AI, make or break for our Carbon Emissions?
Artificial Intelligence (AI) has experienced a meteoric rise in the last few years. The Global AI market size is projected to reach $407 billion by 2027, a 353.6% growth from its $86.9 billion estimation in 2022. As of 2023, the UK AI market alone was worth more than £16.8 billion, projected to reach £801 billion by 2035. This growth results from the widespread integration of AI into everyday life; half of US mobile users use AI-powered voice search daily, and its potential business applications, 97% of US business owners believe ChatGPT will enhance their business.
This adoption has been a worldwide phenomenon. A Reuters survey showed that 54% of global correspondents employed generative AI. Forbes reports that China leads company adoption of AI, with 58% of respondents utilising the technology.
The rapid growth of AI has raised concerns about its potential risk. The current regulations attempt to ensure safety when employing AI, mitigating the security risks posed and ensuring fairness in its employment. Meanwhile, each country and company races to play a leading global role. [1] [2] [3] [4]
However, the Climate Action Against Disinformation coalition highlighted concern in a report that mapped the risks that AI poses to the climate crisis, describing the current landscape as lacking regulation and relying on voluntary, opaque, and unenforceable pledges.
A study conducted by Carnegie Mellon University Study on data usage delved into the energy and carbon footprint associated with AI tasks, measured per 100 inferences into a metric known as Code Carbon. The study found generating a single image can equate to around half an average smartphone charge. Per 1000 images generated, carbon emissions equate to driving 4.1 miles in a gasoline-powered car. Text generation, on the other hand, equates to around 2 smartphone charges or 0.026 miles of a journey per 1000 inferences.
The energy consumption of AI varies significantly depending on the model’s purpose and complexity. [5] A Bloom study found that generalised AI, such as ChatGPT, has a substantially higher energy footprint compared to more specialised models, only taking a couple of weeks for their usage emissions to exceed emissions produced by a specialised trained AI. Goldman Sachs further emphasised the energy demands of AI, stating a ChatGPT query requires nearly 10 times the amount of electricity as a Google search.
The rapid expansion of AI is driving a surge in demand for data centres – the giant computing warehouses that power AI systems. Data centres are responsible for a substantial portion of global electricity usage, posing future issues for energy consumption.
The International Energy Agency (IEA) predicts electricity consumption from data centres could double by 2026, reaching 1000 TWh, adding the equivalent of ‘one Germany’ to global electrical demand. Goldman Sachs research supports this projection, expecting a 160% increase in demand by 2030. At a minimum, AI is predicted to represent 19% of this data centre energy demand. As a result, global data centre emissions are forecast to reach ‘2.5bn metric tonnes of Carbon Dioxide equivalent by 2030’, according to Morgan Stanley research (Reuters).
These increasing energy demands have prompted concern for the ability of the world to meet such demands whilst pressures to transition away from fossil fuels continue. As of the 22nd of September, The United Nations adopted a pact to phase out fossil fuel consumption. António Guterres, UN Secretary-General, described the
world as “heading off the rails”, calling upon all countries to act, whilst acknowledging emissions were still rising, to ensure their 2050 net zero emissions goal.
Michael Khoo, Climate Disinformation Program Director at Friends of the Earth, described AI as being such a power drain as to cause “America to run out of energy”. Goldman Sachs analysts predict the US will be required
to ‘invest around $50 billion in new generation capacity for data centres’ additionally causing 3.3 billion cubic feet of natural gas demand by 2030.
The same Goldman analysts predict Europe's demand for power to grow as much as 50%, requiring $800 billion in transmission and distribution costs and up to $850 billion in investment in renewable energy sources. The EU’s aim of climate neutrality by 2050 will likely be hindered if not prevented by these increasing demands.
Microsoft and Google are two leading examples of corporations with rising emissions caused by AI. Microsoft has invested billions of dollars into OpenAI and is building additional AI tools. AI is the predominant cause of its rising emissions, having risen almost a third (29.1%) from its baseline stat in 2020.
Figure 1 2024
Environmental Sustainability Report Sustainability Targets
Figure 2 Microsoft’s
2024 Environmental Sustainability Report fact sheet
Microsoft has made significant development in reducing its direct and energy-related emissions (scope 1 and 2), decreasing by 6.3% since 2020. As a part of its efforts to marry its Carbon-free energy (CFE) goals to its AI ambitions, the company committed to back an estimated $10bn in renewable electricity in May of 2024. Additionally, Microsoft has signed a long-term power purchase agreement (PPA) with a nuclear plant in Pennsylvania as of September 2024.
However, the emissions from its supply chain (scope 3) prevent Microsoft's efforts from being successful. These have increased by 30.9% from its baseline, primarily due to the construction of data centres requiring carbon-intensive materials such as steel, cement, computer chips and hardware. Microsoft's President Brad Smith told Bloomberg Green that the company's climate goals were set "before the explosion in artificial intelligence".
As of February 2023, Google unified all its generative AI products under one name, Gemini. Emissions have increased 48% from its baseline and 13% from the previous year.
Figure 3 Google
Net-zero carbon target
Google has similarly made progress in reducing its direct and energy-related emissions (Scopes 1 and 2). Since 2022, these emissions have been reduced by 13% due to the electrification of its buildings, decreased transport emissions, and decreased generator use in data centres. 64% of Microsoft's energy consumption was carbon-free across all centres, and investment into CFE procurement is ongoing. Additionally, Google has matched its annual electricity consumption with renewable energy purchases (Scope 2) since 2017.
However, emissions from Google's supply chain (Scope 3), which account for 75% of its total carbon footprint, have risen 8%. This has been attributed to emissions from purchased electricity, goods and services purchased, and emissions related to data centre construction. Google acknowledges emissions are likely to continue to rise due to infrastructure investment and growth of AI-related activities, such as its $1 billion investment plans within the UK.
Figure 4 Google
GHG emissions
Figure 5 Google
Scopes 1, 2 and 3
Figure 6 Google
GHG emission per scope
The increasing energy demands raise questions about whether Microsoft and Google can continue to meet their renewable energy targets. Google has purchased carbon credits, including three carbon offtake deals created in late 2023. These equate to approximately 62,500 tonnes of carbon dioxide removal credits, expected to be delivered by 2030. Microsoft has purchased renewable energy credits to offset its current energy consumption. The effectiveness of both these credits is subject to debate. [6] Both are likely to face significant challenges to achieve their goals therefore continued investment in renewable energy and new innovative solutions are critical.
AI does have potential benefits for reducing future carbon emissions. James West, a senior analyst at Evercore ISI, observed a trend with accelerating energy demands with renewable energy development also accelerating at an ever-increasing rate. The IEA predicts that for the first time, renewable energy will surpass coal power output by 2025.
An IEA study found the energy sector can boost efficiency. Smart grids and smart meters send thousands of data points, improving predictions of energy supply and demand. An example is Google's AI subsidiary, DeepMind, allowing for more accurate wind flow predictions up to 36 hours in advance based on historical data. This has allowed Google to alter the timing of tasks such as heavy computing loads from periods of peak energy consumption, or to coincide these tasks with times of peak output.
Alternatively, Google can shift the location of computing tasks to other data centres based on national grid intensity. This geographical shifting of these computing tasks means centres are powered by cleaner energy, reducing emissions. The rapid development of AI and the increasing reliance on AI-powered systems are causing a surge in demand for data centres and energy demand. Despite commitments to renewable energy and CFE sources, it remains uncertain whether supply can sustain such demand.
Corporations have framed AI as an unforeseen circumstance, attempting to absolve themselves of responsibility for meeting their carbon reduction targets. As public awareness of AI's potential carbon emissions increases, pressure on companies to address this issue will intensify.
Whilst steps are being taken in the right direction to mitigate AI's environmental impact, achieving our current climate objectives will depend on our ability to manage future energy demands. If we fail to address this challenge, it will be down to our failings, with severe consequences for our planet.
Oliver Windridge