It is no secret that the rapid advancement of artificial intelligence is transforming the world in which we live and how we create. It is giving us tools that allow us to explore new skills. Coding is no longer the realm of a select few; mundane or routine tasks can be efficiently executed, and complex subjects or theories can be broken down and understood by all.
But this comes at a cost. As AI technologies become more sophisticated, they demand vast computational resources. Data centres are being built at a furious rate, and they have a voracious appetite for the energy that keeps the lights on and spins up the chips and cooling systems.
Often called the Fourth Industrial Revolution, the speed of current breakthroughs has no historical precedent. When compared with previous industrial revolutions, the Fourth is evolving at an exponential rather than linear pace and it will have societal implications at home and security issues abroad. Emerging technology in fields such as artificial intelligence, robotics, the Internet of Things, autonomous vehicles, materials science, energy storage, and quantum computing are coming at an extraordinary rate.
In the United States, forecasts suggest the AI sector alone could require at least another 50 gigawatts (GW) of additional generation capacity by 2028 to sustain the current growth rate and maintain their technological leadership. The challenges are huge. Federal legislation, state zoning and permits, and social considerations all have the potential – if not likelihood – to cause delays.
To put this incremental generation into perspective, 50 GW (gigawatts) is approximately 4% of total US power generation in 2024, which is estimated to have been 1,300 GW. The demand side of the equation paints a similar picture, having accelerated sharply in 2024 to mark a stark departure from the modest annual increases of the last 20 years, during which electricity demand was stagnant.
Escalating demand is not merely a technical issue but a national imperative, intertwining economic prosperity, security, and societal progress. Yet, it brings challenges: regulatory bottlenecks, infrastructure delays, and supply chain vulnerabilities threaten to progress.
However, there is a precedent. China, added over 400 GW of new generation capacity in 2024, a 21% increase over 2023 and perhaps eight times that of the US.
What does the rise in electricity demand mean?
Currently it is estimated that the US accounts for nearly 50% of global data centre power use, which is accelerating at approximately 3 to 5% annually due to AI’s integration into everyday life. Globally, data centre consumption is projected to double to 920 TWh by 2030, representing approximately 10% of global electricity demand. Recently, Alphabet announced that it is now processing 980 trillion tokens monthly, more than double the level only three months before.
As models require larger clusters of semiconductors for training and deployment, data centres – once focused on cloud storage – are now hubs for energy-intensive AI compute. Estimates vary; however, it is anticipated that training a single-frontier AI model might necessitate between two and five GW of capacity by 2028 to 2030. With multiple US companies competing at the AI frontier, this could total 20 to 25 GW for training alone.
Inference – the use of already developed AI models for specific tasks – is less energy intensive than that required for training. But the compute and energy needs of inference compound quickly the more widely AI is deployed into applications like autonomous vehicles, medical diagnostics, and cybersecurity.
Building a sustainable grid to address surging demand
Solar continues to be rapidly deployed, contributing almost 80% of the incremental amount of power generation while battery and wind continue to grow. However, these sources are less reliable than traditional forms such as gas. In its last quarter, GE Vernova (GEV) announced strong growth in gas turbines, receiving contracts for nine GW of new equipment agreements, with seven GW in slot reservations and two GW directly converted into firm orders, alongside five GW of equipment shipped.
As if to confirm the data centre-driven demand, GEV also announced that it had received orders for almost US$500 million over six months, comparing favourably to a total of US$600 million for the prior 12 months.
Nuclear advancements, particularly in small modular reactors (SMRs), promise clean, scalable energy. Progress in manufacturing TRISO (TRi-structural ISOtropic) – a safer fuel source for reactors – may allow a clearer path to approval. This suggests the interest in nuclear as a reliable technology may eventually move forward at scale. China again has the march here, with Gen4 reactors being readily constructed and effectively leaving the rest of the world in their wake.
In addition to generation, grid enhancements are critical, and in the US transmission projects are accelerating, with 2,400 kilometres of high-voltage lines planned for 2025 to 2030, compared to 90 kilometres added in 2023. As with power generation, traditional zoning rules have caused delay; however, there appears to be a will to address these, removing red tape from the process. Other challenges need to be overcome, notably the requirement to train almost 250,000 skilled workers by 2030.
Barriers to growth in electricity supply
The significant obstacles that may hinder a timely expansion need to be overcome, but it will not be a smooth path. Regulatory hurdles are paramount: building data centres, power plants, and transmission lines involves navigating federal, state, and local approvals, often extending timelines by three to five years. Environmental concerns and reviews under the National Environmental Policy Act (NEPA) can overlap with air quality, endangered species, and water regulations, adding to delays or cancellation.
Transmission and interconnection issues exacerbate the problem. High-voltage line additions have slowed from an average of 2,700 kilometres per annum 10 years ago. Although this activity is expected to reaccelerate, they are still subject to a lengthy approval process, while grid interconnections also add to complexity, averaging four to six years to complete. While a data centre build may take only 18 to 24 months to construct, the preconstruction approvals take a significant amount of time and need to be streamlined while ensuring diversity and other social considerations are addressed.
Given nascent power demand growth over the last 20 years, the equipment manufacturers have not invested heavily in spare capacity and are unable to scale production significantly to meet the explosive order growth they are seeing in the near term. For example, GEV has been taking slot reservations for 2030 delivery and has a total backlog of almost US$129 billion. Of this, the equipment backlog rose US$5 billion over three months to US$50 billion. Supply chain vulnerabilities add complexity.
Globally, transformers and circuit breakers are in short supply and have lead times approaching three years, a significant problem given the US imports almost 70% of these from abroad. In addition, there remains a significant and growing shortage of trained electricians and engineers.
Will technology save technology?
The obvious question is whether we can innovate to resolve rising energy demand – and indeed, innovative technologies are emerging that may have the potential to curb data centre energy use. One such technology, from Coherent, may enhance efficiency without sacrificing performance. Optical transceivers, such as Coherent’s 100G ZR modules, enable high-speed data transmission over 100 kilometres with 20 to 30% less power than traditional systems, reducing energy loss in fibre-optic networks. Liquid cooling systems are increasingly replacing that of air, reducing cooling energy by up to 30% in dense cluster environments, and advanced chip designs, incorporating photonics and AI-specific accelerators, are increasingly more efficient for any given task.
In addition to hardware improvements, software will lead network optimisation, deploying machine learning algorithms to dynamically allocate resources, powering down idle servers, and predicting workloads, while Edge computing will allow processing in local nodes, reducing data centre compute. While still relatively early in the improvement, these technologies should lower data centre energy consumption significantly while simultaneously extending chip and equipment lifespans.
Navigating the AI-driven energy demand
Whether it be your preferred chatbot or self-driving cars or virtual assistants, progress in the Fourth Industrial Revolution has been impressive. The driving force behind this has been the exponential increases in computing power and by the availability of vast amounts of data harvested from the open internet.
This success has raised social and ethical questions, but the genie is out of the bottle and will not return to it. Progress is often disruptive, and the non-linear nature of this industrial revolution will create change on a scale we have not seen before. The change has put pressure on our electrical system, the same infrastructure that powers our homes, hospitals, and schools. As time passes, human ingenuity will rise to the challenge and everyday obstacles such as permitting delays and supply constraints will be overcome.
With geopolitical tensions high, AI has been singled out as the frontier where second place is not good enough. Today’s technology is amazing, but success will come from the ability to fully harness what was popularised 146 years ago by Thomas Edison, when he demonstrated the electric light bulb.
Tim Chesterfield is the long-time CIO of the Perpetual Guardian Group and the founding CIO and Director of its investment management business, PG Investments. With $2.6 billion in funds under management and $8 billion in total assets under management, Perpetual Guardian Group is a leading financial services provider to New Zealanders.
Perpetual Guardian Group recently acquired boutique fund manager Castle Point, which now forms part of the Group’s investment management suite of businesses.
Disclaimer
Information provided in this publication is not personalised and does not take into account the particular financial situation, needs or goals of any person. Professional investment advice should be taken before making an investment. The information provided in this article is not a recommendation to buy, sell, or hold any of the companies mentioned. PG Investments is not responsible for, and expressly disclaims all liability for, damages of any kind arising out of use, reference to, or reliance on any information contained within this article, and no guarantee is given that the information provided in this article is correct, complete, and up to date.
This article was originally published by the NBR. You can read the original piece here.
