The Upstream Squeeze

When a market commoditizes, the upstream layers—the ones that feed into the commodity—get squeezed. We've seen it with compute, with APIs, and now with model access and AI-native tooling.

• Compute and infra. Training and inference get cheaper per token; margins thin. Value shifts to who uses the compute (models, apps) and who owns distribution and data.

• Models and APIs. As frontier models converge in quality and multi-model routing becomes normal, "which model" matters less than "how you use it"—prompts, workflows, guardrails, and product fit. The squeeze is on undifferentiated model wrappers.

• Data and distribution. The scarce assets are proprietary data, user graphs, and distribution. Builders who own a loop (content → engagement → data → better product) sit upstream of the squeeze. Everyone else is in the commodity layer.

For product and strategy: assume the layer you're in will get squeezed. Your moat is either further upstream (data, distribution, trust) or downstream (user experience, integration, community). The middle—generic API calls, generic UIs—is the hardest place to stay.