Conclusion: what remains scarce?
The terminal question is ownership through migration
The thesis is not that AI eliminates rents. It is that AI recursively relocates the bottlenecks supporting rents. Some scarcities survive: legal permission, sovereign authority, unique physical resources, genuine counterparties, trusted state and irreducible physical scale are plausible candidates.
Durable value creation = invested capital × (ROIC − WACC) × competitive-advantage period and fade
Ownership of Sₜ matters only insofar as it preserves a positive ROIC–WACC spread. The financial question is not scarcity in the abstract, but how the bottleneck changes incremental returns, required reinvestment and the rate at which excess returns fade.
The ultimate valuation winner is therefore not necessarily the provider of the smartest model. It is the actor that owns the binding scarcity long enough to extract rent before the next migration—and can continue owning, acquiring or controlling the bottlenecks as they move.
The investigation began with a valuation question and repeatedly encountered the same distinction. A technology can be extraordinarily important without granting its provider a proportionate claim on the value it creates.
World purchasing power constrains simultaneous monetization even while real output can grow. Population must be filtered through relevance, affordability, purchase frequency and share. B2B expenditure must ultimately justify itself through useful output or productive capacity. High multiples require not only growth but long-lived excess returns.
AI then complicates every layer. It can expand real output while reducing unit prices. It can replace labor while also redistributing productive capability to individuals and small teams. It can automate enterprise workflows and then redesign them so the workflow disappears. It can use general intelligence to discover deterministic or specialized execution paths that require less general intelligence. It can make software cheap enough that users generate capabilities instead of renting packaged products. It can accelerate replication, rediscovery and functional substitution. It can lower organizational scale even where physical scale remains.
None of these mechanisms proves that a particular AI company cannot be worth a very large amount. They change what must be proved.
AI importance is the beginning of the valuation argument, not its conclusion.
The durable bull case must identify a residual scarcity that survives sufficient alternatives, customer optimization, process redesign, open knowledge, cheaper organization and technological substitution. It must show that the provider owns that scarcity, can price it, can preserve margins around it, and can do so for a duration long enough to support the present value.
After intelligence has successfully created the value attributed to it, what remains scarce?
If the answer is a durable bottleneck, extraordinary corporate value can follow. If the answer is that capability becomes reproducible, substitutable, interoperable or cheap, AI may still be one of the most valuable technologies humanity has produced—while much of that value appears as lower costs, broader productive power and consumer surplus rather than as permanent rent to the supplier of intelligence.