It takes time, NV's journey began with V100, it took years and a few generations of products before a customer managed to make a killer app that skyrocketed demand.
Even then, it took some time before their new supply chain was online, then revenue rose like mad.
There are plenty of potential headwinds for everyone coming though. China was a massive share of the demand, and that market is nearly closed now.
MI300 is ramping faster than A100 did, quite a lot faster. Only H100 is a faster ramp once the cork was pulled, qualification for that mother took a long time.
AMD had virtually zero share until after MI100, that was just there to establish the CDNA chain for the future, MI200 was the first class leading part, but only in HPC but supplying those contracts certainly helped prepare for higher demand with a more general purpose accelerator, which is what MI300 is. It is the best part outside of training, even then it can basically reach parity in training if optimised, but training isn't nearly as agnostic as inference is right now. MI400 with the new on package Pensando DPU and Broadcom xGMI switches is gonna be litty though, the target there is comprehensive leadership.
NV may be a software first company, but they make nearly all their money on hardware.
They focus so hard on software to be ready for new markets, and then to provide the best software for said markets, software that only runs on their hardware or is exceptionally well optimised for their hardware.
NV's software is to lock people into their hardware ecosystem, which is where they make their money. Full vertical integration to extract obscene GM's, and to hope their software is comprehensive enough to discourage customers from migrating. Plus some anti-competitive stuff with smaller OEM's but I digress.
They do want a big SaaS presence too, but they aren't MS/Alphabet/Amazon/Meta/Oracle...
People distrust IHV software of that nature because of the lock-in threat, 3rd party software will work for the hardware that makes sense which makes customers more receptive.
The fact is though, the dominant software in AI is developed by the SV ISV titans, companies that NV or any other IHV will likely never surpass in revenue.
So in the hardware game, the biggest stick wins, AMD has multiple generations of SoIC already, and CoW_L coming next gen, and the disaggregation just keeps going from there.
Nobody knows stuff further out than that, or if they do, any mention of it would end with a R9X delivered straight from Austin.
NV has a lot to prove that they can scale their sticks to the same degree, the fun has just begun.