Is AI Really Eating the World? (philippdubach.com)

Benedict Evans framed generative AI as a potentially new platform shift—important but with an unclear trajectory—and this piece pushes back, arguing the evidence points toward commoditization rather than a winner-takes-all model-provider monopoly. Hyperscalers are pouring capital into models (roughly $400B in AI capex in 2025; Microsoft now spends >30% of revenue on capex), producing systems that are more capable (GPT-4’s complex reasoning, Claude’s 200k-token contexts, Gemini’s multimodal inputs) but less defensible. Barriers remain—DeepSeek showed a $500M budget can build frontier models—but costs and API prices have collapsed (OpenAI API down ~97% since GPT-3), and many models now cluster around similar benchmark performance. That economic reality matters: most current value is coming from integration, process redesign and distribution, not raw models—consulting firms and system integrators are booking billions (Accenture expects $3B GenAI bookings in FY25). Adoption is uneven—92% of developers use AI coding tools, but many enterprise pilots remain experimental—so we’re largely in “absorb” stage. Crucial technical-open questions remain (do LLMs truly reason or just pattern-match?), which determines whether data network effects persist or model capability unbundles recommendations from scale. Even if AGI arrives, the author argues competition, price pressure, and commoditization mean economic gains will likely accrue to AI users and integrators who control distribution and domain value, not necessarily to model developers.