The miners are digging up the gold and selling it. Say they manage to sell $100 worth of gold.
Meanwhile a guy sets up shop selling shovels. For the sake of argument he made $100 selling shovels to the miners.
How much money did the miners make?
How long can that continue until something changes?
TLDR: For those that don't care for analogies a toolmaker can't be making more money then the companies using those tools to make money... or some very obvious things happen.
nvidia can't make more money than the client companies make altogether, but they certainly can make more money than any(/all) of the client companies make using nvidia's products. It could be the case(and currently is the case) that other non-AI revenue streams are just being handed over to nvidia in the hopes the AI revenue streams take off eventually.
Right now companies are spending a ton of money trying to find gold, and that costs capital. That's why they're slapping the word "AI" onto everything to see if it sticks. And re:
> a toolmaker can't be making more money then the companies using those tools to make money.
Every historical gold rush has had exactly that happen which is the point of the analogy. You have a bunch of broke pioneers, a few people that struck it rich on their stakes, and then a saloon owner that rakes in steady income from alcoholic prospectors.
In the historical gold rush, you also had speculators funding prospectors. That might just be personal savings, friends and family or investors. Similar to the current AI rush, you have VCs and corporate profits from companies pivoting into AI that, theoretically, could all go bust. But there still is a whole lot of capital to expend even if the investments are never profitable.
There are a bunch of others:
People start getting shovels from somewhere else even if they aren't quite as good (AMD).
They realise there is actually only a small group of them buying literally all the shovels and they use that position to force more favourable pricing.
Another company works out how to make a better shovel, or worse something that isn't even a shovel anymore and is much more efficient at mining. (These don't exist for training yet but there are inference chips that make NVidia look terrible in comparison)
etc.
The fact of the matter is that it's not a stable situation, something has to give. Hype can allow these sorts of massive imbalances to exist for short periods but there is always balance in the end.