef984fad9a1d9b9015/c.html
Note 10: https://www.news.cn/liangzi/20260603/5903c373ae314a32927258a971424f27/c.html
