The AI Hype: Navigating the Resource Demands of a Digital Future
Audio summary HERE [18:57 mins]
The relentless march of artificial intelligence (AI) is being hailed as the foundation of the next economy, spurring massive investment and an “arms race” to build the necessary infrastructure. However, this rapid expansion, particularly the proliferation of data centres, is creating unprecedented strain on already scarce power and water resources, prompting comparisons to past speculative tech bubbles and raising serious concerns about environmental and economic stability.
The Unprecedented Demand for Power
The migration of numerous services to the digital realm, from education and healthcare to gaming and shopping, combined with the rise of AI, is driving a revolution in data consumption. AI algorithms are “much more power-intensive” than previous computing, necessitating “round-the-clock supply” and a new generation of “much more power-hungry data centres”.
Soaring Electricity Consumption: Australia, for example, is witnessing its long march towards greater power needs resume with a “vengeance”. Forecasts by the Australian Energy Market Operator (AEMO) predict a “doubling of electricity consumption” by 2050, largely driven by decarbonisation efforts. However, energy advisory firms like Rennie suggest that demand from data centres alone could be “two, three times what AEMO was forecasting”, potentially reaching 4.9 gigawatts (GW) by 2035 in an “accelerated data centre scenario,” compared to AEMO’s 1.5 GW. To put this in context, a large coal-fired power plant typically has about 1 GW of capacity. Data centre operator Interactive expects to “double the capacity — another 175 facilities” in Australia by 2030. In addition to the existing ~250 Australian Data Centres estimated to be consuming the power & water equivalent of 10 million homes.
Global Strain: Globally, data centres are projected to “double the electricity by 2030 — and that’s driven by AI”, according to the International Energy Agency (IEA). In the US, AI-related data centres alone could “consume as much as 10 per cent of power… by 2030”. Some US regions have even declared an “energy emergency” due to the massive demand. The UK is expected to see its number of data centres increase by almost a fifth, with concerns about the huge energy they will consume potentially driving up consumer bills.
Supply Challenges: Experts like Matt Rennie highlight that the demand from data centres is already “sneaking up and risking the assumptions” used for electricity forecasts. There is a concern that “not enough energy projects developed, announced, funded today that will [meet demand]”. While renewables are seen as the market winners for new capacity due to their speed and cost, there’s also a push to justify more gas, coal, and nuclear power, which are often “unneeded, uncompetitive, and likely to raise everyone’s costs and risks”.
The Looming Water Crisis
Beyond energy, data centres are “power-hungry creatures” that also require “vast amounts of water to cool their servers”. The heat generated by servers, especially for AI training and inference, necessitates efficient cooling systems, often leading to significant water loss through evaporation. The generation of electricity to power these facilities also carries a substantial embedded water footprint, particularly when sourced from fossil fuels.
Australian Concerns: In Melbourne’s west, Greater Western Water is assessing 19 data centre applications requesting nearly “20 gigalitres of water a year”, an amount equivalent to the annual water usage of “330,000 Melburnians”. This raises fears of water shortages and limitations on new housing. Similarly, Sydney Water is preparing for data centres to use the equivalent of “25 per cent of the city’s yearly drinking water supply by 2035”, a significant jump from less than 1 per cent currently.
UK Water Stress: The UK faces a projected daily water deficit of “nearly 5 billion litres by 2050”, a challenge exacerbated by climate change and population growth. Data centres intensify this vulnerability, especially as most existing facilities rely on “evaporative cooling systems that consume potable water and lose up to 80% of it through evaporation”. Some regions, like Oxfordshire and Cambridgeshire, are already designated “seriously water stressed” areas where new data centres could worsen local scarcity.
Lack of Transparency: A critical issue is the “lack of reliable data on the total water use of data centres” in the UK, with only two-fifths of operators tracking their consumption. This opacity makes effective planning and regulation difficult. Globally, a single ChatGPT user session can consume “up to 500 milliliters of water”, and global AI water demand could reach the equivalent of “more than half the annual water usage of the U.K.” by 2027.
The Shadow of Hype and Uncertainty
The current AI boom bears striking resemblances to past periods of “irrational exuberance”. Experts warn of a “wildly uncertain” future, shaped by unproven concepts, disputed performance, and unpredictable adoption. The technology’s rapid evolution, with computing energy efficiency roughly quadrupling each year, means that “depreciation and amortization schedules — decades in the world of atoms — may shrivel in the radically faster-moving world of bits”.
Overbuilding and Stranded Assets: The speculative surge in data centre investment risks a “12-figure overbuild”, potentially creating a “new electricity bubble with hundreds of billions of dollars of overbuilding”. If AI demand plateaus or if more efficient models emerge, many of the massive investments in training-focused data centres could become “stranded assets”.
Unknowable Demand: “Nobody understands the fundamental long-term demand for AI’s services”. The market for AI’s profitable use cases remains speculative, with concerns about whether customers will truly pay for AI features or if the benefits outweigh the costs.
Trust and Quality: A deeper concern is that AI tools “fabricate and then deny they did” or produce “buggy code,” leading to a lack of trust crucial for business adoption. There are fears that recursive training on errors could “corrupt the quality of AI models and of the entire scientific/technological enterprise”, diluting sound literature with lower-quality machine inferences and summaries.
Towards a Sustainable Digital Future
Addressing these challenges requires a multifaceted approach, combining technological innovation, strategic planning, and robust policy.
Energy Solutions:
Decarbonisation and Renewables: A rapid decarbonisation of the power sector is crucial, especially given that electricity consumption during manufacturing and operation of electronic devices is the primary contributor to climate impacts. Companies like Interactive are determined to rely totally on clean energy. Renewables, particularly solar, wind, and storage, are identified as the fastest and cheapest options for new capacity.
24/7 Carbon-Free Electricity (CFE): This approach involves matching electricity demand every hour with carbon-free generation 24/7, where and when consumption occurs, pushing for innovation in battery storage and flexible technologies.
Demand Flexibility: Implementing flexible and intelligent demand response can optimize when data centres are used, deferring compute-heavy tasks to times when power is cleaner or cheaper, and transforming data centres from liabilities into grid assets.
Efficiency Gains: Advances in cooling efficiency have dramatically reduced data centres’ energy consumption for cooling. Further efficiency gains in software and algorithmic progress are expected.
Water Solutions:
Efficient Cooling Technologies: Mandating and incentivising high-efficiency cooling technologies such as liquid immersion, direct-to-chip, and closed-loop systems can significantly reduce water usage.
Non-Potable Water Sources: Requiring data centres to use recycled or non-drinking water, such as industrial wastewater, greywater, or harvested rainwater, can reduce the strain on municipal drinking supplies.
Strategic Siting: Restricting new data centres in “seriously water stressed areas” and requiring site-specific water risk assessments are vital.
AI for Water Management: AI itself can be a tool for water management, used for pollution detection, water quality forecasting, and optimising infrastructure.
Policy and Transparency:
Mandatory Reporting: Introducing mandatory, location-specific reporting of water, energy, and carbon use for data centres is essential for accountability and informed planning.
Integrated Planning: AI infrastructure development must be integrated into national water and energy planning frameworks to accurately account for future demands and avoid deficits.
Government Leadership: Governments, as major procurers of digital services, can lead by example by embedding water efficiency targets in their own ICT policies and mandating disclosure from suppliers.
Circularity: Extending the lifetime of electronic devices plays a crucial role in reducing the embodied impacts (from raw materials extraction, manufacturing, etc.) and the use of mineral and metal resources.
Conclusion
The “AI hype” signals a transformative era, but the scale of investment and the speed of development are outpacing prudent planning for resource management. The escalating demands for power and water from AI data centres pose significant environmental and economic risks, especially in already stressed regions. To avoid an overbuilt, unsustainable future, disciplined foresight, accurate risk pricing, and market-led investment in proven, least-cost solutions are essential. This means prioritising renewable energy, water-efficient technologies, transparent reporting, and integrated resource planning to ensure that AI’s advancements contribute to a resilient and sustainable future, rather than exacerbating existing challenges.
References
[ABC News] AI data centres need round-the-clock energy and could be more power-hungry than we think
[ABC News] AI is driving data centre growth — and it’s bringing environmental challenges
[ABC News] Calls for guidelines after Greater Western Water documents reveal potential data centre water usage
[ABC News] Data centres are vital for the future and AI but their environmental footprint can be a problem
[BBC News] Data centres to be expanded across UK as concerns mount
[Financial Times] Who pays for the $3tn AI building boom
[Financial Times] Inside the relentless race for AI capacity
[Podcast Computer Says Maybe] UK Groups Sue To Block Data Center Expansion
[Nature Communications] The environmental sustainability of digital content consumption
[Climate Group] What is 24/7 carbon-free electricity (CFE)?
We Found the Hidden Cost of Data Centers. It's in Your Electric Bill
[High Yield YT Channel] How AI Datacenters Eat the World