“The next battle turns on who controls the economics of intelligence.”
Three developments reported on November 2–3 materially strengthen our central thesis: America leads the AI frontier, while China is challenging the economics and diffusion of intelligence.
The competition is evolving beyond model capabilities and semiconductor performance. It increasingly concerns control over computing access, distribution channels and the financial infrastructure supporting AI expansion. America is strengthening its control over the infrastructure required to produce frontier intelligence; China is accelerating the commercialization and distribution of intelligence itself.
Three developments
November 2–3
-
On November 3, the Financial Times reported that Nvidia instructed international cloud and data-center partners to sell at least 51% of their computing services to non-Chinese customers. The requirement applies to future contracts and was communicated verbally rather than through documented regulation. Chinese companies have increasingly rented overseas capacity to access processors unavailable for direct import; this corporate commercial restriction operates alongside U.S. export controls. The second-order effect could be an acceleration of technological separation rather than permanent reinforcement of American dominance. signal: America's advantage now includes influence over computing access beyond its borders — strengthening China's incentive to build domestic alternatives.
-
Reuters reported on November 3 that AI computing credits have become a significant product category in China's Singles' Day festival. During a November 2 livestream, retailer Austin Li sold 100,000 monthly memberships to Alibaba's QwenWork — roughly 2 trillion AI tokens — in about 30 minutes. Alibaba reported AI-token package sales rose 450% over October 1–20 versus the prior month. Tmall, JD.com and others now distribute packages from DeepSeek, Moonshot, MiniMax and Z.AI. These are company-reported figures; prepaid tokens do not establish realized consumption. Intelligence is beginning to acquire the characteristics of a standardized, commercially distributed input. signal: China's opportunity is not the most powerful model but inexpensive, accessible intelligence integrated into everyday commerce.
-
On November 2, Marvell announced its proposed acquisition of optical-components manufacturer Lumentum at an enterprise value of approximately $115 billion including debt — one of the largest AI-infrastructure bets yet. The rationale is optical connectivity: as more accelerators operate together, information transfer between processors becomes critical. The financial reaction was unfavorable — Marvell fell more than 10% and Lumentum more than 15% on valuation and execution concerns. The deal awaits regulatory approval and is expected to close in 2028. signal: The bottleneck is shifting from processors to integrated systems — but identifying it differs from paying an attractive price to control it.
The Second-Order Effect — Two AI Economies, One Financial Constraint
divergence under pressure
The American model is defined by frontier capabilities, advanced infrastructure and substantial capital intensity; the Chinese model by efficient models, broad distribution and commercial diffusion. Neither is financially unconstrained.
At the Hong Kong Global Financial Leaders' Investment Summit on November 3, Xiao Yuanqi, deputy head of China's National Financial Regulatory Administration, warned that concentrated AI investment could raise borrowing costs for traditional industries and increase financial risks. Executives from Blackstone, Goldman Sachs and JPMorgan emphasized continued demand while acknowledging valuation and capital-allocation risks.
The deeper implication: the AI Supercycle can accelerate technologically while becoming more difficult to finance profitably. Expanding token consumption does not guarantee profitability for model developers; rising demand for computing does not guarantee attractive infrastructure returns; growing adoption does not automatically establish measurable productivity. No independently verified November 3 DeepSeek release or frontier-model breakthrough establishes a reversal of American technological leadership. The winners will convert technological capability into sustainable financial performance.
Five Decisive Signals To Monitor
what changes our assessment
Computing access
Whether offshore cloud restrictions materially reduce Chinese access to advanced accelerators.
Chinese semiconductor independence
Commercial improvements in domestic accelerator performance, advanced-memory yields and system costs.
Token economics
Whether Chinese token sales translate into recurring consumption, stronger enterprise revenues and lower cost per completed task.
Capital-market discipline
Whether infrastructure spending generates improving utilization, operating cash flow and returns on invested capital.
Frontier-model convergence
Independently verified Chinese model improvements that materially narrow the gap with leading American systems.
The Final Take
separate the two scorecards
America is competing to control the production of frontier intelligence; China is competing to accelerate its economic diffusion. Both strategies can succeed simultaneously, but their returns will depend on capital efficiency, market access and measurable productivity.
For investors and boards, technological leadership and economic value creation must be evaluated separately. For policymakers, restrictions that protect technological advantages must be weighed against their potential to accelerate rival ecosystems.
The next great battle in artificial intelligence will not be decided solely by who builds the most powerful intelligence, but by who controls its economics—and who captures the value when intelligence becomes ubiquitous.