I think in short term, it will get better, as more funding flows from industry to academia, you will see a spur in well-compensated PhD positions. However, in the long run, while I doubt that mathematicians will go extinct, they might have to move on to industries like in trading, chip-making, where they oversee AI models write code and also proof that the code works and is consistent with other parts of the software. However, until we have infrastructure for human-AI math research, as Tao himself said that the "roads" for human-AI research is yet to be built, humans would simply be working FOR AI models, and not working WITH AI, which will truly scale discoveries at a massive scale.