AI has tremendous potential value, but there are important consequences and tradeoffs to consider. In our last blog post, we discussed the evolution of AI technology and explored how sustainability professionals are beginning to incorporate AI into their work. This blog post focuses on surfacing the emerging social and environmental impacts of AI technology. Capturing the full benefits of AI will require understanding and managing its impacts on nature and people and carefully considering AI use. Our final blog in the series (coming in September) offers precautions and recommendations to help sustainability teams to manage AI-related risks and make the most of emerging AI solutions.

What are the emerging impacts of artificial intelligence?

Approximately 12,000 data centers are already operational worldwide, and another 800 data centers are under construction, with a growing number facing public opposition.

A broad and growing coalition of communities from around the world are mobilising against data center construction, taking legal – and, increasingly, physical – action. Companies who develop and use AI, and especially those where workers are displaced in favour of AI automation, also face a steep increase in reputational risks.

This coalition is motivated and bound by shared concerns over the scale of energy, water, and other natural resources that AI consumes, as well as the effects that AI is having on the rights and resilience of their communities and on workers.

AI and water

AI systems are thirsty, and their water needs continue to grow.

AI’s water usage includes supply-chain water for component manufacturing, onsite water for datacenter cooling, and offsite water for electricity generation

The process for fabricating the technological components required by AI, such as semiconductor manufacturing, requires significant amounts of water, and particularly ultrapure water, which is water that has been filtered to stringent specifications. Taichung, one of the world's most advanced semiconductor plants, uses 100,000 metric tonnes of water per day for manufacturing.

Data centers require water to cool computer servers to prevent overheating. Large data centers can use up to 5 million gallons of water daily, and while some of this water is contained in closed-loop systems, roughly 80% of it is lost to evaporation.

Many data centers are built in dry, inland areas where low humidity reduces the risk of metal corrosion. Often, they draw from local water supplies, competing for freshwater that should be available to residents, which is especially vital in water-scarce regions. As a result, poorly planned data centers increase the threat of water insecurity, and thus increase the risk of dehydration and poor hygiene. Water use also varies based on location and timing. Data centers in cool, humid areas can rely on outside air and will require less evaporative cooling capacity than those in hot, arid areas. Similarly, AI training during the heat of the day – and during stretches of warmer seasonal weather – will have a greater toll on water than AI training during the evening and cooler months.

AI’s water needs also continue to evolve. Recent research indicates that even just 10 medium-length responses with ChatGPT can consume the volume of a standard bottle of water. By 2030, the global data centres powering AI are expected have a water footprint equivalent to the basic annual water needs of all 1.3 billion people in Sub-Saharan Africa.

AI and climate change

The expansion of AI is also having a significant impact on the release of greenhouse gas emissions. AI chip manufacturing primarily occurs in regions heavily reliant on fossil fuels for electricity, and demand for power for AI data centres is outpacing renewable energy capacity and investments.

To meet the present and future power needs of AI, companies are turning to fossil fuel energy capacity. Despite growing efforts to require data centers to source most or all of their power from new green energy builds, much of the new – or revitalised – power infrastructure is powered by fossil fuels, including coal, diesel, and natural gas.

This has the potential to lock in long-term dependence on fossil fuels and increase the release of carbon emissions, nitrogen oxides, methane, volatile organic compounds, and fine particulate matter. Beyond climate change impacts, these pollutants also have a significant toll on the health of local communities. A 2025 model indicates that U.S. data centers in 2030 could contribute to approximately 600,000 asthma symptom cases and 1,300 premature deaths, exceeding one-third of asthma deaths in the U.S. each year and resulting in public health costs in excess of $20 billion.

ChatGPT produces an estimated 4.32 grams of CO₂ for every query, and Google receives 5 trillion queries per year. The carbon footprint per query is growing rapidly, with new research indicating that agentic AI consumes between 62 to 136 times more energy per query than a single conventional AI inference.

New research also finds that AI could help to produce more oil and natural gas, producing a climate impact that could increase the concentration of greenhouse gas emissions in the atmosphere and significantly outweigh the technology’s benefits for renewable energy. Assessments of AI’s climate impact are often framed as a tradeoff between data center energy use and the emissions AI may help avoid through renewables and efficiency gains – unfortunately, such assessments generally fail to account for emissions being enabled from using AI to make fossil fuel production cheaper and more profitable.

AI and its impacts on community resilience

Globally, AI is having a broad and growing range of impacts on communities.

Unintended biases in training data and algorithms can reinforce and amplify distortions, contributing to discriminatory outcomes that reflect systemic biases.

Malicious actors are increasingly exploiting AI technologies to spread sophisticated misinformation and disinformation and to manipulate and influence people’s decisions and actions. AI contributes exponentially to the consolidation of power, threatening democratic society with reinforcing and increasingly irreversible authoritarian controls.

AI is also helping to enable sophisticated and novel cyberattacks that are compromising privacy and security. Bad faith actors are using AI tools for phishing scams and to mimic identities (and steal them through synthetic identity fraud). They are also manipulating AI technology to violate security and privacy measures to gain access to sensitive user data, such as by tricking chatbots into providing or re-assigning log-in credentials.

Data privacy and intellectual property infringement are likely to be ongoing issues. LLMs are dependent on massive volumes of training data, which is often obtained by web crawlers scraping data without user consent. There is also a complex and developing issue of determining ownership of AI-generated content.

Impacts on electricity costs are also a concern. Despite efforts to regulate power rate increases or to allocate costs to large load customers, data centers are driving up rates for customers – sometimes significantly – when operators add new data center infrastructure to the utility grid. In the U.S., electricity can cost as much as 267% more for a single month than it did five years ago in areas located near significant data centers, and some citizens are losing power entirely as energy providers redirect electricity to data centers.

Data centers are also a growing source of noise pollution, with diesel generators and HVAC systems producing a steady hum that can be highly disruptive to neighbouring residents and wildlife.

AI and worker wellbeing

AI has already disrupted the job market, and there are growing concerns of AI automation displacing additional workers and driving a net loss in jobs. Although AI is driving growth in some tech roles, such as machine learning specialists, robotics engineers, and digital transformation specialists, it is also driving a decline in junior developer, clerical, secretarial, data entry, and customer service roles.

In the US alone, it is estimated that AI is already creating a net loss of 16,000 jobs per month. The WEF estimates that 92 million jobs will be displaced by 2030, and although it is estimated that AI could create 170 million new jobs, the technology is evolving quickly. AI is becoming significantly more capable than predictions can account for, which exposes a growing number of roles to AI-related disruption and possible obsolescence. 

In our next blog, we highlight key precautions and recommendations to help your organisation limit its exposure to AI-related risks and to prepare for evolving assurance expectations.