dwarkesh patel

Frontier lab revenue is compounding at 10x/year while compute supply only scales 3x/year (Moore's Law 1.4x + new fabs 1.2x + wafer reallocation 1.8x). This structural imbalance forces either lab margins >90% or compute price increases — with supply inelasticity (EUV bottlenecks, wafer ceiling) making sustained compute price inflation the likelier outcome.

The 3x annual compute growth relies on three components (Moore's Law 1.4x, new fabs 1.2x, wafer reallocation 1.8x) all hitting physical limits by 2025-2030, making supply unable to absorb demand shocks unlike commodity markets.

The 3x annual compute growth decomposes to: 1.4x from Moore's Law (slowing, miracle to sustain), 1.2x from new fabs (bottlenecked by ASML EUV output through 2030+), 1.8x from wafer reallocation from smartphones/PCs to AI (AI taking 60%→86% of TSMC N3, hitting 100% ceiling). All three face hard physical limits within 1-5 years.

Models cannot learn from deployment data without updating weights — in-context learning scales poorly. On-policy self-distillation (OPSD) offers dense, sample-efficient weight updates without verifiable rewards, while 'dreaming' (test-time training against self-built simulators) could become a fourth scaling axis alongside pretraining, RL, and inference compute.

Labs are betting billions on RL from Verifiable Rewards (RLVR) across containerized environments to build general agents, but Dario Amodei's comments suggest short-horizon RL may not generalize to long-horizon real-world tasks like building businesses, making the trillion-dollar scaling bet uncertain.

As AI models become more capable, they can monetize compute far more effectively, driving up compute prices despite 3x annual supply growth. The marginal value of compute could reach $250K/year per H100 equivalent (15x current spot) if AI achieves human-level software engineering.

When models improve from each user interaction, switching AI providers becomes like firing an experienced employee. This lock-in lets labs charge high margins, similar to cloud providers, and they can use subsidies or access restrictions to incentivize data sharing.

As compute costs rise, labs with more token-efficient models can charge large premiums because weaker models burn more expensive compute for same results, creating winner-take-most dynamics in model quality.