The rapid adoption of AI is creating a significant strain on resources that often remains unseen by the average user. By 2030, AI energy demand could increase fivefold, reaching close to 1,000 TWh (terawatt-hours), or roughly 3% of total global power demand, according to a new report from Arthur D. Little.
Not unlike other data-heavy tech services, running AI models requires an extensive physical foundation of minerals, manufacturing, electricity, and water. While the technology may seem weightless, its expansion is placing an unprecedented strain on global resources, leading to systemic vulnerabilities for businesses and governments alike.
Environmental impact is indeed one of the primary concerns around AI as the industry continues to grow. There has been a major increase in data center construction and expansion, fueled in large part by the growth of AI. A full quarter of the gigawatts powering data centers will serve AI workloads by 2030, according to the report.
“AI feels cheap today because its real economic and environmental costs are essentially hidden, but once dependence sets in, those costs will surface – and companies should be strategically prepared,” said Albert Meige, partner at Arthur D. Little.
Huge resource usage
Global data center emissions are projected to double by 2030, potentially reaching around 1% of all global emissions. This increase is driven by a fivefold rise in energy demand, moving from 90 TWh today to nearly 470 terawatt-hours by the end of the decade.
In addition to that staggering energy use, the physical requirements for cooling these systems are no less significant: A single large data center can consume as much water daily as a medium-sized town. Despite these rising costs, transparency is declining, as fewer than 3% of new AI models disclosed any environmental data in 2024.
Energy supply and grid stability have also emerged as critical bottlenecks. In major technological hubs, data centers could soon account for up to 40% of local electricity consumption. This concentration has already led to significant delays in power connections, with some regions facing wait times of up to seven years for new infrastructure.
More than 40% of the world’s data center capacity is located in the US, though Europe is also home to a large number of data centers, with energy demand expected to reach up to 200 TWh by 2030, according to Arthur D. Little.
Most of the demand for data centers in Europe is concentrated in just a few cities (Frankfurt, London, Amsterdam, Paris, and Dublin). In fact, Ireland represents an extreme case: Data centers consume over 20% of national electricity, prompting a moratorium on new connections in Dublin.
When it comes to new growth in data centers, the Nordic countries are rising stars. That is because Norway, Sweden, Finland, and Denmark have relatively low energy prices and there is an abundance of low-carbon energy, among other perks.
Access to energy and chips
Many large operators are increasingly taking direct control of energy sourcing to bypass grid constraints, sometimes extending the life of fossil fuel plants to meet immediate demand. In the US, several tech giants have struck deals to expand nuclear power and even restart energy production at mothballed nuclear plants.
The infrastructure supporting AI is highly concentrated, creating potential supply chain choke points. For example, a few dominant players currently control the design and fabrication of the specialized chips required for advanced computing. In response, some nations are beginning to treat compute capacity as a new form of geopolitical power.
The United States, China, and Europe are all investing billions to secure domestic production and talent, leading to a fragmented global market where access to technology is increasingly tied to state policy.
Experts warn that the current era of relatively cheap and abundant AI is temporary and supported by early subsidies. As these hidden costs become apparent, businesses face the risks of economic instability and strategic lock-in with dominant providers.
To maintain resilience, the report suggests that organizations must gain better visibility into their actual environmental footprint and prioritize resource efficiency. Building flexible architectures that can move between different providers and jurisdictions will be essential for companies to navigate the uncertain future of this critical infrastructure.
“AI is becoming the world’s newest critical infrastructure,” said Meige. “Its exponential rise poses a question that has so far been underexplored: what are the hidden resource dependencies of AI, and what systemic vulnerabilities do these dependencies create?”