Offline with Jon Favreau Endless Slop, Cancer Cures, or Robot Apocalypse? Derek Thompson on Our AI Future
70 snips
Mar 7, 2026 
Derek Thompson, journalist and author who writes on economics, technology, and culture. He unpacks why AI’s labor impact is so hard to measure. He debates whether AI is a bubble or a boom. He explores AI in medicine, national-security risks like the Anthropic-Pentagon clash, and his 'Everything Is Television' take on how media and politics are changing.
AI Snips
Chapters
Transcript
Episode notes
GPU Economics Could Expose An AI Leverage Problem
- Chip and capital economics could make today's AI industrial build unsustainable if costs must be refreshed faster than accounting depreciation.
- Thompson notes GPUs depreciate over five to six years while analysts fear re-up cycles could be every two years, squeezing profits.
Fast Revenue Growth Counters The Bubble Claim
- Rapid revenue growth from frontier labs makes a counterargument that this is not a pure bubble.
- Thompson cites projections of Anthropic and OpenAI revenues (tens of billions) and Stripe data showing AI firms grow faster than previous cohorts.
AI Turned Five Hours Of Analysis Into Minutes
- Derek tested Claude Code on the American Time Use Survey and produced ten graphs and analysis in minutes.
- An Aspen Institute economist found Claude's output within 1–2% of expert work that took five hours, and called it terrifying.