Geoff Woo: Distribution is the Moat After Models Become Cheaper
Geoff Woo believes that software entrepreneurs who only focus on rate limits are missing the point. He predicts that as inference costs decrease and model capabilities improve, access to raw models will increasingly resemble a commodity, while true scarcity will lie in distribution.
He emphasizes that the winners will not be the teams with the flashiest API calls, but those that have established user relationships before the industry flattens. Related English materials and entrepreneurial discussions consistently point to the same trend: as execution becomes more replaceable by computing power, culture, taste, and channels will become more stable competitive barriers.
This viewpoint is highly consistent with the real changes at the AI application layer. As models become stronger and calls cheaper, the differences between applications will manifest more in entry points, retention, and relationship binding, rather than in the underlying interfaces themselves.
Source: Public Information
ABAB AI Insight
Woo's core argument shifts AI competition from "model scarcity" to "relationship scarcity." As foundational model capabilities gradually commoditize, merely having a better API cannot sustain excess returns; what truly matters is who can embed AI into users' daily lives, creating ongoing usage and switching costs.
This essentially represents a shift in industry focus. Early AI applications attracted users through capability differences, while later AI applications retain users through distribution and positioning. In other words, technological advantages will be quickly neutralized, but entry advantages, habitual advantages, and identity binding will not. Competition at the model layer resembles a short-cycle price war, while competition at the distribution layer is a long-term compounding game.
On a deeper level, this indicates that AI is replicating the evolutionary path of past internet platforms. Search, social media, e-commerce, and mobile applications have all undergone the same process: foundational capabilities gradually commoditize, leaving behind those who control user relationships and traffic entry. AI has simply accelerated this process, as inference costs decline faster, and imitation and substitution occur more rapidly.
From a capital structure perspective, this will further push valuation logic towards "user assets" rather than "technology assets." Teams that can achieve high-frequency touchpoints, retention, and scene embedding will have more stable valuations; teams with model capabilities but lacking distribution control will see their marginal value rapidly compressed. Woo's judgment serves as a reminder: in the AI era, technology can be replicated, but relationships are hard to forge.