According to “Emissions Gap Report 2023”, published on November 20th by UN Environment Programme:

  • As things stand, fully implementing unconditional Nationally Determined Contributions (NDCs) made under the Paris Agreement would put the world on track for limiting temperature rise to 2.9°C above pre-industrial levels this century. Fully implementing conditional NDCs would lower this to 2.5°C which we need desperately.

It turned out that Greenhouse Gas (GHG) emissions hit new high, temperature records tumble and climate impacts intensify.  Consequently, temperatures hit new highs, yet world fails to cut emissions (again) finds that the world’s heading for a temperature rise far above the Paris Agreement goals unless deliver more than they have promised.

Here’s a reality check – Global carbon emissions from fossil fuels have risen again in 2023 — reaching record levels, according to new research from the Global Carbon Project Science team.  The annual Global Carbon Budget projects fossil carbon dioxide (CO2) emissions of 36.8 billion tonnes in 2023, up 1.1 percent from 2022.

Nevertheless, it’s worth noticing that fossil CO2 emissions are falling in some regions, including Europe and the USA, but rising overall – and the scientists say global action to cut fossil fuels is not happening fast enough to prevent dangerous climate change.

According to the International Energy Agency (IEA), the majority of CO2 emission reductions through 2030 will come from technologies on the market today. Achieving net-zero by 2050 is nothing short of a momentous endeavor, and the energy sector is playing a key role in reaching it. Technology is critical in achieving carbon neutrality, and to no surprise, startups have a unique role to play. 


While there’s an acknowledgement for the impressive widespread use of Artificial Intelligence (AI) tools and techniques around the world, scientists are also concerned about CO2 emissions being emitted into the atmosphere as a result of creating and using Machine Learning (ML) models in the name of AI.

Here’s an image to illustrate the characteristics of Machine Learning Language Model (MLLM):


As a background, A language model uses machine learning to conduct a probability distribution over words used to predict the most likely next word in a sentence based on the previous entry. Language models learn from text and can be used for producing original text, predicting the next word in a text, speech recognition, optical character recognition and handwriting recognition.

A recent study was conducted by researchers at the University of Massachusetts to determine how much energy is used to train certain popular large AI models. According to the results, training can produce about 626,000 pounds of CO2, or the equivalent of around 300 round-trip flights between New York and San Francisco – nearly 5 times the lifetime emissions of the average car.

The following graph exemplifies the comparison of carbon footprint associated with “Training an AI Model” with four other entities to highlight the magnitude of the footprint :

In 2019, the University of Massachusetts Amherst analysed various natural language processing (NLP) training models available online to estimate the energy cost in kilowatts required to train them. Converting this energy consumption in approximate carbon emissions and electricity costs, the authors estimated that the carbon footprint of training a single big language model is equal to around 300,000 kg of CO2 emissions. This is of the order of 125 round-trip flights between New York and Beijing, a quantification that laypersons can visualize.

According to recent estimates, the global CO2 emissions of the information and communications technology (ICT) sector account for around 2 percent of global CO2 emissions, but this figure is hard to estimate precisely given the distributed nature of global computing infrastructure.

It’s only logical that as datasets and models become more complex, the more energy needed to train and run AI models becomes massive. This increase in energy use directly affects GHG emissions, aggravating climate change. According to OpenAI researchers, since 2012, the amount of computing power required to train cutting-edge AI models has doubled every 3.4 months. By 2040, it is expected that the emissions from ICT industry as a whole will reach 14 percent of the global emissions, with the majority of those emissions coming from the ICT infrastructure, particularly data centres and communication networks. These data demonstrate the urgent need to address AI’s carbon footprint and role in environmental deterioration.


In addition to CO2 emissions generated as a result of using Training AI models, the e-waste produced by AI technology poses a serious environmental challengeE-waste contains hazardous chemicals, including lead, mercury, and cadmium, that can contaminate soil and water supplies and endanger both human health and the environment. 

By 2050, the World Economic Forum (WEF) projects that the total amount of e-waste generated will have surpassed 120 million metric tonnes. To avoid environmental harm and minimize the release of dangerous compounds, proper e-waste management and recycling are essential. To ensure the secure processing and recycling of AI-related electronic waste and significantly reduce its negative environmental effects, more stringent laws and ethical disposal practices are required.


Furthermore, AI applications like driverless automobiles and delivery drones pose a threat to animals and natural environments. Automation fuelled by AI may result in greater consumption as well as increasing waste in certain sectors, such as the e-commerce industry, which has normalized the rapid and frequent delivery of goods. 

The rising use of AI in agriculture could result in the overuse of pesticides and fertilisers, contaminating the soil and water, and harming biodiversity. Implementing AI in agricultural practices to increase yields at the expense of maintaining ecosystem health could lead to monocultures and biodiversity loss. 


Another perspective, AI seems destined to play a dual role. On the one hand, it can help reduce the effects of the climate crisis, such as in smart grid design, developing low-emission infrastructure, and modelling climate change predictions. On the other hand, AI is itself a significant emitter of carbon.

Using AI for environmental management also raises ethical questions. Decisions made by AI systems could be biased if they were presented with inaccurate or incomplete data. For instance, if an AI system received instructions to value economic growth over environmental protection, it might choose to put short-term financial gain ahead of environmental sustainability.


A main problem to tackle in reducing AI’s climate impact is to quantify its energy consumption and carbon emission, and to make this information transparent. Crawford and Joler wrote in their essay that the material details of the costs of large-scale AI systems are vague in the social imagination, to the extent that a layperson might think that building a ML-based system is a simple task. Part of the enigma lies in the absence of a standard of measurement.

AI scientist Richard Sutton, often called the ‘father of reinforcement learning’, wrote a blog post in early 2019 titled ‘The bitter lesson’, saying that AI methods that leverage computation are better and more accurate than those that leverage human knowledge. This mindset divides the AI industry, but what is indisputable is that relying on increasing amounts of compute and data requires ever-increasing power and other infrastructure, which is directly proportional to a rising carbon footprint. For a full grasp of AI’s carbon impact, it is not enough to scrutinize the compute costs incurred by training large models.

Here’s the fact, companies are hesitant to share data about their energy mix. Greenpeace’s Clicking Clean report from 2017 says that many companies who had committed to a 100 percent renewable future are more in a state of status quo than on a transformational path — in fact, despite net-zero-by-2040 pledges, Amazon’s emissions increased by 15 percent last year. The report also points out that despite a drive towards renewable energy in significant markets, in others there has been a concomitant push for fossil-fuel-based energy. One such example is Virginia, USA, the data centre hub of the world, where only 1 percent of electricity comes from renewable sources. Then, there is the nexus of Big Data with Big Oil, the report says. Amazon, Google, Microsoft, Royal Dutch Shell and many others market their AI solutions to companies that work on fossil fuel extraction and use.

The Copenhagen Centre on Energy Efficiency, a partnership between the United Nations Environment Programme and the Technical University of Denmark, is a research and advisory work on climate, energy and sustainable development. Gabriela Prata Dias, head of the centre, and Xiao Wang, programme officer, stress that environmental sustainability should be considered as one of the principles towards responsible development and application of AI:

  • “It is important to note that AI is not only just a tool but a resource demander…[and] the benefits of using such technology should outweigh its drawbacks.”

At the same time, AI development and utilisation breeds a lack of transparency and accountability regarding its environmental impact. Certain companies put their financial well-being and competitive edge ahead of any potential negative effects that AI technologies may have on the environment. Users find it challenging to completely appreciate their environmental footprint due to the complexity of AI systems. Accurate evaluation of their carbon footprint or potential environmental impact is hampered by the secretive methods and hidden data used to train AI models. 

Finally, in order to solve this, more transparent procedures and laws that ensure the creation and application of AI are in line with environmental concerns, are necessary. A responsible approach to AI that prioritizes sustainability will be made possible by working toward greater accountability.

Greely, Ontario, Canada 17 February 2024