Executive Summary:
- The AI environmental impact is broader than carbon alone. Recent investor and policy work increasingly frames data-centre growth around water use, affordability, grid reliability, and local community impacts, not just emissions. Ceres says roughly one-third of U.S. data-centre construction is concentrated in areas already facing serious water-availability challenges, and that one facility can use up to 550,000 gallons of water a day for cooling.1
- This update revisits the observations in the 2024 Russell blog on AI & the energy transition2, which still hold, but the impacts are now clearer and better quantified across electricity demand, local grid strain, water use and buildout bottlenecks. The IEA estimates data centres used around 415 TWh of electricity in 2024 and projects around 945 TWh by 2030 in its base case - roughly a doubling in six years, taking the sector to a level of power demand slightly larger than Japan’s entire electricity consumption today. In parallel, Ceres’ Phoenix case study suggests that, if planned facilities come online over roughly the next six years, annual water use linked to data-centre electricity consumption could rise from about 2.9 billion gallons to more than 14.5 billion gallons.3
- Local impacts matter more than the global averages suggest. The IEA stresses that data centres are geographically concentrated, so even if their global electricity share still looks manageable, the local burden on grids, infrastructure and communities can be much more acute. Globally, data-centre electricity demand is growing more than 4x faster than electricity demand from all other sectors combined.4
- Physical bottlenecks are now part of the story. This is not just about how much electricity AI needs, but whether the required grid and electrical equipment can be delivered. The IEA says grid constraints could delay around 20% of planned global data-centre capacity by 2030. Recent Bloomberg reporting adds that more than half of U.S. data centres planned for this year are expected to be delayed, with shortages of transformers, switchgear and batteries a major factor.3,5
- Australia-specific developments now make the issue much more immediate and locally relevant. AEMO6 has begun forecasting data centres as a standalone demand segment, estimating around 4 TWh of NEM electricity use in FY2025 and around 12 TWh by 2029-30 under Step Change. The Australian Government’s March 2026 expectations say new data centres should not put upward pressure on energy prices, should secure new and additional clean generation and/or storage, and should provide transparent reporting on water usage and efficiency. The AEMC7 has also proposed new technical standards because large data-centre loads can affect grid stability.8,9
1. Update to Russell’s 2024 blog, “Is AI slowing down the energy transition?”
- The biggest change is that the evidence base is much stronger. The IEA now estimates data centres consumed about 415 TWh of electricity in 2024, equal to about 1.5% of global electricity demand, and projects demand of around 945 TWh by 2030 in its base case. It also says data-centre electricity demand is growing more than four times faster than electricity demand from all other sectors combined.10
- The IEA’s main nuance is still critical for investors: the global share can look manageable while the local impacts are much sharper, because data centres are geographically concentrated. In the IEA’s executive summary, the United States is expected to account for by far the largest share of the demand increase, and local grid stresses become more important than global averages.
- The U.S. evidence base is now firmer as well. Berkeley Lab says U.S. data centres consumed about 176 TWh in 2023, or 4.4% of U.S. electricity demand, up from 58 TWh in 2014, and could rise to 325-580 TWh by 2028, or 6.7%-12% of U.S. electricity demand. It also says data-centre power demand more than doubled between 2017 and 2023, largely because of AI servers and cooling requirements.11
- The supply-side picture remains mixed rather than clean. The IEA says electricity generation serving data centres rises from 460 TWh in 2024 to over 1,000 TWh in 2030 in the base case. Renewables meet nearly half of the additional demand, but natural gas and coal together still meet over 40% of incremental demand. In the U.S., natural gas already supplies over 40% of data-centre electricity and is the largest source of incremental supply through 2030.12
2. The issue is broader than carbon
- The environmental question is now wider than “AI raises emissions.” The United Nations Environment Programme (UNEP13), says the boom in AI and data centres is increasing demand on electricity systems and can also affect water resources depending on design, cooling technology and location. It also argues that if expansion is unmanaged, it can push up local electricity demand, raise power prices for households and businesses, strain power and water systems, and increase emissions.14
- Ceres gives the clearest investor shorthand, framing the issue around grid reliability, affordability, water use and local community impacts, and says data-centre growth needs to be fast, fair, affordable, efficient and clean. It also says roughly one-third of U.S. data-centre construction is concentrated in areas already facing serious water-availability challenges, one facility can use up to 550,000 gallons of water per day for cooling, and data centres account for about 4% of U.S. electricity demand, with demand expected to triple over the next decade. 15
- The water story is more severe than many 2024 discussions assumed. Ceres’ Drained by Data report says indirect water use from electricity generation can matter even more than direct cooling water. Using Phoenix16 as an illustrative example, Ceres estimates annual water use in the coming years, tied to data-centre electricity consumption could rise 400%, from about 2.9 billion gallons to more than 14.5 billion gallons, while water use tied to cooling could rise 870%, from 385 million gallons a year to more than 3.7 billion gallons a year. It also says data-centre growth could raise water stress in already strained basins by up to 17% annually.17
- Investor organisations are starting to grapple with the frameworks and mechanisms needed to assess these issues. IIGCC says the scale of AI-driven data-centre growth, the prominence of technology companies inside “sustainable” funds, and the growing private-markets allocation to real estate and infrastructure make these issues increasingly material for investors. It argues for an integrated lens spanning decarbonisation, nature and resilience.18
- The climate balance is still nuanced. The IEA says data centres account for about 180 Mt CO2 of indirect emissions today, amounting to approximately 0.5% of global combustion emissions, but it is among the few sectors whose direct and indirect emissions are projected to rise into 2030. At the same time, the IEA says that widespread adoption of existing AI applications in end-use sectors could reduce emissions by around 1,400 Mt CO2 in 2035.
Implication: The emissions evidence still supports a mixed conclusion. AI is not automatically climate-positive or climate-negative; its impact depends on how and where the underlying infrastructure is built and powered.19
3. Physical buildout bottlenecks are now part of the story
- Physical bottlenecks are already slowing data-centre buildout and complicating wider electrification. Bloomberg reports that more than half of U.S. data centres planned for this year are expected to be delayed, and that shortages of transformers, switchgear and batteries are a major reason. It also says those same components are needed for broader electrification, including EVs and heat pumps.20
- The article is especially useful because it shows how small components can become system-wide constraints. Bloomberg says electrical infrastructure adds up to less than 10% of the total cost of a data centre, but the whole project can be held up if that part of the supply chain slips. It also says U.S. manufacturing capacity cannot keep up, forcing builders to rely on imports.18
- Lead times have become meaningfully worse. Bloomberg reports that high-power transformers that once took around 24-30 months now have lead times of up to five years, and that some developers are refurbishing old transformers as a stopgap.18
- This is consistent with the IEA’s broader system warning. The IEA says grid constraints could delay around 20% of global planned data-centre capacity by 2030, and that wait times for critical grid components such as transformers and cables have doubled in the past three years. It also warns that, unless these risks are addressed, data-centre load growth could create trade-offs with other goals such as electrification, manufacturing growth and affordability.21
Implication: AI growth is no longer just an electricity-demand story; it is also a grid, equipment and execution-risk story. That matters for environmental outcomes because the same bottlenecks can delay both data centres and the wider infrastructure needed for the energy transition.
4. Australia-specific AI/Data Centre Developments
- Australia-based industry actions and data are now available to strengthen the analysis. The Australian Energy Market Operator (AEMO) now forecasts data centres as a standalone demand segment, rather than grouping them inside broader commercial load. It estimates data centres consumed around 4 TWh in FY2025, about 2.2% of NEM22 grid demand, and could reach around 12 TWh by 2029-30 under Step Change, or about 6% of NEM grid-supplied electricity. Over the longer term it sees around 34 TWh by 2049-50.23
- The Commonwealth (national government) has now published an explicit expectations framework for the data centre industry. The Department of Industry says these expectations are intended to form the basis of data-centre developers’ social licence to operate in Australia, and that energy-intensive projects not closely aligned with them will not be prioritised in Commonwealth regulatory assessments.24
- Those expectations are highly relevant to ESG analysis. The Commonwealth says new data centres and AI infrastructure should not place upward pressure on energy prices, should make a positive contribution to Australia’s energy transition, should secure new and additional clean generation and/or storage, should cover their share of transmission and distribution infrastructure costs, and should improve grid stability through demand flexibility, peak-load management and data sharing.
- The water and community side is equally explicit. The same Commonwealth document says data centres should use innovative, efficient and sustainable solutions to minimise water use, engage early with water utilities, communities and First Nations peoples, use appropriate local water sources, and provide ongoing transparent reporting on water usage and efficiency.
- NSW is also treating this as a live planning issue. NSW announced on 27 March 2026 that 15 data-centre projects worth $51.9 billion would progress through the Investment Delivery Authority, while also releasing a consultation paper to steer sustainable development. The government said it wanted to capture the economic benefits while managing the challenges associated with data-centre operations, especially around power and water.25
- There is also now a technical grid-stability angle in Australia. The Australian Energy Market Commission (AEMC)’s March 2026 draft rule says many data centres use inverter-based technology similar to wind, solar and batteries, and warns that if many such facilities disconnect simultaneously during a grid disturbance, this could increase the risk of cascading outages or instability.26
Implication: In Australia, this is already a policy, planning, water, grid and social-license issue, not just a macro theme.
5. Impacts from the Middle East Conflict
- The main channel from the Middle East conflict to data centre economics is energy-market disruption. The IEA says the Strait of Hormuz carried about 20 million barrels per day of oil and oil products in 2025, or around 25% of global seaborne oil trade, and that LNG from Qatar and the UAE passing through the Strait represented about 19% of global LNG trade. It warns that disruption would have major oil-market consequences and would fuel strong volatility in gas prices.27
- Australia’s official position is that current supply is secure, but risks rise if the conflict persists. The Prime Minister’s March 2026 National Cabinet statement says Australia is well prepared and current supply is secure, but that the longer the conflict continues, the more significant the likely impact on global supply chains, fuel prices and the wider economy.28
- The main implications for data centre economics are indirect. The most relevant channels are likely to be higher fuel and backup-power costs, higher logistics and construction costs, and greater pressure around energy affordability and resilience. That is an inference from the energy-market shock rather than a specific data centre finding.
GLOSSARY
Additionality
Whether a company’s renewable-energy procurement leads to new clean generation or storage being built, rather than simply claiming existing supply.
AI (Artificial Intelligence)
Software systems that perform tasks such as prediction, pattern recognition, optimisation, or content generation. Large AI models often require significant computing power.
AEMC (Australian Energy Market Commission)
The body that makes and amends electricity and gas market rules in Australia.
AEMO (Australian Energy Market Operator)
The organisation that operates and plans Australia’s main electricity and gas systems and publishes demand forecasts.
Broader related exposure
Exposure to businesses that may benefit from AI or data-centre growth indirectly, such as utilities, electrical-equipment suppliers, cooling providers, or industrial property owners.
Contractual renewable claim
A renewable-energy claim based on contracts or certificates rather than the actual electricity physically supplied to a facility at a given time.
Cooling water
Water used to remove heat from servers and other equipment in a data centre.
Data centre
A facility that houses servers, storage and networking equipment used to process, store and transmit data.
Demand flexibility
The ability of a large electricity user to reduce or shift its electricity use in response to grid conditions or price signals.
Direct exposure
Exposure to companies or assets whose business is clearly tied to AI or data centres, such as cloud providers, chipmakers, data-centre operators or data-centre REITs.
Electricity demand
The amount of electricity used over a period of time.
Ex-fossil / fossil-fuel exclusion
An investment approach that screens out companies with specified involvement in fossil fuels.
Grid constraints
Limits in the power system, such as network capacity, interconnection delays or equipment shortages, that restrict new connections or raise costs.
Grid reliability
The ability of the electricity system to provide stable and continuous power without major disruptions.
Indirect water use
Water used upstream to generate the electricity consumed by a facility, rather than water used directly on site.
Interconnection
The process of connecting a large electricity user or generator to the power grid.
Local grid strain
Pressure on a particular part of the electricity network caused by concentrated new demand.
Look-through transparency / full holdings disclosure
The ability to see the underlying holdings inside a pooled fund or investment structure.
Material / materiality
Important enough to influence investment decisions, financial outcomes, or stakeholder concerns.
Mt CO2
Million tonnes of carbon dioxide.
NEM (National Electricity Market)
Australia’s main interconnected electricity market across the eastern and southern states and territories.
Passive management with exclusions
An investment approach that broadly follows an index or rules-based strategy while removing certain prohibited exposures.
Peak-load management
Reducing or shifting electricity use during periods of highest demand on the grid.
Physical power mix
The actual sources of electricity serving a facility through the grid, such as gas, coal, solar, wind or hydro.
Pooled funds
Investment vehicles in which money from multiple investors is combined and managed together.
Private markets
Investments that are not traded on public stock exchanges, such as private equity, private credit, real estate and infrastructure.
REIT (Real Estate Investment Trust)
A listed vehicle that owns or finances income-producing real estate. Some REITs specialise in data centres.
Social licence to operate
The ongoing acceptance of a project or company by communities, regulators and other stakeholders.
Stewardship
The use of engagement, voting and escalation by investors or managers to influence companies on governance, environmental or social issues.
Switchgear
Electrical equipment used to control, protect and isolate power systems.
Transformer
Electrical equipment that changes voltage levels so electricity can be transported efficiently and used safely.
Transmission and distribution infrastructure
The poles, wires, substations and other equipment that move electricity from generators to end users.
UNEP
The United Nations Environment Programme (UNEP) is the leading global authority on the environment, established in 1972 to coordinate UN environmental activities and address the "triple planetary crisis" of climate change, nature/biodiversity loss, and pollution. Headquartered in Nairobi, Kenya, it sets the global environmental agenda, promotes sustainability, and supports 193 member states.
Water stewardship
The responsible management of water use and its impacts on local communities and ecosystems.
1 https://www.ceres.org/data-centers
2 https://russellinvestments.com/uk/blog/2024/09/ai-energy-transition
3 https://www.iea.org/reports/energy-and-ai/executive-summary
4 https://www.iea.org/reports/energy-and-ai/energy-demand-from-ai
6 The Australian Energy Market Operator (AEMO) manages electricity and gas systems and markets across Australia.
7 The Australian Energy Market Commission (AEMC) is the expert statutory body responsible for making and amending the rules that govern the electricity and gas markets in Australia.
9 https://www.industry.gov.au/publications/expectations-data-centres-and-ai-infrastructure-developers
10 https://www.iea.org/reports/energy-and-ai/energy-demand-from-ai
12 https://www.iea.org/reports/energy-and-ai/energy-supply-for-ai
13 The United Nations Environment Programme (UNEP) is the leading global authority on the environment, established in 1972 to coordinate UN environmental activities and address the "triple planetary crisis" of climate change, nature/biodiversity loss, and pollution. Headquartered in Nairobi, Kenya, it sets the global environmental agenda, promotes sustainability, and supports 193 member states.
14 https://www.unep.org/technical-highlight/how-make-ai-data-centres-more-sustainable
15 https://www.ceres.org/data-centers
16 This analysis uses Phoenix, Ariz. — a water-stressed region, fast-growing metropolis, and booming data center hub - as an illustrative example to highlight the impacts of data center water use on regional water stress.
18 https://www.iigcc.org/insights/ai-data-centres-climate-and-nature-risks-for-investors
19 https://www.iea.org/reports/energy-and-ai/ai-and-climate-change
21 https://www.iea.org/reports/energy-and-ai/ai-and-energy-security
22 The National Electricity Market (NEM) is one of the world's longest interconnected power systems, spanning Australia's east coast, Tasmania, and South Australia. It operates as an energy-only market, balancing supply and demand every 5 minutes across five state-based regions to deliver electricity, with management by AEMO.
24 https://www.industry.gov.au/publications/expectations-data-centres-and-ai-infrastructure-developers
25 https://www.nsw.gov.au/ministerial-releases/data-centre-investment-sustainable-development
27 https://www.iea.org/about/oil-security-and-emergency-response/strait-of-hormuz