Leonard Τraistariu
Romania
A short timeline of data centers
Depending on how we define them, data centers have existed since before the advent of the internet. As early as the 1960s, large corporations and public institutions used mainframe computers housed in dedicated facilities to process payrolls, conduct scientific research, and manage records. Since then, technological advances, particularly the development of the internet, have transformed both the scale and the purpose of these facilities.
The emergence of the World Wide Web and online services created a need for infrastructure capable of keeping websites, applications, and digital services continuously available. Building and operating this infrastructure is expensive. Companies must acquire computing equipment, such as servers, storage arrays and networking to connect them, provide reliable power and cooling, maintain their systems, and employ specialized personnel. They must also ensure sufficient capacity to accommodate fluctuations in demand and implement redundancy to minimize the risk of service interruptions. Nobody wants a power outage or a failed hard drive to take their business offline.
For many companies, maintaining their own computing infrastructure became increasingly costly and impractical. This created an opportunity for specialized providers to build and operate large-scale data centers, offering computing power and storage to multiple customers. Companies such as Amazon, Microsoft, and Google developed cloud computing platforms that allow businesses to rent these resources on demand rather than build and maintain their own infrastructure. By pooling demand across many customers and exploiting economies of scale, these providers, called hyperscalers, can offer computing resources without requiring every company to make the same substantial upfront investment.
The result has been the growth of enormous data center facilities that underpin much of the modern digital economy. While most data centers are owned by individual companies, more than 48% of the total global computing power is concentrated in data centers owned by hyperscalers.
More recently, the rapid development of artificial intelligence, particularly large language models (LLMs) and AI agents, has created additional demand for computing capacity. This has intensified investment in data center infrastructure, adding to the electricity, cooling, and material resources required to sustain the expansion of the digital economy.
The scale of this expansion can be difficult to imagine for someone whose experience of the internet is limited to browsing websites, downloading games, or using everyday online services.
Between 2018 and 2025, the number of data centers operated by hyperscale providers nearly tripled, from around 450 to almost 1,300. Ownership of this infrastructure is highly concentrated: Amazon, Microsoft, and Google together account for approximately 58% of global hyperscale data center capacity.
Power consumption
The infrastructure housed in these facilities has also become significantly more power-hungry. Traditional workloads, such as hosting websites, managing databases, and running business applications, now coexist with AI workloads that rely on specialized hardware, particularly graphics processing units (GPUs) and other specialised accelerators. According to the International Energy Agency, the power density of AI servers increased approximately elevenfold between 2020 and 2025. To put this into perspective, an individual rack in an advanced AI data center could require as much power as 65 households.
Meanwhile, demand for computing capacity is growing rapidly. Although chip manufacturers, such as Nvidia and AMD, continue to improve the performance of their processors, these improvements do not keep pace with the growth in demand and are also limited by how small transistors can be made.
So, one way data center operators meet this demand is by deploying more chips, often in increasingly large clusters. This expansion requires not only more computing hardware but also additional electricity, cooling capacity, and supporting infrastructure, like high-bandwidth networking to enable communication between them.
Although some data center operators invest in their own energy infrastructure, in some cases (China is a good example) including renewable energy, this does not necessarily eliminate their dependence on the wider electricity grid.
Data centers require a reliable supply of electricity around the clock. Wind and solar generation are variable and do not always coincide with demand. Unless sufficient energy storage and other sources of electricity are available, facilities must continue to rely on the wider energy system to meet their needs. They must also be able to maintain operations during power outages. To this end, many data centers install backup generators, often powered by diesel, which can supply electricity if the grid fails. Large facilities may have dozens of generators, and some have considerably more, reflecting the enormous power requirements of their equipment and the need for redundancy. Although these generators are generally intended for emergencies and periodic testing rather than continuous operation, their use produces greenhouse gas emissions and local air pollutants, including nitrogen oxides and particulate matter. These pollutants can contribute to respiratory and cardiovascular health problems, creating additional environmental and public health concerns for nearby communities.
As demand for electricity grows, data centers are in competition with households and energy-intensive industries for available electricity generation and grid capacity, contributing to higher electricity prices. These pressures are particularly significant in regions where energy infrastructure is already constrained. These rising energy costs can actually lead to factory closures, which is something that is already happening in parts of the United States, namely in the rust belt, generating job losses in an area already affected by deindustrialisation.
At the same time, the employment benefits of data centers can be much more limited than the enthusiasm in some (not all) bourgeois media might suggest. Although their construction creates temporary jobs, once operational, data centers typically require relatively few employees compared with other industrial facilities of similar capital cost. Their operations rely on specialised staff to maintain servers, electrical systems, cooling infrastructure and security.
Data centers also tend to cluster geographically. Once a region has developed the electricity connections, fibre-optic networks and other infrastructure needed to support them, it can become an attractive location for additional facilities. This creates a reinforcing cycle: new data centers encourage further infrastructure investment, which in turn makes the region more attractive for further development. The consequences extend beyond the boundaries of individual facilities. Where data centers concentrate, their combined demand for electricity, water and land place increasing pressure on local resources and communities. While the computing services they provide may be distributed globally, many of the environmental costs are borne locally.
The expansion of AI infrastructure has also contributed to pressure to keep fossil-fuel power plants operating for longer. In September 2025, US Energy Secretary Chris Wright said that most US coal-fired power plants were expected to delay their retirement to help meet rising electricity demand. Although this demand has multiple causes, the expansion of data centers has become an increasingly important factor. The AI boom has also encouraged investment in new natural gas-fired power generation, including facilities intended to supply data centers directly. Of course, this is not universal. There are countries, such as the Nordic countries, or Canada which have an abundance of energy. However, data centers, due to constraints like the latency introduced by transferring data over long distances or regulatory constraints that require companies to keep parts of their data within their home country, cannot all be built where excess energy is available. In consequence, there are situations in countries like Brazil and Mexico where the energy consumption of data centers causes multiple blackouts per day.
Cooling impact
Most of the energy consumed by an AI data center is ultimately converted into heat. To prevent overheating, which can damage semiconductors and cause hardware failure, data centers employ various cooling technologies. As AI servers become more densely packed with GPUs and other high-power components, conventional air cooling can become insufficient or less energy-efficient, encouraging the adoption of liquid cooling. Some systems use water directly, while others circulate specialised coolants to transfer heat away from the hardware.
Globally, data centers use hundreds of billions of litres of freshwater yearly for cooling their systems (not including the water used to generate the power for their operation).
The water requirements of data centers can influence where they are built, with some facilities located near reliable water sources. Depending on the cooling technology, they may consume substantial quantities of freshwater, competing with agriculture, households and ecosystems for a limited resource. In water-stressed regions, such as areas experiencing droughts, this can intensify existing pressures on local communities and wildlife.
Romania provides a striking example of the vulnerability of water and energy infrastructure to drought. In the summer of 2026, the Danube’s flow fell to exceptionally low levels, disrupting agriculture and river transport. The crisis also affected electricity generation: Romania was forced to shut down both reactors at its Cernavodă nuclear power plant temporarily because of the low water levels, threatening electricity supplies and increasing the need for imports. The Danube is also home to the Iron Gates hydropower complex, an important contributor to Romania’s electricity production. These events demonstrate how water scarcity can affect several essential systems simultaneously and can potentially become one of the causes of a blackout.
At the same time, Romania is planning a significant expansion of its data center infrastructure. The Cluster Power campus near Craiova, for example, has announced plans to expand to approximately 512 MW of capacity, as part of a wider proposed 800 MW development. This is planned capacity, not current electricity consumption, but it illustrates the potential scale of future demand. Other projects are also being proposed, including data centers near Bucharest with a combined planned grid connection of 380 MW.
This is just one of many examples around the world. Where water scarcity is already a problem, or risks becoming a problem because of climate change, data centers, can become an additional environmental and social stressor, raising competition for a resource that is already limited.
India is another example. In 2016, India had to close 18 thermal power plants, because of conditions of severe drought which made the water needed for cooling and power generation insufficient, resulting in a loss of 14 Terawatts-hour of energy generation potential. Since then, India has become a major hub of data center development, its operational capacity increasing from 375 MW in 2020 to 1.6 GW in 2026, making it the second largest data center market in Asia-Pacific. This increased demand for water and energy can make a crisis similar to 2016 even more serious.
Water-related impacts do not end with consumption. Where cooling systems discharge heated water into rivers or lakes, they can cause thermal pollution, raising water temperatures and disrupting aquatic ecosystems.
Warmer water holds less dissolved oxygen, placing additional stress on fish and other aquatic organisms. Temperature changes can also interfere with reproduction, migration and the life cycles of aquatic species, while favouring organisms better adapted to warmer conditions. The severity of these effects depends on the volume and temperature of the discharge, the characteristics of the receiving waterway and the sensitivity of its ecosystem. Although some facilities use cooling systems that avoid direct discharge into natural waterways, their water consumption and the energy required for cooling still carry significant environmental costs. And even in this case, the heat that escapes from a data center, particularly one with high energy consumption, can raise the temperature of the surrounding environment, which can still affect the local environment and farming.
Opposition to Data Centers
Because of all the reasons above, there have already been numerous mobilisations against the expansion of data centers. These have taken different forms: grassroots campaigns raising awareness of their environmental impacts and their connection to broader political questions, such as climate change; local struggles over immediate material concerns, including electricity prices, noise pollution and water consumption; and protests against the construction of specific facilities. The consequences discussed in this article are therefore neither theoretical nor warnings about some distant, bleak future. They are already being felt.
However, localised opposition, when it remains local and does not expand on a national level, has limited power when confronted with the economic and political influence of large corporations. Within the capitalist mode of production, the state plays the role of safeguarding the conditions for capital accumulation, even when doing so conflicts with environmental protection and the interests of local communities. The states in the center of accumulation, the main imperialist countries, are compelled to invest vast amounts of money and resources to develop this new technology, in order to secure their position and the competitiveness of their capital, while states in the semi-periphery and periphery, even if they may be pulled in different directions by the tensions between local and central interests, are ultimately compelled to participate in the advent of this new technology, but with even less control over regulations or where these projects are built.
Although local campaigns have sometimes succeeded in halting developments or securing temporary moratoriums, these victories do not necessarily challenge the broader process driving the expansion of data centers. A project blocked in one locality may simply be relocated elsewhere, while the underlying incentives for further construction remain intact.
The struggle against the environmental consequences of AI infrastructure must therefore develop a mass, political character. It must move beyond isolated disputes over individual facilities and challenge the economic priorities that subordinate energy, water, land and human labour to the pursuit of profit. This means connecting local struggles with the wider workers’ and environmental movements, and fighting for democratic control over how computing infrastructure is developed, what purposes it serves and how its environmental costs are checked. This has to go beyond demands for local workers’ management and control. Data centers, as all other means of production and social reproduction, have to be under the democratic control of a society under the leading role of the working class.
Conclusion
In the context of accelerating global warming, and capitalism’s persistent failure to implement the necessary measures for a transition to less polluting energy sources, particularly during periods of economic crisis, the expansion of AI data centers aggravates the climate crisis in two ways.
First, their growing demand for electricity and water intensifies existing environmental pressures, including water scarcity and the degradation of natural habitats.
Second, this demand prolongs dependence on fossil fuels, contributing to decisions to delay the retirement of coal- and oil-fired power plants and encouraging investment in natural gas generation.
Governments operating within a global system of economic and geopolitical competition face pressure to secure the energy and infrastructure required to attract investment and maintain the competitiveness of their national economies. In this context, commitments to decarbonisation are being subordinated to the immediate demands of capital accumulation, further undermining the supposed energy transition.
The consequences are not distributed equally. Working-class communities face higher energy costs and the loss of industrial jobs, while farmers/peasants and agricultural workers are exposed to crop failures and worsening water scarcity as climate change intensifies. These outcomes have multiple causes, and they cannot all and solely be attributed to AI infrastructure. Nevertheless, the expansion of energy-intensive data centers deepens existing inequalities.
Under a socialist society organised around human needs rather than private profit, AI and associated data centers could instead be developed and deployed according to collectively determined priorities, with its energy use and environmental consequences assessed alongside its potential social benefits.
This would not automatically make AI environmentally sustainable: that would still require democratic planning, investment in low-impact energy systems, and decisions about which applications are genuinely worth their material costs. But freed from the imperative to maximise private profit and capital accumulation, its development could be directed towards socially useful purposes within ecological limits. For example, AI could be used in medicine for applications such as automatic detection of cancer in medical imaging, or in the processing of large databases, which could help in formulating production and development plans, applications that already exists, but are chained by capitalist production relations to serve only the role of increasing profit margins.
In fact, under capitalism, AI is increasingly being developed as another instrument for raising productivity, reducing labour costs and accelerating capital accumulation – even when the resulting environmental and social costs, borne by workers and communities, are huge, while those responsible for this damage benefit and accumulate vast profits.
Moreover, the immense investment that we see nowadays in AI bears the hallmarks of other similar developments in capitalism, namely the developments of railroads or of the internet. A lot of money is poured into the fixed capital necessary for an industry before it has matured in any way, as many companies do not know how to profit out of AI. A crisis is almost inevitable, in which case the immense costs of the loans necessary for the construction of data-centers and supporting energy infrastructure will be paid by the working class, both through austerity, rising energy bills, blackouts and government bailouts and through the environmental destruction that threatens the livelihood of large parts of humanity.
