Summary
Theo digs into a viral claim that Lauren, the creator of the P stack and a Cursor engineer who previously worked on the React compiler at Meta, is burning through roughly $132 million of AI tokens per year. Her Cursor profile shows about 1.5 trillion tokens used in 30 days, far beyond what Theo reaches even when he tries to use as many as possible.
He questions whether the dollar figure makes sense and explains how labs and companies such as OpenAI, Anthropic and Cursor actually spend tokens, and why heavy spend can reflect pioneering agentic workflows rather than waste. He also pushes back on online critics who argue that people at the labs can’t justify this level of usage.
A major thread is the economics of intelligence. Theo argues that the cost of a given level of capability keeps falling quickly, citing newer models such as GLM-5.3 Flash and DeepSeek-4.1 Flash that match or beat last year’s top model, GPT-5 High, on benchmarks while being cheaper and better at long-running agentic tasks. The video is useful for developers and followers of the industry who want context on token costs and agent-driven development.
📺 Source: Theo – t3․gg · Published October 27, 2026
🏷️ Format: Opinion Editorial
