“The AI race is no longer about who builds the smartest machine — it is about who builds the most profitable economy around intelligence.”
America is building maximum capability. China is optimizing maximum diffusion.
The significance of recent developments is not simply that both countries are investing more in AI. It is that they are investing against different economic constraints. The U.S. constraint is increasingly physical: frontier intelligence requires more advanced chips, larger clusters, more power and unprecedented capital. The Chinese constraint is technological: restricted access to the most advanced U.S. compute forces China to extract more capability from domestic hardware while driving down the cost of deployment.
The Evidence
Two buildouts, two constraints
Two Production Systems
Different flywheels, different constraints
The U.S. Flywheel
Better models → more compute → larger capital commitments → frontier capability → premium economic value. The constraint is physical, and the machine is extraordinarily capital intensive.
The Chinese Flywheel
Lower-cost models → cheaper inference → more deployment → higher utilization → faster learning and industrial adoption. The constraint is technological, met by domestic substitution.
The U.S. Production Model
Frontier models, frontier chips, hyperscale compute, dedicated power, deep private capital — the most powerful intelligence system ever assembled.
The Chinese Production Model
Near-frontier models, domestic accelerators, lower inference costs, open software, rapid industrial deployment — intelligence made cheap and pervasive.
The Core Tensions
Where the race is actually decided
-
The American advantage rests on four reinforcing layers: frontier models, accelerated computing, capital formation, and power. Together these create an unprecedented industrial machine for producing intelligence — but one that is extraordinarily capital intensive. signal: Can America preserve enough pricing power to earn attractive returns on the capital required to build the frontier?
-
China's strategy is driven by three levers: cost compression (lower inference costs expand viable workloads), domestic substitution (Huawei Ascend and local software reduce U.S. dependence), and industrial deployment. The larger prize is not model-provider margins but productivity. signal: China's bet is that AI creates more value as an input to the economy than as a premium product in its own right.
-
Lower prices do not necessarily mean lower total economic value. Cheaper inference drives more viable workloads, higher usage, deeper integration, higher productivity and a larger surplus — but that surplus can accrue to model providers, cloud and semiconductors, platforms, industrial companies, or consumers. signal: Diffusion creates the surplus. It does not determine who captures it.
-
The U.S. system is absorbing extraordinary capital across chips, memory, networking, data centers, power, private credit and infrastructure funds. If intelligence becomes cheaper faster than capital intensity declines, revenue can grow while returns on capital fall. signal: The distinction that matters is profit per token versus economic value per token — the winners combine both.
Why It Matters for the Velocity Edge
The core thesis, updated
The Investment Implication
A new question for capital
The two AI economies
- The U.S. is maximizing the returns on frontier intelligence; China is maximizing the economic impact of lower-cost intelligence.
- The competitive question is shifting from who owns the frontier to who owns the economics of intelligence.
- Diffusion creates the surplus but does not determine who captures it — the winners combine capability, diffusion and capital efficiency.
- Both models can create enormous value; they distribute that value differently.
- The decisive measure of the next AI era: how much economic surplus does AI create — and who captures it?