English (UK)
Deutsch (Sie)
English (UK)Deutsch (Sie)All in; that is the new Modus Operandi in Big-Tech. Since the release of ChatGPT the world has thrown rationale out the window and set out to create a machine god capable of solving the world formula. While the modern alchemists have yet to deliver anything close to 'General Intelligence', many seem to conveniently overlook the major risks this goose-chase entails.
The popular rise of Large Language Models (commonly referred to as AI) has created widespread consequences for the climate. The subsequent AI-sprawl created a carbon accelerant for the already spiraling climate ecosystem. These risks are created by a powerful few and born by the powerless individual. Regulators must therefore take preemptive measures for the common good before it is too late.
AI capacity has become state agenda across the globe. The required energy influx is yet to be met by carbon neutral sources, resulting in a carbon footprint matching New York City. Not only has AI become a second awakening for the obsolete fossile fuel industry, it threatens the scarce fresh-water resources this planet has left. This is especially worrying, as we are entering an era of water bankruptcy, where key parts of the water system could not be rehabilitated.
The explosion of AI computing capacity demand necessitates preemptive regulation.
Prodromes---early signs of illness---signal a pre-crisis phase, where proactive action can mitigate larger consequences [zotpressInText item="{21070564:6N9FXVFV}"].
This policy-focused framework ensures regulations address environmental risks preemptively.
This essay combines Jaques' crisis prevention framework [zotpressInText item="{21070564:CIEI2J96}"] with the FIVER issue management model, a flexible framework for addressing diverse issues. It consists of 5 steps: Understand the situation and interpreting signals, identify specific issues, assess risk, adapt changes, Respond and communicate the issue. [zotpressInText item="{21070564:E4KBT9C7}"]
1. Assessing the impact and proactively adressing underlying systemic causes of each driver
2. Identifying key stakeholders and their perspectives
3. Establishment of effective signal detection mechanisms
4. Policy changes that should be implementedThe International Energy Agency projects datacenter electricity consumption to reach 945 TWh by 2030, a 2.3-fold increase from 2024. Datacenter demand could trump Germany's current electricity demand (2025) by 2030. Wider AI adoption likely leads to exceeding the electricity needs of over 200 million people. Half of the EU's population.
These projected figures are potentially still underestimating the growing energy demand. From 9.4 Gigawatts in 2024, the demand is expected to reach 23 GW in 2025. Empirical data shows this projection underestimated the demand by a significant margin. Tangible data depicts it to be approximately 29.6 GW of power. Therefore in 2025 AI drew more much power than all Germany households combined. The grave short-term underestimates do not imply a trend of moderation befalling big-tech.
The climate activism NGO, Greenpeace warns about the impact of AI on global warming and rejects the claim that AI carries benefits for the climate. They found the claim "AI benefits the climate" to be an attempt of greenwashing.
Enthusiasts proclaim the short term emissions are a necessary trade-off, postulating AI could in fact solve the climate crisis. This claim is overly optimistic. While specialized applications could contribute to a climate transition, the models attracting major investment---generalized, power-hungry Large Language Models---consume vast resources without providing any climate benefits.
The current lack of transparency surrounding AI hampers the ability to spot potential issues, as most companies do not share the figures required to detect an impending crisis. Without comprehensive reporting by datacenter operators and AI enterprises, monitoring the problem and identifying early warning signs becomes an impossible task.
Through the lifetime of an AI-Service the overwhelming majority of energy consumption stems from inference, not training. Inference is the process of using a trained AI model to make predictions on new data. In contrast to the training process, which is offset over the lifetime, inference related electricity consumption occurs each time end-users interact with the service. Therefore, as adoption of AI services increases, the carbon footprint scales manifold, driven by a rising user base and the exponential growth of model sizes, as predicted by AI scaling laws.
The rapid expansion of the AI industry and the subsequent scramble for computing power hampers the goal of climate neutrality at best and amplifies the use of dirty energy at worst. The International Energy Agency projects that carbon intensive energy sources, required to meet additional electricity demand of AI services, could contribute over 40% of power (2030). While this share would see a relative decline by 15% (in comparison to 2026) the total dirty energy contribution could increase twofold. Consequently more coal and natural gas power plants will be built, accelerating the climate crisis even further rather than containing it.
The AI industry remains unconcerned about their impact on the environment. Cases like the xAI Datacenter imply that if not for regulatory intervention, a significant part of the additional electricity could stem from carbon intensive sources. This indifference necessitates mandatory transparency and carbon neutrality standards in order to enforce climate goals.
Energy hungry datacenters need to be cooled. The cooling systems required for increasingly energy intensive AI-systems pose major risks for increasingly scarce freshwater resources. The most popular solution 'evaporative cooling' is marketed as being less energy demanding. While effective at dissipating heat, large amounts of water are squandered and permanently removed from the water cycle. This method is encouraged by active legislation (Germany). The government is requiring energy efficiency, but not cooling efficiency. Thus shifting the problem from water-intensive energy production to a water intensive cooling system.
Some argue the benefits of artificial intelligence will outweigh the water stress it poses. While significant advancements have been made, they've typically been in specialized models, with a negligible environmental footprint. For Large Language Models however, the adverse is true. Generalized use-cases are expected to provide little value, all the while delivering mediocre results.
The United Nations estimates the water footprint of datacenters to reach 9.3 trillion liters by the end of the decade. This estimate is born of pure optimism. Should the AI promise be kept, the footprint could grow up to 12 trillion liters. Datacenters would consume as much fresh-water as 1.7 billion people (2030).
Virtually every 2nd wasted liter stems from fossile energy sources. Carbon intensive energy sources are the primary culprit in AI's trail of destruction. Consequently one can not overstate the importance of transitioning to clean energy sources.
The subsequent stress AI puts on fresh water resources exacerbates the state of water bankruptcy. The UN warns that water resource loss may be irreversible. As freshwater supplies dwindle, billions of people face famine, conflict, and health crises, making AI's contribution to water bankruptcy a matter of life and death for billions of human beings.
The AI industry affirms it is committed to water efficiency. Critics call it disingenuous. A plausible conclusion given the track record of major actors lobbying EU lawmakers to hide data behind confidentiality.
Without targeted policy on the issues in the AI age, the damage of artificial intelligence to the climate crisis could be significant. While the issues AI poses are discussed in the public domain, the extent of its impact is commonly neglected. Can an arms race be won? What is the price of victory? Climate change is not visiting, it is knocking at the door and here to stay and shape our lives. The only question that remains: How bad will it get?
Projections signaling the expansion of dirty energy sources to meet AI's power demands should alarm regulators. To combat this issue, regulations need to reflect the goal of carbon neutrality. Tackling this problem would have positive secondary effects in regards to freshwater shortages, which have become an increasingly bigger problem.
Wagering water on the tech-bros' prophecies is a reckless gamble. Especially in a world already in the midst of water-bankruptcy. Current legislation is centered around energy efficiency. Energy efficiency is important. However, without simultaneous regulations on water usage, water consumption accelerates in effort to save energy on cooling. This gap must be closed. Datacenters must be both energy- and water-efficient.
A circular economy must also be established in the tech industry. Stockholm, Odense and Frankfurt are demonstrating what studies have already confirmed. By contributing excess heat into the district heating system, operators would make a positive impact, rather than degrading scarce water resources.
People should not be deprived of freshwater in favor of tech enterprises. Recent heat waves, even in moderate climates, highlighted the importance of protecting existential resources. The AI industry appears to disregard this assessment.For every US datacenter in safe regions, two more are being built in areas already under acute groundwater stress. The AI industry is unfit to self regulate, thereby necessitating intervention by lawmakers. Should AI enterprises nonetheless consider extensive water usage to be of necessity, they need to be able to provide it.
Sustainability requires transparency. Currently the worst offenders are not obligated to transparency, just profits. This greed hinders efforts to asses the damage. Critics might argue that this is not an oversight, but rather a calculated strategy in order to prevent public scrutiny and evade regulations. To create the most effective strategy, decision-makers should be aware of all the facts. Regulators need to order transparency by the industry.
Environmental risks caused by AI are accelerating at an alarming rate. The signals are clear; without intervention, AI is a fire accelerant for the climate crisis and a precursor for Arrakis on earth. Record temperatures, wildfires, and heat-related deaths demonstrate that humanity is at the end of its rope. But there is hope! AI's acceleration is still in a pre-crisis phase. Proactive policies like carbon neutrality mandates, cooling alternatives, and transparency obligations can still avert the worst outcomes.
This article is adapted from the essay 'Artificial Intelligence and the Risks for our Environment'. For a deeper dive and literature analysis download the complete essay below.