Generative AI's Power Bill

The energy consumption of large AI models has been discussed in academic circles for several years — a 2019 paper from the University of Massachusetts estimated that training a single large NLP model could emit as much carbon as five cars over their lifetimes. But it was not until the mainstream deployment of ChatGPT and GPT-4 in early 2023 that the energy implications of generative AI entered the mainstream investment conversation. The numbers are striking: a single ChatGPT query reportedly consumes approximately ten times the energy of a Google search. At the scale of hundreds of millions of daily queries, the aggregate energy demand is substantial.

The power implications extend well beyond the training of individual models. The inference infrastructure required to serve AI queries at scale — the data centres running 24 hours a day to respond to user requests — is where the bulk of AI energy consumption actually occurs. Training a model is a one-time (or periodic) event. Serving that model to millions of users is a continuous process that requires sustained power at significant scale. As AI becomes embedded in more applications and more workflows, the inference energy demand will grow continuously, even if training efficiency improves.

For data centre investors and operators, the energy challenge is both a constraint and an opportunity. The constraint is that power availability has become the binding limitation on data centre development in many markets — utilities are struggling to provision the grid connections that new AI data centres require, and the lead times for new power infrastructure are measured in years, not months. The opportunity is that operators who have secured long-term power agreements — particularly for renewable energy — have a competitive advantage that is difficult to replicate and that commands a premium in the market.

At Lensbridge, we have been incorporating power analysis into our digital infrastructure due diligence for several years, recognising that the energy supply chain is as important as the physical infrastructure itself. The generative AI power bill is making this analysis more important, not less — and it is creating opportunities for investors who understand the intersection of energy infrastructure and digital infrastructure in ways that generalist investors do not.