I asked Meta’s Artificial Intelligence (AI) tool to suggest a title for this article. By doing so, I may have used 10 times the electricity required to answer a traditional Google query*. As the use of AI becomes more widespread, the demand for electricity is expected to rise. Currently, data centres account for about 2% of US electricity consumption, according to the US Department of Energy. However, the Electric Power Research Institute (EPRI) estimates that data centres could grow to consume 4.6%-9.1% of US electricity generation annually by 2030.
Before ChatGPT arrived in November 2022, the growth in power demand from computing and data storage had been largely offset by data centre efficiency gains, as small, relatively inefficient corporate data centres were replaced by large cloud computing facilities. Furthermore, Koomey’s Law also suggests that as computing devices advance, the amount of energy necessary to do a set amount of computing falls by half every two and a half years. The arrival of AI computations, however, represent a steep increase in power demand that could well exceed efficiency gains. Part of the uncertainty in EPRI’s estimated range is due to varying assumptions on the extent of efficiency gains in the future.
Big tech is preparing
The ‘big tech’ companies, which are investing heavily into AI, are already responding to the anticipated rise in power demand. With aggressive clean energy targets, many of the technology companies are likely to seek low or zero-carbon solutions. Several nuclear power deals have been making the headlines. Nuclear plants do not emit carbon. They also have the advantage of operating continuously, unlike renewable power sources, such as solar and wind, that have dependencies on the weather. However, nuclear power plants can take a long time to build and can be very expensive, not to mention safety concerns.
In March 2024, Amazon Web Services (the largest cloud service provider) bought a data centre campus in Pennsylvania for USD 650 million which is directly connected to a nuclear power facility. Then in October 2024, Amazon said it would finance the construction of several small modular nuclear reactors (SMRs) in Washington state and invest in a start-up which will build the reactors. SMRs do not currently operate in the US but it could enable faster and less-costly construction compared to large nuclear plants.
Earlier in October, Alphabet’s Google also announced it is backing the construction of seven SMRs, to add 500 megawatts of nuclear power in total. The first reactor is expected to be online by 2030, followed by others through to 2035, to power Google’s global data centres and offices.
Meanwhile, Microsoft announced in September 2024 that it was backing the restart of a nuclear plant in Pennsylvania to provide electricity for its data centres in a 20-year supply deal.
Besides nuclear power, the big tech companies have also been investing in solar and wind power, which are more readily available today compared to nuclear power, as well as geothermal power, another form of renewable energy. But given the significant increase in power demand driven by AI, new nuclear capacity is seen as essential by the big tech companies.
Investment implications
The growth in US electricity consumption this year and the anticipated future increase have been major drivers of the strong performance of the utilities sector, one of the top performing sectors in the US this year. We believe demand for clean power, especially from big tech companies, favours independent power producers who have the ability to craft customized power contracts including long-term power purchase agreements. Nuclear power operators and the associated supply chain are also well positioned in this environment. However, in the near term, the US utilities sector appears overbought, which is likely to result in some consolidation. We would use any pullback in US utilities to build a core allocation within a portfolio.
In addition, there are companies in the industrial sector that can benefit from growing electrification and the expansion of utilities’ power capacity. There are also attractive companies that benefit from the buildout of data centres, such as companies that provide cooling systems which are critical for the efficient operation of the data centres, as well as power and monitoring solutions. An uninterruptible power supply system is seen as essential for data centres, given the need to provide backup power and prevent costly downtimes.
AI development remain in its early days, in our view, with further software and hardware applications to come. This should be supportive of the demand for advanced semiconductors used in AI applications, as well as software companies that can adopt AI to improve efficiency and enhance value for their users. We have a favourable view of the US technology sector, driven by ongoing investments in AI. Although valuations are elevated, we expect interest rate cuts to mitigate such concerns and earnings growth to drive the sector higher.
Risks to our view would include a significant slowdown in AI investments, weaker-than-expected economic growth that curtails investments into technology and escalation in geopolitical tensions.
(*an EPRI report notes that, at 2.9 watt-hours per ChatGPT request, AI queries are estimated to require 10x the electricity of traditional Google queries which use about 0.3 watt-hours each. This article’s title remains what the author already had in mind)
Written by Fook Hien Yap, Senior Investment Strategist, Standard Chartered Bank’s Wealth Solutions Chief Investment Office.