Ask HN: What features does a chip require to beat Nvidia in the AI race?
Nvidia has a lot of strengths that puts them over other competitors. I'm curious to hear from some technical experts in data center, AI labs, and enterprise. I think there's a lot of nuance here beyond just "faster and more energy efficient".
Most of Nvidia's "strength" is the heavy lifting done by the folks over at TSMC.
Apple and AMD both have access to the same TSMC technology, but neither of them are willing to invest any more into AI than they are currently. Apple ships an inference solution that is a combination of GPU and NPU and mostly okay for inference at home, and AMD ships a high performance enterprise compute solution that scales the whole way up to the needs of supercomputers without needing to make a purely AI-only product line.
If this was the next big money making frontier, then AMD would be chasing after it too. Instead, they sell a pretty damned good enterprise compute product that actually has a life outside of the AI bubble, and will keep selling it to happy customers long after the AI bubble pops.
The question you need to ask is how bad will Nvidia's collapse be. Like, I'm pretty sure some form of the company will still exist, but they have nothing stopping their valuation going from $4T back down to something more reasonable like $250m, they don't do anything any better than anyone else, its all 100% Jensen wooing people with his leather jacket collection.
Nvidia's products have been the biggest cash grab of all time. I don't think it's a matter of other companies believing it is just a bubble and therefore not attempting to compete, but that Nvidia hasn't left them room to compete effectively. That's what I'm trying to get to the bottom of.
Nvidia lost the contract for now two generations of consoles.
AMD was XBone, XSX, PS4, and PS5.
Nvidia was only able to get in on the Switch, a console that "sold well" because it stretched across two generations, and in a units/yr basis, sold less than either of the four. Nvidia sold the Tegra X1 to Nintendo at a break-even just to get the console down to $300, else it was a no-go for consumers.
The Switch 2? Basically DOA, I'm not sure how either Nvidia or Nintendo is going to downplay this.
Realistically, pretty much anything.
Most of Nvidia's "strength" is the heavy lifting done by the folks over at TSMC.
Apple and AMD both have access to the same TSMC technology, but neither of them are willing to invest any more into AI than they are currently. Apple ships an inference solution that is a combination of GPU and NPU and mostly okay for inference at home, and AMD ships a high performance enterprise compute solution that scales the whole way up to the needs of supercomputers without needing to make a purely AI-only product line.
If this was the next big money making frontier, then AMD would be chasing after it too. Instead, they sell a pretty damned good enterprise compute product that actually has a life outside of the AI bubble, and will keep selling it to happy customers long after the AI bubble pops.
The question you need to ask is how bad will Nvidia's collapse be. Like, I'm pretty sure some form of the company will still exist, but they have nothing stopping their valuation going from $4T back down to something more reasonable like $250m, they don't do anything any better than anyone else, its all 100% Jensen wooing people with his leather jacket collection.
The software side is also important. No discussion of Nvidia's moat is complete without also mentioning CUDA.
It's far more believable that AMD aren't competing because they can't.
AMD was XBone, XSX, PS4, and PS5.
Nvidia was only able to get in on the Switch, a console that "sold well" because it stretched across two generations, and in a units/yr basis, sold less than either of the four. Nvidia sold the Tegra X1 to Nintendo at a break-even just to get the console down to $300, else it was a no-go for consumers.
The Switch 2? Basically DOA, I'm not sure how either Nvidia or Nintendo is going to downplay this.