Something Is Changing in the Unit Economics of Software
Nicolo argues that AI weakens one of SaaS's favorite properties: near-zero marginal cost. A conventional feature can be copied endlessly once built; an AI feature keeps consuming inference for each user and request. At the same time, cheaper software creation gives buyers and competitors more credible alternatives, putting pressure on prices while compute costs push from below. That is a real squeeze, especially for products whose value is little more than a model call behind a polished interface.
Inference is only one line in the ledger, though. Support, security, compliance, integrations, distribution, and the cost of earning trust often dominate the code itself, while model prices may eventually behave more like ordinary hosting. The deeper threat to undifferentiated SaaS is not merely a worse gross margin; it is that a customer can assemble a sufficiently tailored substitute. Durable value moves toward proprietary workflows and data, reliable operations, distribution, and outcomes that remain difficult after implementation gets cheap.