Sustainability & CSR
AI's hunger for (green) power
Artificial Intelligence has the potential to deliver multiple benefits to society but it should not come at the expense of net zero ambitions.
By Bart Linssen
Hardly anybody will have missed the images of Jensen Huang strolling Taiwan’s night markets. His company, Nvidia, is a trillion-dollar business and symbolizes the huge attention that Artificial Intelligence (AI) is getting. AI is everywhere and if it hasn’t entered your daily life yet, it will soon do so.
The growth in AI usage is expected to be substantial over the coming years and Taiwan stands to benefit from it, as a semiconductor powerhouse. The AI market is projected to expand at an impressive compound annual growth rate (CAGR) of 37.3% from 2023 to 2030. By 2027, the market size is anticipated to reach US$407 billion, up from US$86.9 billion in 2022. This rapid growth is driven by AI's increasing adoption across various industries, with 72% of organisations now regularly using AI technologies. The impact of AI is also expected to be significant on the economy, and will, according to Forbes, lead to a 21% net increase to the United States GDP by 2030.
These developments in AI may come at a cost to the climate, however. The power consumption of AI data centres is significantly higher than that of traditional data centres due to the intensive computational requirements of AI workloads. Traditional data centres typically consume around 10-16 kilowatts (kW) per server rack. These data centres primarily handle standard IT operations, web hosting, and storage. In contrast AI Data Centres can consume 40-100 kilowatts (kW) per rack due to the use of resource-intensive Graphic Processing Units or GPU’s, the heart of AI computing power, and other specialized hardware required for AI and machine learning tasks.
The anticipated yearly increase in electricity consumption by AI data centres globally is substantial, with projections indicating a dramatic rise in demand. According to the International Energy Agency (IEA), global data centre electricity demand is expected to more than double from 2022 to 2026, driven significantly by AI applications, with AI contributing to this surge as it becomes an integral part of various industries. Currently, data centres account for about 1-2% of global electricity use, but this figure could rise to 3-4% by the end of the decade due to the escalating energy requirements of AI technologies.
Moreover, the power demands of next-generation AI hardware are escalating rapidly. For example, Nvidia's upcoming GPU generations are expected to consume significantly more power than their predecessors, with some chips projected to draw up to 2,700 watts. This trend indicates that as AI continues to evolve and expand its applications, the energy requirements will not only increase but may do so at an accelerating rate.
Taiwan is facing a significant increase in electricity consumption driven by the rise of AI technologies, with projections by the Ministry of Economic Affairs indicating an annual growth rate of about 3%. The increase in energy demand in Taiwan not only driven by the construction of AI data centres but also from the manufacturing sector, particularly in areas like GPU production and server facilities.
By the end of 2025, up to six AI data centres are set to be constructed in Taiwan. The combined power consumption of the six planned AI data centres isn't specified in detail. However, considering that a typical AI data centre can consume between 20 to 100 megawatts (MW), we can estimate that the combined energy consumption of these six data centres could range from 120-600MW or the electricity consumption of up to 1.5 million households.
Government figures further indicate that it expects AI data centres to consume about 2.4 Gigawatts (GW) of power in 2028, increasing eight-fold from the current figure of 240MW. The power consumption of AI data centres in 2028 would be equivalent to 21TWh, which is almost nine times the power production of Yunlin’s 640 MW offshore windfarm (2.4TWh), or if you prefer the nuclear option, equal to the electricity production of 2 nuclear power reactors.
Will the rise of AI make Taiwan’s net zero 2050 ambitions impossible to achieve?
While AI data centres and new demand for GPUs, servers and others will massively drive up energy demand and make net zero targets harder to achieve, AI can play a significant role in helping achieve this goal. For example, by optimising energy use, improving operational efficiencies, and enabling better carbon management, AI can accelerate the transition to a low-carbon economy.
Some examples of where AI can help the energy transition are:
- Energy optimisation: AI algorithms are used to optimise energy consumption in buildings and industrial processes. For instance, Google uses AI to manage its data centres' cooling systems, reducing energy use by up to 40%.
- Smart grids: AI helps in managing and balancing electricity grids more efficiently. By predicting energy demand and optimising the distribution of renewable energy, AI can reduce the need for fossil-fuel-based power plants.
- Transportation: AI is used to optimise routes for delivery vehicles, reducing fuel consumption and emissions. Companies like UPS and DHL use AI-driven route optimisation to cut down on their carbon footprints.
- Agriculture: AI-powered systems help farmers make data-driven decisions that reduce the use of fertilizers and water, leading to lower greenhouse gas emissions.
- Carbon footprint tracking: AI tools can track and analyse carbon emissions across supply chains, helping companies identify areas for improvement and reduce their overall carbon footprints.
- Renewable energy forecasting: AI is used to predict the output of renewable energy sources like wind and solar, improving the integration of these sources into the power grid and reducing reliance on fossil fuels.
On the other hand, the numbers are discouraging. If indeed power demand by AI alone will add another 21TWh to Taiwan’s energy consumption by 2028 as anticipated, half of the entire buildout of offshore wind between now and 2028 (In the best-case scenario of 9.5GW) will go to AI and maintaining the current status quo.
To keep Taiwan’s goal of net zero by 2050 on track, measures to address the topic are necessary. The policy document released by the Energy Administration on 3 October 2024 indicates the government aims for 30% of electricity to be generated from renewables by 2030 of which 31GW is solar, 15GW is offshore wind, 1GW of biomass, 2GW is hydro and a minor contribution by geothermal. New AI data centres will take a big bite out of that.
Recommendations
Here are some of the things that need to be done to address the situation:
- Get Taiwan’s Corporate Power Purchase Agreements (CPPA) and energy market trading on track to ensure that offshore wind parks and are getting constructed.
- Mandate the pairing of new data centre construction with green energy development: To mitigate the environmental impact of AI, the government should require that new data centre constructions be coupled with the development of renewable energy sources. This can be implemented as a requirement for either the municipality granting construction permits or for the data centre owner. For example, the government could mandate that each new data centre must be powered by a dedicated renewable energy source, such as a solar or wind farm or geothermal power plant, with a capacity sufficient to meet its energy needs.
- Encourage data centre owners to invest in additional renewable energy sources: While many companies that own data centres are members of RE100 and have committed to 100% renewable energy, this commitment does not necessitate the use of new, additional renewable energy sources. The government should encourage these companies to go beyond simply purchasing renewable energy certificates and instead invest in the development of new renewable energy projects to support their data centres. The government could offer incentives such as tax breaks or subsidies to encourage such investments.
- Promote research and development of energy-efficient AI technologies: The government should invest in research and development of energy-efficient AI hardware and software. This could include supporting the development of new algorithms that require less computing power, as well as investing in the development of more energy-efficient hardware. The government could also collaborate with universities and research institutions to foster innovation in this area.
- Raise public awareness about the energy consumption of AI: The government should launch public awareness campaigns to educate the public about the energy consumption associated with AI and the importance of sustainable energy practices. This could include information on how individuals can reduce their own energy consumption related to AI, such as by using energy-saving settings on their devices. Public awareness campaigns can also highlight the benefits of supporting policies that promote renewable energy development.
By implementing these recommendations, the Taiwan government can help to ensure that the growth of AI is sustainable and does not compromise the island's efforts to achieve its net zero 2050 goals.
Conclusion: Balancing AI advancement with sustainable energy in Taiwan
In conclusion, the rapid growth of AI technology presents both opportunities and challenges for Taiwan. While AI has the potential to revolutionize various industries and boost the economy, its increasing energy consumption raises concerns about the island's ability to achieve its net zero by 2050 goals. The construction of six new AI data centres by the end of 2025 and the projected eightfold increase in power consumption for AI data centres by 2028 underscore the urgency of addressing this issue.
To ensure a sustainable future for AI in Taiwan, it is crucial to prioritise the development of renewable energy sources alongside the expansion of data centres. Mandating the construction of green energy sources alongside new data centres, either at the municipal or ownership level, could be a viable solution. This approach would help mitigate the environmental impact of AI by ensuring that its energy needs are met by clean and sustainable sources.
While some argue that AI can contribute to the energy transition through optimisation and efficiency improvements, the sheer scale of projected energy demand from AI necessitates a proactive and comprehensive approach. Failing to address the energy needs of AI could significantly hinder Taiwan's progress towards a low carbon future, potentially diverting resources intended for renewable energy expansion towards maintaining the status quo. Therefore, striking a balance between fostering AI innovation and ensuring sustainable energy practices is paramount for Taiwan's long-term success. Embracing policies that promote both renewable energy development and responsible AI deployment will be critical in navigating this complex landscape and securing a sustainable future for Taiwan.
Bart Linssen is the Director of Renewable Energy at RCI Engineering.
