AI’s Climate Impact Is Smaller Than Expected, New Research Shows

The artificial intelligence (AI) frenzy has sent the market to new highs. 

Billions of dollars continue to flow into chipmakers and data centers, as the ongoing AI mania keeps the turbocharged market going for more than two years with no signs of slowing. However, many have begun to question if we are in an AI bubble.

Analysts such as those at JPMorgan (JPM ) are actually calling for investors to brace for turbulence ahead. But while froth is building in both private and public markets, the mania surrounding AI isn’t without its merit, as the technology has a lot of economic value in terms of creating accessible, powerful intelligence, which is believed to resemble the creation of the internet. 

In its 2026 outlook, JPMorgan called AI “the most transformative technology since computing,” and “driving more GDP growth than consumer spending.”

At the same time, the bank warned about power shortages, water constraints, and regulatory scrutiny.

More broadly, the rapid expansion of AI comes with significant environmental pressures that are found to be lower than expected. That’s because the energy intensity and emissions of a query depend on factors like the model’s type and size, the output being generated, the energy grid powering the data center handling the request, the time of day it’s processed, and other variables.

JPMorgan also estimates that 60% of jobs in the developed world face some level of automation risk from AI, but as old roles disappear, new roles should emerge.

Overall, the biggest risk, according to the bank, “is not having exposure to transformational technology.”

AI Adoption, Energy Use, and Climate Burden

The ability of AI to automate repetitive tasks, improve decision-making, and enhance efficiency and productivity across various domains has led to its widespread adoption.

According to a survey by McKinsey, 88% of respondents reported regular AI use in at least one business function, a 10% jump from a year ago. However, it noted that at the enterprise level, the majority are still in the experimentation stage, with about one-third saying their companies have begun scaling their AI programs.

Currently valued at about $400 billion, the AI market is projected to reach $1.8 trillion by the end of this decade.

When it comes to investment, private AI funding in the US surpassed $109 billion in 2024, which is about 24 times that of the UK at $4.5 billion and 12 times that of China’s $9.3 billion. Looking ahead, a whopping 92% of companies plan to invest in generative AI over the next three years.

While AI adoption is growing thanks to its promises of unprecedented productivity, it comes with serious energy and environmental challenges. The thing is, AI is extremely energy hungry. 

Training AI models requires tons of energy, enough to power several hundred households annually, and it can be much higher in inference. It is actually estimated that 80% to 90% of computing power for AI is used for inference.

So, AI’s energy usage is primarily associated with computational power required to train and run these models within data centers, where AI models are loaded onto groups of servers equipped with GPUs like Nvidia’s (NVDA ) Blackwells.

Data centers are facilities housing computer servers, data storage systems, power supplies, cooling systems, and networking equipment. All of this infrastructure is key to not only providing the latest digital services like asking ChatGPT a question, but also sending an email or streaming a video.

So, data centers have been around for a long time, but it’s in the last few years that they have expanded substantially. 

Today, there are more than 100,000 data centers, spread out all over the world, with the US hosting the most at over 4,200, followed by the UK and Germany, both of which have about 500 each. In the US, a third of data centers are in just three states: California, Texas, and Virginia, with the last two heavily dependent on fossil fuels for their energy needs.

Broadly, data centres fall into three buckets. Traditional enterprise facilities are run in-house by companies for their own workloads: colocation providers lease rack space and power to many different customers in shared buildings. At the high end are “hyperscale” campuses—vast, warehouse-like sites packed with tens of thousands of servers that handle the heaviest AI and cloud computing jobs.

According to JPMorgan, hyperscalers’ capital spending is projected to top $500 billion next year as companies race to build data centers and secure scarce power.

So, how much energy do these data centers use? Well, a lot. Data centers in the US consumed 183 TWh of electricity last year, when the nation’s total annual electricity consumption hit a record high.

The energy consumption of these data centers is estimated to grow by 133% by 2030, hitting 426 TWh. Globally, the IEA expects electricity generation dedicated to data centres to more than double in its base case, from around 460 TWh in 2024 to just over 1,000 TWh by 2030.

In the US, data centers accounted for 4% of the total electricity consumption, which is equivalent to the annual electricity demand of Pakistan. After remaining stagnant for more than a decade, the adoption of AI is projected to increase this share to 12% by 2028.
Swipe to scroll →

As for how much of this data center energy consumption is AI responsible for, that’s hard to tell, as data centers handle different types of workloads. A typical AI-optimized hyperscaler, however, annually consumes as much electricity as 100,000 households.

According to the recent projections by Lawrence Berkeley National Laboratory, in the next three years, more than half of the electricity consumed by data centers will be used for AI, which will be equivalent to the annual electricity usage of 22% of all US households.

The Energy Mix Powering the Core of AI 

With the rapid rise of AI, data centers are now experiencing a strain, facing growing pressure as demand for electricity surges.

Most of the electricity, an average of 60%, that is used by data centers actually powers the servers processing and storing digital information. Then, cooling systems are responsible for the second-most energy use, ranging from 7% to 30%, depending on the efficiency of a facility.  

These cooling systems prevent servers from overheating, which requires a large amount of water. In 2023, the US data centers directly consumed about 17 billion gallons of water, with 84% of it consumed by colocation and hyperscale facilities, which alone are expected to consume about 16-33 billion gallons of water annually by 2028.

As for sources of the energy used by data centers, natural gas supplied the most (over 40%) of electricity to data centers in the US, followed by renewables like solar and wind (24%), nuclear power (20%), and coal (15%).

Tech giants like Google (GOOG ), Amazon (AMZN ), and Meta (META ) have actually pledged to use more nuclear power, which currently only accounts for 20% of electricity supply in the US, to reduce the carbon emissions of data centers.

This is why IEA projects CO2 emissions from electricity generation for data centers to peak at around 320 Mt CO2 by 2030, before seeing a decline to about 300 Mt CO2 by 2035.

According to IEA:

“Despite rapid growth, data centres remain a relatively small part of the overall power system, rising from about 1% of global electricity generation today to 3% in 2030, accounting for less than 1% of total global CO2 emissions.”

But the public doesn’t think so. As per a 2024 Pew Research Center survey regarding AI’s broader environmental impact over the next two decades, a quarter of US adults think the impact will be negative, and the same share says the impact will be equally positive and negative.

Both companies and researchers are also continually seeking ways to reduce the energy use of both computation and data centers.

In fact, considerable energy efficiency gains have been achieved in the hardware used for computation. However, the rate of efficiency advantages has been decelerating while the AI’s computational demands have been accelerating.

The thing is, the effects of AI on energy use go beyond direct electricity use for computing. Energy, after all, is a critical input in almost all economic activities, powering industries and supporting modern life’s infrastructure.

And studies show a strong correlation between energy consumption and economic output, indicating that energy use is closely linked to GDP growth. So, if AI enhances economic productivity, it may also lead to a spike in aggregate energy use. Not to mention, the continuous use of fossil fuels to power the economy will amplify the environmental effects of electricity production by contributing to climate change.

But when it comes to AI, it can actually reduce energy use. That can be achieved through demand-side management or improving energy infrastructure resilience.

In the near term, though, reliance on fossil fuels for energy generation results in increased air pollution, degraded water quality, and exacerbated climate change.

While natural gas is expected to remain the primary power source for data centers in the near term, the global research and advisory firm Gartner anticipates “rapid growth in battery energy storage systems to balance the fluctuations of solar and wind energy” over the next three to five years. The firm said the following in a research note:

“New clean on-site power alternatives – such as green hydrogen, geothermal and small modular reactors – are beginning to emerge and will become viable fuel alternatives for datacentre microgrids by the end of the decade.”

AI’s Green Potential and Climate Emissions Trade-Offs

While AI has clear and massive environmental consequences, new research has found that the environmental risks of current levels of AI use are lower than we think. Also, it can actually support environmental progress and economic growth.

In order to project the potential environmental outcomes of AI if it continues to expand at its current pace, in the study titled “Watts and bots: the energy implications of AI adoption¹” published in Environmental Research, scientists from the University of Waterloo and the Georgia Institute of Technology combined data on the US economic activity with estimates of how widely the technology is being adopted across different occupations and industries.

At the industry level, the scientists estimate annual increases in energy use to range between 0 and 12 petajoules (PJ) while carbon emissions could range from 0 tonnes to 272 kt (ktCO2).

How Much Energy and CO₂ Could AI Add at the National Level?

Another 28 PJ of energy use, which is about 0.03% of annual national energy consumption, could result from AI adoption when aggregated across the economy. It would also add 896 ktCO₂ in annual emissions, equivalent to about 0.02% of the country’s yearly CO₂ output.

This is because, according to data from the US Energy Information Administration, 83% of the nation still depends on fossil fuels, i.e., coal, petroleum, and natural gas.

Formed from the fossilized remains of organisms, they take millions of years to form and are considered non-renewable resources. While finite in supply, fossil fuels serve as crucial energy sources for electricity generation, transportation, and industrial processes. Notably, the use of these fuels releases heat-trapping greenhouse gases (GHG) and contributes to climate change.

Researchers have found that although AI-related electricity consumption in the US is comparable to Iceland’s total energy usage, this amount remains too small to register meaningfully at the national or global level.

“It is important to note that the increase in energy use is not going to be uniform. It’s going to be felt more in the places where electricity is produced to power the data centers,” said environmental economist, Dr. Juan Moreno-Cruz, a professor in the Faculty of Environment at Waterloo and Canada Research Chair in Energy Transitions. From the local perspective, he noted it could be a “big deal,” with some places seeing double the amount of electricity output and emissions. 

At a larger scale, though, “AI’s use of energy won’t be noticeable,” Moreno-Cruz added.

While the researchers didn’t look into the effects on local economies where the data centers are located, they found some encouraging results.

“For people who believe that the use of AI will be a major problem for the climate and think we should avoid it, we’re offering a different perspective,” he said. “The effects on climate are not that significant, and we can use AI to develop green technologies or to improve existing ones.”

It can contribute to solutions for energy efficiency and emissions reduction by optimizing renewable energy sources and industrial processes, the study noted.

To draw their conclusions, researchers analyzed different sectors of an economy, jobs in those sectors, and what part of them could be done by AI. They plan to repeat the study beyond the US, in other countries, to measure the impacts of AI adoption globally and get a more comprehensive picture of the technology’s impact on energy use and emissions.

Noting the limitations of the study, the researchers pointed to limited data availability and varying granularity affecting the analysis, a lack of information on the spatial distribution of energy use, and presuming that AI impacts productivity only through tasks that were previously done at a lower cost, and not accounting for the technology introducing new tasks or affecting other forms of production like capital. 

With their study, the researchers aim to provide a helpful base to understand the broader implications of wider AI usage across the economy.

And their findings indicate that the magnitude of increase in energy usage and emissions from AI adoption is “relatively modest compared to overall economic activity.”

So, as AI revolutionises different sectors and reshapes the various aspects of our society, the study calls for the need to balance the economic benefits and productivity gains with potential increases in the demand for energy and related carbon emissions. 

In order to strike that balance, the study mentions investing in renewable energy sources, emphasizing those AI technologies that are energy-efficient, developing strategies to offset rising emissions in AI-heavy industries, and using AI to mitigate vulnerabilities to climate change.

Through constant analysis and monitoring of AI’s energy and environmental impacts, we can achieve sustainable development of transformative technologies, said the study.

Investing in AI: Data Centers, Chips, and Climate Risk

The biggest winner of the ongoing AI mania is Nvidia (NVDA ), a full-stack computing infrastructure company that offers AI solutions and software in addition to having a gaming segment, professional visualization, and robotics.

Nvidia is the world’s most valuable company with a market cap of $4.3 trillion, which surpassed $5 trillion last month when its shares hit a 52-week high of $212. As of writing this, NVDA shares are trading at $179.5, up 33.2% YTD and more than 1,450% in the last five years.

As a result of massive spending by AI companies on infrastructure, chip maker Nvidia’s data centre business posted a record $51.2 billion in revenue, increasing 25% from the previous quarter and 66% from a year ago. Its clientele includes Google, Amazon, Meta, Microsoft, and Oracle.

According to the firm, this business has been fueled by an acceleration of powerful AI models, agentic applications, and computing. The sales of Nvidia’s Blackwell GPU chips have also been “off the charts,” and cloud GPUs have sold out.”

As a result, the company announced AI factory, specialized computing infrastructure, and other infrastructure projects totaling 5 million GPUs, spanning “every market, CSPs, sovereigns, modern builders, enterprises, and super computing centers.”

“Compute demand keeps accelerating and compounding across training and inference — each growing exponentially. We’ve entered the virtuous cycle of AI. The AI ecosystem is scaling fast — with more new foundation model makers, more AI startups, across more industries, and in more countries. AI is going everywhere, doing everything, all at once,” said CEO Jensen Huang in the recent Nvidia Q3 earnings statement.

Latest Nvidia Corporation (NVDA) Stock News

Conclusion: AI’s Climate Impact and Investment Takeaways

With AI becoming more personalized and capable, gaining the ability to reason and solve complex problems, its adoption and energy consumption are still in their early stages. The expansion of this technology is driving massive buildouts of data centers, pressing existing power systems, and raising concerns about electricity demand and carbon emissions. 

However, as the latest research suggests, AI’s nationwide energy impact remains modest, though its local footprint can be significant. It all comes down to the underlying power mix shifting toward low-carbon sources. So, AI is not just a driver of energy consumption but also a tool for energy optimization, helping modernize grids, improve efficiency, and accelerate climate-focused innovation while unlocking the economic value it promises.

Click here for a list of undervalued disruptive tech stocks.

References

1. Harding, A. R. & Moreno-Cruz, J. “Watts and bots: the energy implications of AI adoption.” Environmental Research Letters 20 (11), Article 114084 (2025). https://doi.org/10.1088/1748-9326/ae0e3b