Dylan Patel's conversation with Dwarkesh Patel argues that compute economics are pushing frontier AI toward greater concentration.
The discussion projects rising lab demand, a shift of capacity from inference toward research and training, and a world in which OpenAI and Anthropic can outbid other buyers for a large share of usable compute. It also explores a multi-trillion-dollar capital cycle and possible sovereign and market consequences.
These are forecasts, not settled infrastructure facts. For builders, the useful move is to treat capacity, pricing, and provider concentration as revisable dependencies. The next checks are realized deployments, measured utilization, financing conditions, and compute that actually reaches applications.