Dwarkesh Podcast Dwarkesh Podcast

Dylan Patel – Anthropic & OpenAI will have most of the world’s compute by 2028

Aug 25, 2026 · 1h 16m

Summary

Dwarkesh Patel interviews Dylan Patel of Semi Analysis on the rapid centralization of AI compute, where labs like OpenAI and Anthropic are capturing half of incremental global capacity by 2027. They discuss how these labs are shifting from venture-funded losses to massive profitability, enabling them to outbid competitors for infrastructure. The conversation covers supply chain bottlenecks, regulatory impacts on model releases, and China’s lagging but accelerating domestic chip production.

Topics discussed

Intro: Semi Analysis founder Dylan Patel and AI CapEx trends AI Labs profitability and the shift from VC funding to revenue Compute centralization: SpaceX, OpenAI, and Anthropic's infrastructure The economics of compute: $6B CapEx for $100B revenue potential Supply chain bottlenecks and arbitrage opportunities in hardware Future compute projections: 100 gigawatts for top labs by 2028 Regulatory impacts on model release and revenue per megawatt Value capture shifts from chipmakers to AI labs and hyperscalers Sponsor: Using GrokBot to automate podcast editor recruitment Revenue forecasts and the timeline for Recursive Self-Improvement Investor pressure vs. safety: Allocating compute for training vs. inference Global compute growth: US dominance vs. China's constrained expansion China's semiconductor subsidies and long-term AI competitiveness Training vs. Inference ratios and the $10 trillion CapEx forecast Funding the AI boom: Debt issuance by hyperscalers and suppliers Sponsor: Antithesis deterministic software testing platform Sovereign debt crisis risks and the crowding out of other industries Interest rate impacts and the valuation of non-AI stocks Economic doubling and the potential default of non-AI nations Political constraints and the possibility of slowing AI progress Sponsor: Jane Street intern projects and financial modeling challenges The inevitability of centralization and value capture in the AI economy
Listen ad-free on Castria