Model Models Disruption

When Cheap AI Meets Premium Pricing: Goldman's Uncomfortable Question

Goldman Sachs is asking the question the entire AI investment thesis depends on: if open-source and low-cost models continue improving at their current rate, what happens to the pricing power of frontier model providers? The historical analogy is the commoditization of any technology — from disk drives to cloud computing — where initial high margins erode as production efficiency increases and competition intensifies. OpenAI burning through $3.7 billion in the first quarter of 2026, more than half its revenue, while ChatGPT's market share fell below fifty percent for the first time, gives Goldman's concern concrete numerical grounding.

The counterargument from frontier model providers is that enterprise customers pay for reliability, safety certifications, and integration support rather than raw benchmark performance, and that applications built on top of these models create switching costs that sustain pricing. That argument depends on the capability gap between frontier and near-frontier models remaining large enough to justify the price differential. If GLM-5.2 performs at ninety percent of GPT-5's benchmark scores on tasks that actually matter to enterprise customers — code generation, document analysis, customer service automation — and does so under an MIT license that allows unlimited fine-tuning on proprietary data, the pricing power argument weakens substantially.

Nvidia's Jensen Huang told AP that society needs 'new social norms' for the AI age, comparing the disruption to the arrival of automobiles. The analogy is apt in some respects — cars created new industries, reshaped cities, and required decades of regulatory and infrastructure development — but has limits Huang did not address. Cars displaced horses and some transportation workers while creating millions of jobs in manufacturing, maintenance, and adjacent industries. AI's displacement of knowledge workers, a category that previously required significant education and training, faces less obvious retraining pathways.

Meta's CTO Andrew Bosworth called the company's AI-driven workforce reorganization 'atrocious' and promised better communication and morale-boosting measures after weeks of turmoil affecting twenty percent of Meta's workforce. When roles are eliminated and staff reassigned at that scale and speed, cultural damage accrues that takes much longer to repair than a financial model anticipates. A leaked connection to the Thiel Dialog Society — an invite-only network of over two hundred figures from politics, technology, and finance — offered a rare window into the informal relationship-building environments where investment decisions, regulatory posture, and political access are shaped outside public view.