I could see the case for Palantir purely from the ridiculous P/E ratio. On the other hand hey, so does Tesla and it's stayed irrational longer than many shorters have stayed solvent.
But Nvidia seems highly risky to bet against. Also a very high P/E ratio, but not too crazy. And the demand is extremely real and keeps growing.
Until the latest earnings, all the hyperscalers were basically telling us quarter after quarter that they were capacity constrained and backlogged largely due to AI workloads. And capacity generally translated to GPUs. Only now has MSFT reported being constrained on power, which might signal a shift.
But that doesn't mean lower demand, it just means folks lower down the totem pole can finally get their hands on GPUs. Large sovereign nations, including US and China, are jockeying over these things.
Even if this bubble pops, I still don't see demand going down. If you look honestly for indicators, there is tons of data showing rocketing usage which is translating into real productivity wins. In a capitalistic world, that dependency is going to be impossible to wean off of.
I also don't see any real competition on the horizon yet. I kinda understand NVDA's moat is more than just their chips, it's also the ecosystem. I don't claim to deeply understand this (as another comment points out: https://news.ycombinator.com/item?id=45826309), but there are other signs I see:
1. Huawei was forced to train their models on non-Nvidia hardware, and they couldn't get a single training run completed, AIUI due to showstopping bugs and other issues in the stack.
2. Anecdotally I hear AWS has been unsuccessfully trying to get customers to use Trainium. They all prefer Nvidia.
3. Google announced OpenAI would use TPUs on GCP and now is leasing Nvidia capacity from Coreweave to support the deal.
With everyone locked in a Red Queen's race to train the most powerful models ASAP, they can't afford any delays inevitably introduced by new platforms.
The real threat right now is that behind all the smiling partnership announcements, everyone is trying desperately to diversify away from this monopsony. They will eventually succeed, but doesn't seem it will happen anytime soon.
But Nvidia seems highly risky to bet against. Also a very high P/E ratio, but not too crazy. And the demand is extremely real and keeps growing.
Until the latest earnings, all the hyperscalers were basically telling us quarter after quarter that they were capacity constrained and backlogged largely due to AI workloads. And capacity generally translated to GPUs. Only now has MSFT reported being constrained on power, which might signal a shift.
But that doesn't mean lower demand, it just means folks lower down the totem pole can finally get their hands on GPUs. Large sovereign nations, including US and China, are jockeying over these things.
Even if this bubble pops, I still don't see demand going down. If you look honestly for indicators, there is tons of data showing rocketing usage which is translating into real productivity wins. In a capitalistic world, that dependency is going to be impossible to wean off of.
I also don't see any real competition on the horizon yet. I kinda understand NVDA's moat is more than just their chips, it's also the ecosystem. I don't claim to deeply understand this (as another comment points out: https://news.ycombinator.com/item?id=45826309), but there are other signs I see:
1. Huawei was forced to train their models on non-Nvidia hardware, and they couldn't get a single training run completed, AIUI due to showstopping bugs and other issues in the stack.
2. Anecdotally I hear AWS has been unsuccessfully trying to get customers to use Trainium. They all prefer Nvidia.
3. Google announced OpenAI would use TPUs on GCP and now is leasing Nvidia capacity from Coreweave to support the deal.
With everyone locked in a Red Queen's race to train the most powerful models ASAP, they can't afford any delays inevitably introduced by new platforms.
The real threat right now is that behind all the smiling partnership announcements, everyone is trying desperately to diversify away from this monopsony. They will eventually succeed, but doesn't seem it will happen anytime soon.