The utmost important question to ask about the AI infrastructure boom is how long will the capacity actually be relevant? Whilst many are investing based on assumptions that increased AI adoption will lead to a sustained demand for new facilities, power and cooling systems, this is not necessarily the case. As AI economics change with different model architectures, accelerator capability and inference requirements, facilities built for highly dense training workloads may now be operating in a different market where inference and cost effective compute are more important. The facility may have full power and cooling capabilities but the actual economics of the site depend on what customer requirements and costs are. This means the question to ask in the industry is how relevant will our infrastructure investments be in the future?