AI’s exponential energy boom: Can clean energy keep up?

At a glance:

• Energy use from data centres has skyrocketed to power the AI race, which has led Big Tech to re-evaluate their long-standing carbon footprint commitments.

• To understand how the AI boom impacts achieving the Paris Agreement goals by 2050, we compare clean energy electricity demand and supply, grid readiness, and policy environments in countries such as the U.S. and Canada.

• We discuss how Microsoft and Google, two of the largest AI developers, are pivoting their climate strategies and whether this will accelerate renewable energy production or continue reliance on fossil fuels.

• With Tech playing a prominent role in ESG indexes given the sector’s historically low carbon footprint, we provide key considerations for investors in this new AI-era.

Computational demand for artificial intelligence (AI) is expanding at an unprecedented rate. It is estimated that the computing power required for AI doubles every 100 days and is projected to increase by more than a million times in the next five years. This exponential growth promises transformative applications across various sectors including industrials, agriculture, healthcare, finance, and transportation, among others.

In an earlier piece, we addressed the technology sector’s commitments to transition to 100% clean energy by 2030. However, the surge in energy demand due to AI has led to an increase in fossil fuel use, delayed the retirement of certain coal plants, and resulted in a spike in GHG emissions at a time when the pathway to meeting the 2050 goals of the Paris Agreement is rapidly narrowing. Furthermore, the challenges compound when we consider the electricity demand from cryptocurrencies, which is expected to rise by 40% by 2026.

In many regions, the availability of power grid connections has become a challenge as data centres expand amid accelerating electrification, raising concerns that that our current energy infrastructure is unsuitable to support the projected electricity demand growth. Some AI proponents argue that future developments will help make those same grids more efficient, and AI-enabled optimization can lead to exponentially higher megawatt hours of zero-emission energy. However, it is premature to predict the overall net impact of AI as the technology is still at a very early stage of development.To understand whether AI will help or hinder climate goals and the energy transition, we take a look at the sources of energy we use to power it, where data centres are most likely to be housed and what the renewable energy capacity of those areas is, and how the tech titans plan to pivot their climate strategies.

The current state of data centre electricity demand and clean energy supply

For the past fifteen years, electricity consumption has remained relatively flat in the Global North. However, those days are over. Artificial intelligence is a main force of a new age of electricity demand growth, powered by data centres across the globe. Data centres are now among the fastest-growing consumers of electricity due to increased digitalization, artificial intelligence, cloud computing, and crypto mining. In fact, the IEA estimates that global data centre electricity demand will reach 1,000 terawatt-hours (TWh) in 2026 – nearly equivalent to Japan’s total electricity demand in 2023.

The IEA’s 2024 Electricity Report forecasts that by 2026, the AI industry’s electricity demand will expand at least ten times its demand in 2023. It also estimates a 40% increase in electricity demand from cryptocurrencies, though this is subject to uncertainties due to technology innovation in the crypto space. Notably, the Crypto Carbon Ratings Institute (CCRI) estimates that Ethereum, the biggest blockchain network by transaction volume, reduced its energy consumption by 99% in 2022 after changing its consensus mechanism (i.e., how it validates transactions) from proof-of-work to proof-of-stake. In contrast, the Bitcoin blockchain is estimated to have used 120 TWh in 2023, similar to the total electricity consumption of the Netherlands in 2022.

The IEA expects global clean energy investments to reach US$2 trillion in 2024, which while substantial, is insufficient for renewables to keep up with the sharp increase in electricity demand. The burgeoning demand underscores the need for a rapid scale-up in renewable energy infrastructure to meet an exponentially increasing and likely under-forecasted electricity demand.

For now, due to the massive volume of electricity demand, data centre providers are using a mix of energy sources, including natural gas and coal. Electricity demand is simply too high to be met by clean energy alone. As we move forward, to achieve global climate goals and meet data centre electricity demand, governments and the private sector will need to make significant investments to upgrade power grids and build more clean energy supply.

United States

The United States currently hosts one-third of the world’s 8,000 data centres. Projections for future data centre electricity demand are highly uncertain, with consulting firms, investment banks, and other organizations estimating that data centres could account anywhere between 4% and 10% of total U.S. electricity demand by 2030 – ranging from 280 to 580 TWh (Exhibit 1).

Exhibit 1

Forecasts for US data centre electricity demand from now to 2030 vary wildly among consulting firms, investments banks, and other organizations. Estimates for US data centre electricity demand in 2030 range from 280 TWh to 580 TWh, accounting for anywhere between 4% and 10% of total U.S. electricity demand in 2030.

Addressing the urgent need for new infrastructure to meet the rising electricity demand from data centres is a major focus in the U.S. In May 2024, the U.S. government, in partnership with 21 states, launched the Federal-State Modern Grid Deployment Initiative, recognizing the rapid growth of large data centres and the need to modernize and accelerate improvements to the electrical grid.

While the US is actively expanding its renewable energy capacities, the challenge remains substantial given the scale of projected energy demands. For 2024, the U.S. is on track to add 63 gigawatts (GW) of new electric-generating capacity (a 55% increase year-over-year), with solar and battery storage making up 81% of the additional capacity. In addition, the Inflation Reduction Act (IRA) further bolsters economic growth in clean energy, fueling plans for over US$200 billion in clean energy technologies like batteries and solar photovoltaic (PV) power.

Table 1

More than 40% of the U.S. electricity generation mix comes from natural gas; almost a fifth of the mix comes from nuclear power, following by coal with more than 15%. Wind power makes up 10% and hydropower and solar energy make up 6% each. In Canada, hydropower makes up almost 60% of the electricity generation mix, followed by natural gas and nuclear power with 15% and 14% of the mix, respectively. Wind is 6% of the electricity mix, followed by coal with 4%.

Sources: Ember

Nonetheless, a possible change in the White House administration this year could pose a new headwind for clean power development, compounding recent challenges such as rapid interest rate hikes, rising commodity costs, and supply chain disruptions.

Canada

Here in Canada, while specific projections for AI-driven electricity demand are less clear, the general trend is similar, with an increasing strain on electrical grids due to the growth of data centres and digital infrastructure. In Budget 2024, the federal government announced a CA$2.4 billion package to support the AI sector, including building infrastructure for new computing power and financial assistance to workers who have been disrupted by AI. However, the plan overlooks investments in energy projects to supply the incremental electricity demand or measures that would help mitigate the environmental impact of AI.

Exhibit 2

Among the G20, Canada has one of the lowest emission intensities of energy generated (represented as grams of CO2 equivalent per kilowatt-hour). In 2023, Canada registered approximately 170 gCO2e per kWh. It is only above Brazil, which registered around 98 gCO2e per kWh in the same year. The United Kingdom lies above Canada, with 370 gCO2e per kWh in 2023.

Sources: Ember, Energy Institute – Statistical Review of World Energy, Our World in Data

Canada has one of the lowest emission intensities among G20 countries (Exhibit 2) and is well-positioned to attract data centres due to its cleaner electricity mix, with a significant hydroelectric power base that accounts for nearly 60% of national electricity supply (Table 1). In fact, in 2023 Microsoft announced a US$500 million investment to expand its cloud computing and AI infrastructure in Québec, incentivized by the province’s abundant hydropower resources and favourable government policies.

Due to hydropower’s main role, Canada’s electricity system has been among the cleanest in the world for decades, and currently, wind energy and solar PV are the fastest growing sources of electricity in the country. By end of 2023, Canada had close to 22 GW of wind, solar, and energy storage installed capacity, which grewopens in new window 11% compared to 2022, driven mostly by Alberta.

Table 2

Canada is one of the few advanced economies that has not significantly increased its share of electricity produced by low-carbon sources since the turn of the century. Between 2000 and 2023, Canada increased its share of low-carbon electricity production from 73% to 81%, a change of 8 percentage points. In comparison, in the same period the UK more than doubled its share from 25% to 62%, and Spain and Australia increased their shares by 27 percentage points (now sitting at 71% and 35%, respectively). Countries that have increased their share of low-carbon electricity less than Canada include Sweden (+2), France (+2), South Korea (-1), and Japan (-9).

Source: Ember

However, Canada is one of the few advanced economies that has not significantly increased its share of electricity produced by renewable power since the turn of the century (Table 2). That may change with the proposed federal Clean Electricity Regulations, which aims to achieve a net-zero electrical grid by 2035.

In the near-term, Alberta and Saskatchewan are projected to lead in renewable energy growth with new wind and solar capacity. Alberta is experiencing a renewable power boom, driven by developers securing long-term power purchase agreements (PPAs) with tech companies including Meta, Microsoft, and Amazon. Having said that, as the only province with a deregulated electricity market, it is not guaranteed that all PPAs would be combined with clean electricity supply. The province has an ample supply of gas-powered electricity which may be a tempting and reliable source of energy for several customers.

South of the border, the U.S.’ electricity mix remains heavily reliant on fossil fuels, burdened by the politicization of climate action. IRA-fueled investments show potential for strong growth in clean energy technology and swift expansion of renewable energy to better meet future data centre electricity demand. However, based on the projected electricity demand growth from AI, this expansion might not be enough to meet demand. In Canada, the electricity mix is relatively clean albeit flat compared to other nations since the last two decades, and the tech sector has shown interest in PPAs in a few provinces with favourable conditions. Nevertheless, further policy initiatives from provincial and federal governments will be needed to incentivize private capital to build more resilient infrastructure that supports reliable and sustainable energy supply.

AI challenges Big Tech’s climate ambitions

Capital Group analysts predict that total AI investment by Microsoft, Meta, Google and Amazon will reach US$189 billion in 2024, accounting for more than a fifth of total capex among S&P 500 companies. Noteworthy examples of investments include Microsoft’s and OpenAI’s US$100 billion ‘Stargate’ supercomputer data centre and Google’s reported capital expenditure of US$12 billion in this year’s first quarter, up 90% year-over-year, driven by investments in technical infrastructure like servers and data centres to capture AI demand.

Exhibit 3

Tech giant’s electricity consumption, such as Microsoft, Google, and Meta, is growing rapidly and rivaling that of small countries. In 2022, Google and Microsoft consumed more electricity than Slovenia, which was almost 15 terawatt-hours. Meta is on track to surpass Slovenia’s electricity consumption this year.

Below, we compare Microsoft and Google’s climate commitments, how their emissions increased due to AI and how they are pivoting their climate strategies and investing in alternative energy solutions, some of which include big bets on cracking the code for nuclear fusion.

Looking at the comparison above, we can see both companies share the followings strategies:

• Investments and partnerships with low-carbon electricity suppliers via PPAs to accelerate market demand, scale innovation, and secure future sources of energy

• Collaboration with nuclear energy suppliers as a source of low-carbon and reliable base load electricity for data centres

• Off-grid experiments with innovative low-carbon technologies, such as SMRs, for direct electricity supply to data centres

• Minimizing energy use in AI operation through hardware efficiency, improving electricity utilization, and procuring next-generation technology

Directly powering data centres with SMRs would take them off the grid, but building, testing and receiving regulatory approvals could take up to 10 years. Currently, utility companies backfill required grid energy with fossil fuels if other sources cannot meet the demand.

As we shared in Figure 1, coal is around 20% of the U.S.’ energy mix. The U.S. is the fourth largest coal producer globally. It is now estimated that about 54 gigawatts of coal power assets, close to 4% of the U.S.’ total electricity capacity, are expected to be retired by 2030 when previous estimates were 40% higher. This is directly attributed to the rise in data centre demand to power AI, cloud storage and services, and cryptocurrency mining. While some U.S. operators are delaying plans to convert coal-fired plants to gas, others have scrapped targets to phase out coal altogether. Currently, the IEA’s net zero pathway shows that global coal production needs to be phased out by 2040 to be able to reach a 1.5-degree scenario by 2050 (Exhibit 4).

Exhibit 4

Currently, the IEA’s net zero pathway shows that global coal production needs to be phased out by 2040 in order to be able to reach a 1.5-degree scenario by 2050.

Lastly, concerns have been emerging about the visibility and accuracy of Tech companies’ reported carbon footprints. A recent analysis by Bloomberg Green notes that some Tech firms buy what are called ‘unbundled renewable energy certificates’ or unbundled RECs, which are carbon credits that offset fossil fuel emissions and can count positively as part of carbon footprint calculations. Although current global carbon account rules under the GHG Protocol allow for use of such credits, academics have raised concerns that these rules require updating as offsets do not necessarily represent real-world emission reductions in the atmosphere. While Google (Alphabet) phased out their use of unbundled RECs a few years ago, taking unbundled RECs out of the carbon footprint calculations could, for other Tech companies, at a minimum triple the number of emissions currently reported – not counting the added footprint of ramping up AI usage.

Portfolio manager’s take:

“The hyperscalers are definitely thinking about power access seriously and ways it can be acquired in a sustainable and continuous way. Companies in the electrical value chain give credit to the technology industry for backstopping renewable generation investment with their commitment to clean power. At present, there is an inherent disconnect between the intermittency of renewable power (such as wind and solar) and data centre imperative for uninterruptable power. Bridging solutions include connection to baseload capacity, higher back-up generation capacity, power storage, greater grid interconnection, bi-directional grids, distributed power, energy management and new energy sources. These solutions vary in emissions intensity, technology readiness, commercial readiness, cost and scale and all are advancing simultaneously. The anticipated speed of AI build out and the magnitude of electricity requirement may result in some emissions increase in the short term despite lower carbon intensity in the energy mix. Longer term with technological advancement, lower carbon transition solutions should be displaced with.”

Key considerations for ESG investors

The Information Technology sector has played a large role in ESG funds globally, partially due to its relatively low carbon footprint. According to an assessment done by MSCI, it was the largest sector allocation across the top 20 largest ESG equity funds, with Alphabet (Google) and Microsoft being in the top 5 most commonly held stock. Several of these stocks had an average weighting of 5%. The sector also seems to be the highest weighted out of 11 GICS sectors on various MSCI ESG Indexes, such as MSCI World ESG Leaders and MSCI ACWI ESG Leaders, with over 25% allocation.

While these large tech companies continue to commit to climate goals and drive renewable energy investments (notably, as opposed to other heavy data centre users such as cryptocurrency companies where climate does not seem to be a consideration), a change in energy mix usage and emission footprint profiles might affect how the sector will be viewed from an ESG lens going forward.

Key questions and considerations, especially for investors with their own net zero goals, include:

• Is the potential increase of portfolio emissions today, warranted for the promise of AI-enabled reductions and efficiencies in the future, which in certain cases means trusting in technologies that are not currently viable yet (such as nuclear fusion)?

• How can investors ensure that AI technology companies are providing accurate carbon footprint data and have decarbonization strategies in place that achieve real-world emissions reductions, to meet net zero goals?

• What kind of purchase power agreements (PPAs) do tech companies have in place for their data centres and what is the energy mix, including the future potential energy mix of grids in those jurisdictions?

• Certain ESG strategies employ nuclear exclusions; given the tech sector’s increased investments in nuclear R&D and solutions, how could this impact certain tech companies’ ESG assessments?

• Are there other sectors with low-carbon footprints that could overtake the Information Technology sector from a low-emissions standpoint over time, such as healthcare or financials (which tend to be the second and third weighted sectors in ESG indices)?

It is clear that the rise in AI solutions and the surge in data centre-driven energy demand have accelerated real-world decarbonization challenges, narrowing the window to meeting Paris Agreement 2050 goals. We expect that there will be a larger focus on the Information Technology sector’s rising carbon footprint and decarbonization hurdles in the coming future than there has been previously, including from investors.

However, this paradigm shift in energy demand is also a generational opportunity for governments, industry, and communities to collaborate and invest in low-carbon energy solutions. The energy challenges of the AI boom could be an accelerator for sustainable solutions in the future. For this, we will need a coordinated approach and a long-term commitment to a low-carbon energy transition from relevant stakeholders to help navigate the complexities of the energy transition and meet global climate goals.