Not sure what this prediction is based on. Sure a further crash is possible. It could also go the other way. We might see OpenAI applying some of the optimization that DeepSeek used and coming out with something amazing.
That would still mean the resources required for training and maintaining these models would become more efficient/reduced over time, not increase forever. The market value of the AI supply-chain currently depends on these models having a voracious and endless need for more and more underlying resources. Once diminishing returns on model size become evident and/or shrinking & efficiency become the competitive focus, then the jig is up for a number of current players.
If you lower the resource requirements by an order of magnitude, you'll simply see more players using it more commonly. You can see that in the way private spaceflight companies have, simply by reducing launch costs by an order of magnitude, made spaceflight and satellite swarms more than a pipe dream for solving classes of problems that require line of sight to large parts of the planet. There are a lot of applications of these floating point processors beyond LLMs and generative AI, and we're probably going to see more of that in the near future.