Dwarkesh Patel analyzes the widening gap between AI labs' revenue growth (10x/year) and compute growth (3x/year), arguing that this imbalance will force compute prices to rise dramatically, potentially 10x or more. He discusses the economic implications, including the Alchian-Allen effect, where smarter models increase the value of compute, and the challenges of scaling compute production due to physical bottlenecks.
Summarized by Podsumo
Anthropic's revenue has grown 10x year-over-year for three consecutive years and could reach $100-150 billion this year, while lab compute only grows 3x annually. This gap forces either higher margins or higher compute prices.
The price of compute is already rising: spot prices are 40% higher than February lows, and Google pays 2x spot price for 110,000 GPUs rented from SpaceX ($900 million/month).
If a human-level AI software engineer runs on an H100, that H100 could rent for over $250k/year, 15x the current spot price, illustrating compute's potential value.
Compute scaling is bottlenecked: Moore's Law provides only 1.4x, new fabs 1.2x (limited by ASML EUV machines), and wafer reallocation 1.8x, which will hit a wall by end of next year.
The Alchian-Allen effect: as compute becomes more expensive, labs with more efficient models can charge higher margins, since wasting tokens on weaker models becomes costlier.
Dwarkesh draws parallels to the Simon-Ehrlich bet, warning that compute scarcity may persist longer than commodity scarcity due to inelastic supply.
The episode highlights the strong economies of scale in AI models—one training cost is shared across all users—leading to concerns about power concentration.
"If you have a model that can get the same result by using less compute, then you've in some sense created more compute and the value of compute is gonna increase."
"I'm a bit worried that this kind of analysis honestly pattern matches a lot onto the ways that people in the past have been wrong about scarcity."
"The supply of compute is much less elastic and much less capable of absorbing large demand shocks than the extraction of different metals is."