I basically agree, but with the caveat that we need to both work on efficiency and increase renewable energy. We won't be able to bring enough renewable energy on line fast enough to support a thousandfold or ten-thousandfold increase in energy consumption by AI over the next 3-5 years. We would have to improve efficiency over that time period so that if we need 1000x or 10,000x the computational power, we only need to use 10x the energy. Also, to some extent there is a short-run tension between using new renewable energy generation (and dispatchable storage, like grid-scale or distributed battery systems) to support growth of consumption and to retire older, dirtier power generation. In the long-run, we need a carbon-free power system AND we need much more power than we currently have, but in the short-run we have both important climate change pressures to decarbonize rapidly and developmental pressures to continue increasing (and making more equitable) energy consumption.