@pequityresearch Base case is $8.50 hr rental & $1.75 per million tokens. Both of those are the whole ballgame. What happens to the 31% and 46% if realised token pricing halves from here, which is roughly the trajectory we’ve been on?
@TheReRate@TheTranscript_ Qualcomm’s the wrong comp. It’s a mature, cyclical hardware + licensing business with Low single digit growth cyclical, so 4x sales is what it deserves. Databricks is ~$7B ARR growing 80% at software margins with 140%+ net retention. Databricks = mission critical for enterprise
@rickyho_1989 Spot on. The model becomes the commodity, the learning loop is the moat. Big reason I like Meta.Owns the data, workflow,distribution, feedback loop end 2 end. Billions of users whose behaviour trains the system that monetises them. That learning compounds inside Meta only
@Jason Jevons paradox.Cheaper tokens don’t shrink the market, they blow it open. The model itself gets commoditised, like petrol. Value accrues to whoever owns the ecosystem above. Nobody cares which model served the token, same way you don’t care which refinery your fuel came from
@undrvalue@rickyho_1989 Installed compute does commoditise if it’s just GPU hours. Jevons drives more usage, but doesn’t protect unit economics. The hyperscaler bull case is that compute is bundled into cloud, data, workflows, distribution and proprietary silicon, not sold as undifferentiated capacity.
@theallinpod If AI models commoditise and enterprise access is controlled by cloud platforms like AWS Bedrock, how does a $3T valuation hold? Over time, customers will likely be routed to the best-performing or lowest-cost model, not necessarily Anthropic.
@rubbadubDUBS@WillBiddy_ It took over 15 years to get all the permits throughout all the countries, cities airport etc. you think Tesla or Waymo want to go through that process let alone with a driverless car.