The fusion test
Fusion provides a useful extreme because it appears to offer an extraordinary proprietary technology with enormous physical value. Suppose a company discovers a reactor capable of producing electricity at 2 cents per kilowatt-hour.
The first question is not whether fusion is revolutionary. It is whether the plant can recover its capital at the market price. A simplified annual capital-recovery requirement is:
Required annual revenue ≈ Capex × capital-recovery factor + operating cost
If technological competition drives electricity toward the plant's marginal cost, a capital-intensive first generation can become financially unattractive even while its engineering is extraordinary. A cheaper solar-storage system, improved fission, geothermal system or second-generation fusion design can cap the price.
The lesson is not about a specific electricity price. It is structural:
A breakthrough can create abundance that destroys the scarcity rent needed to reward the capital that created the breakthrough.
This is the physical-world analogue of software commoditization. The provider needs not only superior technology but an economic structure that preserves a return on the installed capital.
Two capital-intensity regimes
The fusion case and the declining-capital-intensity thesis are compatible only after capital is decomposed:
Required invested capital = irreducible physical capital + reducible information, coordination and organizational capital
If minimum physical investment, construction time, grid access, permitting and specialized supply remain large, capital intensity survives AI and can support a durable barrier. If AI lowers design and coordination cost, improves utilization, enables modular substitutes or reduces the physical plant required for a sufficient outcome, current sufficient-entry capital can fall even while the first generation remains physically enormous.
Section 19 asks whether useful intelligence must inherit frontier-compute intensity. Section 23 asks whether entrants must reproduce historical invested capital. The fusion test supplies the boundary condition: accumulated capital is a moat only to the extent that a competitor still needs comparable, irreducible capital to deliver the customer outcome.