A bet here is to revisit revenue per deployed H100 by the end of 2027. If the number exceeds one‑hundred‑thousand dollars, the efficiency threshold the short camp predicts has arrived. If it remains close to today’s ten‑thousand, the sceptics will have been vindicated.
I note that CoreWeave has 250,000 GPUs, mostly Hopper series. If net revenue per H100-GPU-year 10xs, they will be in a very good position to benefit from that. My extremely not rigorous back-of-the-envelope estimate is that the CoreWeave valuation makes sense if H-series GPU prices are expected to drop by 10%-30% per year for the foreseeable future. If you instead expect the price of H100s to rise by a factor of 10, that implies that CoreWeave, a public company, is currently trading at far below the correct price.
Not incredibly strong evidence but it does seem like we have a kind-of-prediction-market-ish thing here on net revenue per H100, and the prediction it’s making is quite clear.
Well that aged poorly. Coreweave’s valuation 1.5x’d within 2 weeks of me posting this comment. Mild evidence against EMH here I guess.
(Disclaimer: I do have a small long position in CRWV, entered shortly after posting the original comment on the grounds of “well I’m not positive this is priced in yet”)
Thanks for being transparent; wouldn’t put too much stock in a short-term rise in Coreweave (as I understand, more to do with tariffs and other confounding factors than “they were underpriced before”)
I note that CoreWeave has 250,000 GPUs, mostly Hopper series. If net revenue per H100-GPU-year 10xs, they will be in a very good position to benefit from that. My extremely not rigorous back-of-the-envelope estimate is that the CoreWeave valuation makes sense if H-series GPU prices are expected to drop by 10%-30% per year for the foreseeable future. If you instead expect the price of H100s to rise by a factor of 10, that implies that CoreWeave, a public company, is currently trading at far below the correct price.
Not incredibly strong evidence but it does seem like we have a kind-of-prediction-market-ish thing here on net revenue per H100, and the prediction it’s making is quite clear.
Well that aged poorly. Coreweave’s valuation 1.5x’d within 2 weeks of me posting this comment. Mild evidence against EMH here I guess.
(Disclaimer: I do have a small long position in CRWV, entered shortly after posting the original comment on the grounds of “well I’m not positive this is priced in yet”)
Thanks for being transparent; wouldn’t put too much stock in a short-term rise in Coreweave (as I understand, more to do with tariffs and other confounding factors than “they were underpriced before”)