Even without taking into account the cryptocurrency and artificial intelligence boom, the global Information and Communication Technology (ICT) sector is responsible for as much greenhouse-gas emissions as aviation. The ICT industry has seen tremendous growth in recent decades, driven by trends such as the Internet of Things (IoT), cryptocurrency mining, cloud computing and an overall increasing reliance on the internet and electronic devices. But this growth has raised concerns about the environment, and in particular the carbon footprint. Current estimates indicate that ICT‘s share of global greenhouse-gas (GHG) emissions ranged from 1.8% to 2.8% in 2020, equivalent to that of the aviation sector.
Exhibit 1 shows the International Telecommunication Union’s (ITU) projections for the sector’s emissions in a “business-as-usual” scenario as well as a scenario that is compatible with limiting global warming to 1.5°C, both of which have been extended for this report. In the business-as-usual scenario, assuming the emission-intensity of electricity used remains unchanged, the ICT sector would be responsible for 830 MT of CO2 emissions by 2030.
The data represented here includes emissions from users i.e., Scope 3 emissions, which explains the large values. The ICT sector’s carbon footprint comprises two components – embodied emissions and operational emissions. Embodied emissions cover the emissions originating from the manufacturing and installation of equipment and appliances. Operational emissions stem from the use-phase of these networks and devices, primarily based on the level of electricity consumption and the related emissions from the global electricity mix during that time period. The embodied emissions account for roughly 30% of the total carbon footprint while the operational emissions take a majority share with about 70% of the total emissions.
Exhibit 1: Global GHG emission trajectories of the ICT sector for 1.5˚C scenario
This worrying trajectory does not take into account the boom in cryptocurrency and artificial intelligence, which are contributing their own sizable carbon footprint. Bitcoin, for instance, popularised proof-of-work (PoW) for validating transactions on the blockchain and many others followed suit. But this mechanism has so far proven to be an energy guzzler owing to its high demand for processing power. Bitcoin and Ether alone consume as much electricity as the Netherlands or Austria (Exhibit 2).
Exhibit 2: Combined annualised electricity consumption of Bitcoin and Ethereum vs. that of certain EU member states
Higher electricity prices only allow new-generation energy-efficient computers to stay competitive over time as the production costs for mining each Bitcoin could prove to be too high otherwise. Seeking higher profit margins, crypto miners tend to set up their operations in countries with lower electricity prices, usually developing economies (Exhibit 3), which tend to have a higher share of fossil fuels in their energy mix. As a result, crypto mining contributes heavily to emissions (Exhibit 4).
Exhibit 3: Development of the global Bitcoin hashrate (monthly averages)
Exhibit 4: Bitcoin electricity consumption (yearly)
As of end-April 2023, the global bitcoin hashrate, which represents the amount of mining activity, was double the level in January 2022. If this trend continues unchecked, the energy demand and consequently any associated emissions could prove fatal to climate goals.
The cautionary tale of Bitcoin’s hunger for energy suggests that other emerging and trending technologies should be examined for their environmental impacts. Several researchers at Google, UC Berkeley and Meta, among others, have been studying emissions linked to machine learning workloads (training AI models being one such workload). Focusing on operational energy-related emissions, the training phase for AI is found to be highly energy-demanding and consequently emission-intensive. Their studies compare the energy consumption of various models, one of which is OpenAI’s third Generative Pretrained Transformers (GPT-3), which recorded the highest energy consumption and emissions among the group. For the training phase, the measured energy consumption was 1,287 MWh and the associated operational emissions (location-dependent owing to the energy mix) were calculated to be 552.1 tCO2e.
The operational emissions (from research and development of AI and chips) are localised primarily in the US, given its high research output on AI and AI chips. But the embodied emissions (from manufacturing the chips) are to be found elsewhere. For instance, NVIDIA is emerging as a leader in developing AI chips, but it still relies on Taiwan Semiconductor Manufacturing Co Ltd to produce the chips, which means that the energy mix of the production site and the emissions from transportation should also be considered.
The second part of this piece looks at how the ICT sector can be decarbonised.
