I was watching this interview w Jensen (on Dwarkesh podcast) and he asks this question in some form
The answer Jensen gave was: “yeah well maybe TPUs are faster and cheaper for gen AI workloads, but they can’t do other stuff as well such as video games and molecular biology or mining”
While true isn’t that like… ignoring the fact that 99% of nvidias massive bull run is from gen AI workloads?
No one’s suggesting nvidia will ever be irrelevant but when they dominated gaming and mining they were a 0.5T company not a 5.0T
Am I missing something obvious here?
NVDA stock: Is there a good answer for “how do TPUs *not* pose a threat to GPU”?
byu/Tim_Apple_938 instocks
Posted by Tim_Apple_938
7 Comments
No youre not. I think Jensen was throw off guard by all the tough questions and lost his cool. Thats why I think NVDA stock price will no longer go up much more than the semi index will.
It’s already starting to show up. Listen to Amazon and Google er specifically what they are saying about their own chips.
Jensen is right in that TPUs offer a limited scope in their capabilities. An NVIDIA accelerator is as general purpose as it can be. Hence why a chip like those made by NVIDIA are the ones used for things other than gen AI. It remains to be seen if there is such a large market for non-llm stuff.
Nvidia were never irreplaceable in the chain like ASML or TSM. They were early, but that advantage doesn’t last forever.
TPUs are objectively better than GPUs for LLMs and anything that just requires pure matrix maths, probably by around 20-30%. GPUs in theory are more versatile and if there’s a significant regime change in how these models operate could be better placed however I think this will be unlikely.
Nvidia’s actual advantage is their closer relationship with TSMC and booking all of their manufacturing.
I think what Google, Amazon, and Microsoft are developing in-house are being done so to fit a use case within those companies. Be good at supporting a business area more targeted to themselves so the feedback loop, scale, whatever jargon leads back to improving their business and cash flow. So NVDA is the leader in general, will remain so for minimum of the next few years, and is creating a fortress of a balance sheet. I would expect them to get their eyes on the horizon in the next 5 years, though. Put their cash to work.
The answer is the size of the market. As long as the size of the market is bigger than what can be filled, there is room for everybody to grow.