This nVidia is absolutely all in AI and while I think it is a 'bubble' they will also be the last to be exposed to the pop.
Look at the VC investments and stock prices of the companies delivering AI applications, they are all around that magic 200X earnings number that usually signifies the point beyond which there is not realistic positive return for new investors, unless they happen to get really really lucky and pick the big winner in the specific vertical its kinda like Highlander at this stage, there can be only one.
Think about it like pouring money in IBM because you believed in an OS/2 future in 1988 at that point vs buy MSFT shares; but both were trading at 200X earnings. With MSFT you'd still have made out all right probably, (could not find their actual PE in 88 quickly). My point is there is probably one coding assist, one photo enhancer application, etc to rule them all and everything else will be also-rans in a given vertical. The rest will fold, get acquired cheaply if they have any unique IP, or muddle along without much growth. Unless you picked the big winner you'll never get out what you put in.
On the other hand after each gold rush in a given space peters out, the AI exuberance and capital flows will be pushed into some other vertical in succession. The need for generic-ish hardware good for running ML models will persist. nVidia can expect to keep shoveling chips out the door until either - 1) the market matures and the AI boom full ends, or 2) they hit scaling problems that prevent them from really producing 'better' SI from a performance vs power vs foot print perspective and can't compete with their own previous generation products availible because of the previous vertical app market decline.
The "cloud" is going to be a drag on them too. In the 2000 internet bubble, there was a stigma attached to the used anything, so hardware from failed start ups ended up discounted and sold into secondary markets cheaply. On the hand buying so many units of compute from AWS, well that is a pure commodity and AWS can just leave their compute plant installed and rent it to the next client. Amazon does not need to replace or upgrade it, unless the cost of operation is improved enough to justify the cap-ex. To that end we can expect some flattening of Nvidia sales before they big drop. So you if you watch the quarterly reports, you'll know exactly when to get out. (This is my theory anyway)