What's Still Valuable When Intelligence Is Free

Cliff Sosin argues that AI will make intelligence dramatically cheaper, but that does not mean intelligence becomes unlimited. He frames intelligence as a tool for interpolating between facts, not a magical scaler, and says many important problems are complex dynamic systems where better models still run into uncertainty, time lags, and the need to touch reality through experiments. That leads him to think the biggest winners will not simply be the model builders, but businesses with durable complements to intelligence: data, attention, physical proximity, scale, and regulatory advantages.

He uses Meta as the clearest example, describing it as a business that already spends heavily on intelligence-like inputs while owning the data and attention that cheap AI cannot easily replicate. By contrast, he thinks frontier labs face a mixed outlook: some tasks are commoditizing fast, the market for “good enough” intelligence may be highly price-sensitive, and more intelligence does not automatically create infinite new demand. He is skeptical that the labs’ margins will escape deflation for long, even if they stay important.

Sosin is also more bullish on the second-order effects of AI than on the obvious supply-side infrastructure trade. He thinks America’s capital markets can still reallocate huge amounts of capital quickly, but he worries about overbuild and says it is hard to know which suppliers will capture value. In his own investing process, he expects humans and models to work together for a while, even as agents eventually sharpen prices, automate more tasks, and change how companies like Carvana, Meta, and others operate.