What if US load growth disappoints from here, as the AI era progresses? We have previously published forecasts for +3% pa to 2030. Others are at 5-10% pa. Hence today’s 16-page report asks what assumptions are embedded in our outlook. There are six key risks. Our downside case is as low as +1% pa load growth, which could largely be met behind the meter?
Total net generation of US electricity grew at 0.3% per year in 2004-19, which was below total US population growth of 0.37% pa over the same timeframe.
Growth then accelerated to 1% per year in 2019-24. For two years now, we have been projecting grid bottlenecks and US load growth accelerating to 3% per year, which would itself be astonishing (here, here and here).
Yet over the past two years, we have witnessed sentiment shifting. Others’ numbers have crossed our desks, projecting 5-10% pa load growth, and power plant projects equivalent to 6% pa load growth are in progress (see pages 15-16).
What if load growth is being materially over-estimated, and ultimately the numbers come in much lower, closer to 1-3% per year?
We are somewhat worried that asking this question might make us unpopular. It is remarkable, when we think about our friends and clients across the investment and corporate landscape: we all, in good faith, want to see AI improve the world, and to be called upon to provide the energy and equipment to make that happen.
But ultimately, it is important to sense-check that we are not getting carried away with unrealistic numbers. Avoiding universal optimism also allows us to focus in selectively on the best opportunities.
Hence we have disaggregated US electricity demand across 300 lines, with forecasts to 2050, in our newly expanded models, to evaluate what assumptions are explicitly and implicitly embedded in our load growth forecasts.
Even projecting 3% pa load growth to 2030, then 2% pa load growth to 2050 is making generous assumptions around data center demand and other emergent energy demand in the AI eco-system, based on our research here, here and here. Conversely, on balance, there is room for much greater efficiency gains.
The report includes modeling definitions on pages 2-3, six key factors that decision-makers should monitor on pages 4-13, a scenario for 1% pa load growth on page 14, and closing comments on the US power project pipeline on pages 15-16.