First world (electricity) problems

AI could increase global carbon emissions by a staggering 80%, a startling report by Greenpeace and Friends of the Earth claims. If correct, this would most certainly derail the faltering efforts to limit global warming to  2ºC of the pre-industrial average temperature.

While this prospect is alarming, it is not the only matter surrounding AI and energy that is exercising policymakers. Western nations grappling with the “great stagnation" — economies characterised by persistent low productivity and growth — have been looking to AI as the potential solution. The International Monetary Fund agrees and predicts significant productivity gains from AI, especially in developed countries. Yet Andy Jassy, the chief executive of Amazon, the world's largest provider of data centres, is warning that there is “not enough energy right now" to run new generative AI services.

Training and running the top-end AI models require an eye-watering amount of number-crunching, or compute. Because of that, this almost exclusively happens in data centres dense with specialist microchips. So intertwined has the concept of compute become with power consumption that when engineers at Google want computing power, they ask for it in terms of megawatts. Data centre electricity consumption is projected to rise from 4% of total US demand to 6% by 2026, with generative AI expected to account for a significant portion of this growth. Despite advancements in algorithmic and chip efficiency, the International Energy Agency (IEA) forecasts an “exponential" expansion of AI energy consumption, increasing at least tenfold by 2026 compared to 2023 levels.

One would think that the rapid expansion of data centres and the crypto craze over the past decade would have prepared the US for this energy crunch. But that data centre expansion occurred in the context of a slow-growing economy and advancements in energy efficiency. As a result, US utilities' long-term forecasts predicted a modest 0.5% annual growth in electricity demand.

The rapid advancement and integration of generative AI into applications caught nearly everyone off guard. But this hasn't been the only energy surprise; a broader economy that's flickering into life is the other. The consequences of this double forecasting failure are particularly significant for energy utilities. In 2023, annual peak demand grew at nearly twice the forecasted rate, reaching at least 0.9%, driven by data centres but also by industrial facilities and “other recent investments". According to a Grid Strategies report, even this 0.9% growth in demand is probably less than we can expect in future.

Yet the IEA forecasts that more than enough new energy generation will come online in the US and elsewhere. Most of this involves an unprecedented boom in renewable energy from wind but particularly solar, making the alarming Greenpeace emissions projections an unlikely prospect. To be sure, the IEA fears this clean energy surge will not be sufficient, and carbon-based energy will have to be used at higher levels than previously thought. Microsoft's 2023 carbon emissions report was revealing in this regard. While direct energy-related emissions decreased by 6.3% compared to 2020, supply chain emissions, which include its cloud-based services (second only to Amazon in scale and hosting ChatGPT), surged by a staggering 30.9%. This led to an overall emissions increase of 29.1%.

Yet despite what the IEA says about a surge in renewable energy, a shortage of electricity is real too: unexpected curbs on new data centres have already been instituted in the most popular locations for building them. Unlike the internet, place matters for data centres. They cluster near internet network hubs to save on networking costs. More than 35% of hyperscale data centres worldwide are concentrated in northern Virginia and Goldman Sachs estimates that data centres boosted Virginia power consumption by 2.2 gigawatts in 2023. That's the equivalent to the output of two nuclear reactors. But now Virginia faces a moratorium on new data centres. Here the problem is not an absence of new power generation but getting energy to where it's needed. Existing energy infrastructure — free transmission lines and interconnection points, as well as local distribution networks — has reached its capacity.

The timeline for building this infrastructure is far more daunting than for bringing new power generation online. Part of the problem is that because the local networks are privatised, there are copious amounts of regulation, with rules to ensure interconnection and competition. Regulations to ensure environmental protection, pervasive nimbyism and outdated technical review processes designed for a bygone era are also holding projects back.

Yet among intellectuals on the left and right, mindful of geopolitical competition, there's a new realisation that things take too long to get done — especially compared to China. On the left, Ezra Klein has been banging the drum of how regulation has hampered the state's abilities to build. And on the right, Tyler Cowen has been trumpeting the need for a new kind of right-wing politics which admits that the state is crucial for development. What he calls “state capacity" libertarianism. So in the US economy as well as the realm of ideas, the times, they are a-changin'.

♦ VWB ♦

BE PART OF THE CONVERSATION: Go to the bottom of this page to share your opinion. We look forward to hearing from you.