AI data centers greenhouse gas emissions from 11 planned facilities connected to OpenAI, Meta, and Microsoft could exceed the entire nation of Morocco’s annual greenhouse gas output, according to a recent report. This stark comparison underscores a critical blind spot in the AI industry’s expansion strategy: the massive environmental cost of powering the models reshaping technology and society.
Key Takeaways
- 11 planned AI data centers from major tech companies could emit more greenhouse gases than Morocco annually
- AI chip manufacturing emissions surged over fourfold in 2024 compared to 2023, projected to reach 16.8 million metric tons by 2030
- A single ChatGPT query consumes 5–10 times more electricity than a Google search
- Data centers currently account for 1% of global energy-related emissions and are among the fastest-growing sources
- Training GPT-3 alone consumed 1,287 megawatt-hours and emitted 552 tons of CO2
The Hidden Climate Cost of AI Expansion
Most discussions of AI focus on capability and speed. Few address the electricity demand. The scale is staggering. Training large language models like GPT-3 consumed 1,287 megawatt-hours of electricity during its development, equivalent to powering 120 U.S. homes for an entire year and generating 552 tons of CO2 emissions. That was in 2021. Current models are far larger.
The problem intensifies at inference—when users actually interact with these systems. A single ChatGPT query demands 5–10 times more power than a typical Google search. Multiply that by billions of daily queries, and the electricity footprint becomes genuinely alarming. Yet this demand continues to accelerate. Companies are not slowing deployment; they are racing to build new data centers faster than renewable energy infrastructure can support them.
The report’s revelation about 11 planned data centers exceeding Morocco’s emissions is not hyperbole—it is a projection grounded in facility specifications and grid realities. Most of these centers are sited in regions where electricity grids remain dominated by fossil fuels, particularly coal and natural gas. Even with efficiency improvements, a data center powered partly by a coal-heavy grid will emit far more carbon than one running on renewables. The pace of construction means most new capacity will arrive before renewable energy supplies can be redirected to support it.
AI Data Centers Greenhouse Gas Emissions: The Numbers
Global AI-related emissions from chip manufacturing alone rose over fourfold in 2024 compared to 2023, with projections reaching 16.8 million metric tons by 2030. This is not just about training; it includes the manufacturing lifecycle of processors, which is extraordinarily energy-intensive. New data centers could see emissions rise by up to 160% by 2030, potentially consuming up to 20% of all new renewable energy capacity while still producing as much as 121 million tons of CO2 if powered partly by fossil fuels.
Data centers as a whole currently account for approximately 1% of global energy-related greenhouse gas emissions, but they are among the fastest-growing sources. By 2035, data center emissions could add between 0.4 and 1.6 gigatonnes of CO2-equivalent annually. To contextualize: that is equivalent to adding the annual emissions of a large developed nation to the atmosphere every year, driven entirely by the infrastructure needed to run AI systems.
One concrete example illustrates the problem. xAI’s Memphis data center, which uses turbine power, still emits 1,200–2,000 tons of nitrogen oxides annually, making it a top local air polluter in its region. Even with renewable energy sources, data centers generate significant environmental impacts beyond carbon—water usage, heat discharge, and local air quality degradation.
Why AI Data Centers Greenhouse Gas Emissions Matter Now
The report arrives at a critical moment. AI adoption is accelerating, not decelerating. Every major tech company is committing billions to data center expansion. OpenAI, Meta, and Microsoft are among the largest investors, each betting that computational scale will drive AI capability gains. But the industry has not seriously grappled with whether this expansion is sustainable or even permissible from a climate perspective.
Consider the comparison: the 11 data centers in question could emit more greenhouse gas than the entire nation of Morocco. Morocco is not a small country—it has a population of over 37 million people. The idea that a dozen facilities for private companies could outpace a nation’s total emissions is a failure of regulatory foresight and corporate accountability. Yet no binding commitments exist to prevent this outcome. Voluntary pledges and offset programs have repeatedly failed to deliver real emission reductions.
The gap between stated climate goals and actual behavior is widening. AI companies tout sustainability commitments while simultaneously building data centers in fossil-fuel-dependent regions and accelerating deployment timelines that make renewable energy integration impossible. This is not incompetence—it is the inevitable result of prioritizing speed and scale over environmental impact.
Comparing AI’s Climate Impact to Broader Industry Trends
AI’s environmental footprint must be contextualized within the energy sector. While AI computing currently contributes less than 1% of global greenhouse gas emissions, the trajectory matters more than the current share. Oil and gas companies have spent decades arguing their contribution is manageable; meanwhile, their cumulative impact has driven climate crisis. AI’s growth curve is steeper and less regulated.
The difference is that oil and gas companies operate under at least some regulatory scrutiny and public awareness. AI data centers expand with minimal environmental review. A coal plant faces environmental impact assessments. A massive data center often does not. This regulatory asymmetry means AI infrastructure is being built with less oversight than fossil fuel infrastructure—despite potentially comparable or greater environmental consequences.
Is AI’s Climate Impact Unavoidable?
Some argue efficiency gains in AI systems will offset increased demand. This is the rebound effect fallacy. History shows that efficiency improvements typically lead to increased consumption, not reduced overall impact. More efficient AI means cheaper inference, which means more queries, which means more total energy use. The only way to genuinely reduce AI’s environmental impact is to reduce deployment scale or shift to renewable energy grids—neither of which the industry is willing to do at the necessary pace.
The demand for new data centers cannot be met sustainably. The pace at which companies are building new facilities means the bulk of electricity powering them must come from fossil fuel plants. Renewable energy capacity is finite. AI companies are not waiting for green infrastructure to catch up; they are building now and burning coal and gas to do it.
FAQ
How much electricity does training a large AI model consume?
Training GPT-3 consumed 1,287 megawatt-hours of electricity, equivalent to powering 120 U.S. homes for a year, and generated 552 tons of CO2 emissions. Current models are significantly larger and consume more power.
Why are AI data centers built in fossil fuel-dependent regions?
Cost and infrastructure availability drive location decisions. Regions with cheap electricity—often powered by coal or natural gas—offer lower operating costs than renewable-heavy grids. Companies prioritize expense reduction over environmental responsibility.
Could renewable energy solve AI data centers’ greenhouse gas problem?
Theoretically, yes. Practically, no—not at current expansion rates. Renewable energy capacity is finite and growing slower than data center demand. Even with aggressive renewable buildout, most new AI data centers will be powered partly by fossil fuels for years.
The 11 planned data centers connected to OpenAI, Meta, and Microsoft represent a choice: expand AI capability without environmental constraint, or slow deployment until sustainable infrastructure exists. The industry has chosen the former. Regulators have failed to enforce the latter. The result is a climate liability that will compound for decades, all in service of making chatbots slightly faster and more capable. That trade-off should trouble anyone paying attention to climate science.
Edited by the All Things Geek team.
Source: TechRadar
