It is a sum of parts story but with reasonable assumptions, the HPC/AI segment of $NBIS looks cheap as hell vs Coreweave
- Assume Avride worth $2b ($5b in 2021 before sanctions shut testing down; $SERV now worth $750mm with just delivery robots, whereas Avride has robots and AV)
- Clickhouse raised in late 2021 with market nuts but it gets rave reviews and reportedly doing 50+mm ARR up 5x y/y and possibly raising again soon, down round seems unlikely on heels of NBIS just raising
- Reasonable multiples for Tokla and tripleten, both growing > 100% in '25
- Assume all the cash is depleted through '25, gets them to 60k GPUs deployed +/-, the HPC/AI biz valued at $3.8 billion vs Coreweave valued at $19 billion with 45k GPUs deployed in May '24
- Corweave reportedly 4x'ing revenue to 8b in '25, has more power secured so perhaps more fwd optionality, looking to IPO 35b; NBIS a more experienced HPC/AI team, basically the technical team from Google of Russia, early HPC/AI experts and largest non-US/China customer of NVDA chips pre-sanctions
- Net, net seems like easy double in '25 with all sorts of optionality that avride and clickhouse get prints that reduce value of HP/AI business even further
- Capital intensive biz and will need to raise again in 2025, but $NBIS *should* be looking at 15b-ish valuation heading into '26 for just HPC/AI biz if it follows the Coreweave path
- Feels like some of the shine come off NVDA recently with outlooks from MRVL/AVGO on custom silicon for hyperscaler customers, but it will take years to scale as NVDA continues to improve its offerings