Stratechery 'Agents Over Bubbles': Ben Thompson's Case That AI Compute Demand Is Structurally Immune to Bubble Dynamics

Stratechery·high signal

Thompson argues each LLM paradigm shift (ChatGPT → o1 reasoning → agents) requires exponentially more compute while reducing adoption barriers — agents need fewer decision-makers, not mass consumer adoption. Core thesis: model+harness systems create defensible moats that commoditization arguments miss, putting Google (strong model, weak harness) and Microsoft (abandoned model-agnostic positioning) structurally weaker than Anthropic and OpenAI. Predicts companies will cut employees and rebuild at scale with agents, fueling sustained and accelerating compute demand.