Will model labs have pricing power?

The most rational take on AI you’ll hear this year

Topic 8 of 18

Published May 31, 2026 · Clip 32:39–37:05 (4:26)

Benedict critiques Sam Altman's 'selling intelligence on a meter like water or electricity' by pointing to the margin structure of the utility industry — TV companies and appliance makers don't pay electricity companies a percentage. His telecoms analogy: a ~$1T-revenue mobile industry spending ~$200B a year on capex, with data traffic up 1,500-2,000x since 2010, yet flat stocks for 25 years — a high-growth, low-margin commodity utility where the cool stuff got built further up the stack. The elemental question: do we get three giant models or hundreds of open/local ones, and do models have leverage like Windows or look like AWS, where buyers don't care which cloud software runs on? If the chatbot isn't the UX, apps must be built by others, and models are commodities, why would model companies have pricing power? Today's extreme token bills — like the guy who spent $1.5M in a month — are like a $50K mobile data bill in 2010, temporary artifacts before the steady-state equilibrium.