“Leadership remains essential, but it is no longer sufficient.”
China is accelerating the consumption of intelligence while access to advanced computing remains constrained. America is discovering that technological leadership must also withstand financial-market discipline.
Three developments reported on November 5–6 materially strengthen our central thesis: the United States retains important advantages at the AI frontier, while China is increasingly challenging the economics, distribution and industrial scalability of intelligence. The most consequential is not a new benchmark victory — it is evidence of accelerating Chinese demand for operational AI.
The three developments
November 5–6
-
On November 5, DeepSeek disclosed Harness reached approximately 1.54 million daily active users, up from 271,000 on September 30, with roughly 92 trillion tokens processed daily and 35,000 third-party plugins. These figures are company-reported and unaudited; token volume should not be confused with paid consumption or completed economic tasks. Ant Group also released Ling-3.2-flash, a native multimodal model with 560 billion total parameters, approximately 25 billion activated and a one-million-token context window. signal: China's challenge is evolving from producing intelligence cheaply to distributing sophisticated intelligence across an expanding ecosystem.
-
Reuters reported approximately 600,000 Nvidia H200 processors held in Hong Kong pending mainland import approval. Washington has authorized certain H200 sales, but Beijing restricts imports to reduce dependence even as Chinese developers still need them. Restrictions create incentives to optimize for domestic accelerators and improve inference efficiency, but may also raise costs and slow frontier-model training. signal: The test is whether China can turn computing scarcity into a durable efficiency advantage rather than a persistent handicap.
-
On November 6, Reuters reported Yangtze Memory intends to source at least 60%, potentially 70%, of the equipment for its planned fourth Wuhan factory from Chinese suppliers, producing advanced NAND for data centers and AI. Construction is planned for 2027; negotiations remain preliminary with no formal orders placed. Crucially, NAND is distinct from the high-bandwidth memory used alongside advanced accelerators. signal: An industrial-localization ambition, not proof China has overcome its most consequential AI semiconductor dependencies.
The second-order effect — two different AI capital cycles
Financing the frontier
In China, DeepSeek's expanding agent ecosystem suggests demand for intelligence can accelerate through inexpensive models, developer adoption and commercial integration. In the United States, the financial burden of sustaining frontier leadership is increasingly visible: on November 6 SoftBank shares declined 10.8% while its five-year credit-default-swap spread reached approximately 466 basis points, a four-year high. Investors are questioning the leverage behind major AI investments, including SoftBank's reported $64.6 billion investment in OpenAI.
This does not demonstrate that China's AI industry is more profitable. It reveals two risks: America must prove its enormous frontier investment can generate sustainable returns; China must prove that widespread adoption of relatively inexpensive intelligence can generate sustainable profits despite computing constraints.
What to expect next
Five indicators
DeepSeek's agent economics
Whether rising token consumption produces recurring revenues, attractive margins and measurable productivity gains.
Chinese computing independence
Whether domestic accelerators achieve competitive performance and system-level costs in commercial deployment.
Nvidia's China exposure
Whether Beijing relaxes H200 import restrictions or redirects demand toward domestic alternatives.
Frontier-model convergence
Whether DeepSeek, Alibaba, Tencent or Ant Group demonstrate independently verified improvements narrowing America's remaining advantages.
AI capital-market discipline
Whether infrastructure developers can sustain investment as financing spreads widen and investors demand stronger free-cash-flow visibility.
Demand versus durable profit
- Today's developments strengthen our thesis without establishing a reversal of American technological leadership.
- For investors, the opportunity increasingly lies in identifying where growing demand translates into durable profitability.
- For boards, the relevant measure is shifting from model capability alone toward the cost of successfully completing an economically valuable task.
- For policymakers, the challenge is to preserve advantages without allowing restrictive policies to accelerate competitive alternative ecosystems.
America is competing to extend the frontier of intelligence; China is competing to make intelligence a ubiquitous economic input — and the next decisive advantage will belong to the system that converts intelligence into productivity at the lowest sustainable cost of capital.