This is really exceptional, regardless if your thesis is right. Really appreciate thinking against the grain in such a crowded market. You’re 100% right, due to the positioning in NeoClouds, the opportunity for repricing is strong
Thanks for sharing the analysis. I am also bullish and invested into CoreWeave before reading this series.
This post only seconds my thesis.
I would also like to add another moat - reference architecture for running a data center. As mentioned by Jason, it is one thing to buy equipments but a totally different thing to run everything in sync. The company also has a state of art monitoring solution for Data center operations.
If we get a serious compute shortage I expect the labs would deal with shitty neoclouds/clusters by hiring the best & assigning their own staff to fix the problems. So I don't see why anything other than spot price exposure would matter that much.
I'd be interested to see a comparison of neoclouds modeled out under a fast AGI/useful AI scenario. It seems that CRWV has a very low mix of on-demand capacity. In case of demand shock, they will be unable to capture that convexity and surplus will instead flow to the labs. New capacity doesn't necessarily alleviate that issue either, as there are 2+ year lead times and NVDA will have raised prices post-demand shock. Hyperscalers and neoclouds who reserve a higher capacity of on-demand supply would capture the outsized returns. I like IREN in particular. Inferior pricing power, but a lot of their supply is as yet un-contracted. Furthermore, a lot of that of that demand shock is likely to be on the inference side, which is more fault tolerant and perhaps more forgiving to newer entrants with less cluster management experience. The winners in inference will be those with cheapest inputs (power, data center leases, etc). Think neocloud is right but there may be better names to look at than CRWV.
None of is to mean your calls would be a bad idea, a rising tide raises all ships. Also the assumptions in your analysis don't depend on really any demand shock -- seems like the street is just grossly mispricing current GPU rental rates. Which completely separately, that's interesting, could be worth a further deep dive into the GPU hour pricing numbers.
Great post. Thank you
This is really exceptional, regardless if your thesis is right. Really appreciate thinking against the grain in such a crowded market. You’re 100% right, due to the positioning in NeoClouds, the opportunity for repricing is strong
I like your arguments, but I still like NBIS more. You interested in also looking into NBIS?
Thanks for sharing the analysis. I am also bullish and invested into CoreWeave before reading this series.
This post only seconds my thesis.
I would also like to add another moat - reference architecture for running a data center. As mentioned by Jason, it is one thing to buy equipments but a totally different thing to run everything in sync. The company also has a state of art monitoring solution for Data center operations.
If we get a serious compute shortage I expect the labs would deal with shitty neoclouds/clusters by hiring the best & assigning their own staff to fix the problems. So I don't see why anything other than spot price exposure would matter that much.
I'd be interested to see a comparison of neoclouds modeled out under a fast AGI/useful AI scenario. It seems that CRWV has a very low mix of on-demand capacity. In case of demand shock, they will be unable to capture that convexity and surplus will instead flow to the labs. New capacity doesn't necessarily alleviate that issue either, as there are 2+ year lead times and NVDA will have raised prices post-demand shock. Hyperscalers and neoclouds who reserve a higher capacity of on-demand supply would capture the outsized returns. I like IREN in particular. Inferior pricing power, but a lot of their supply is as yet un-contracted. Furthermore, a lot of that of that demand shock is likely to be on the inference side, which is more fault tolerant and perhaps more forgiving to newer entrants with less cluster management experience. The winners in inference will be those with cheapest inputs (power, data center leases, etc). Think neocloud is right but there may be better names to look at than CRWV.
None of is to mean your calls would be a bad idea, a rising tide raises all ships. Also the assumptions in your analysis don't depend on really any demand shock -- seems like the street is just grossly mispricing current GPU rental rates. Which completely separately, that's interesting, could be worth a further deep dive into the GPU hour pricing numbers.