The Battle for the Economics of Intelligence

“America creates frontier intelligence; China accelerates its diffusion; profits decide the rest.”

America leads in frontier technology and infrastructure revenues. China is accelerating enterprise adoption. The next competitive advantage lies in controlling the workflows that turn intelligence into economic value.

Three developments announced on October 29 materially strengthen our U.S.–China AI thesis — and reveal a new competitive mechanism. The contest is moving beyond frontier-model performance and inference prices toward enterprise distribution, agentic workflows and the financial returns generated by AI deployment.

Since WorkBuddy's March introduction, Tencent reports more than 9 billion task requests, adoption across 50+ industries, penetration of 50% of China's 500 largest enterprises, over 50,000 ecosystem partners and more than 200,000 available agent skills. Overseas cloud customers have doubled over the past year, with international demand growing at a triple-digit rate.

These are company-reported figures, not independently audited measures of productivity or profitability. Nevertheless, they show China's strategy progressing beyond inexpensive model access. The new competitive unit is the enterprise workflow: an organization that embeds AI into communication, document management, procurement and operations can increase consumption without continuously requiring the most advanced frontier model.

The decisive next test is whether reported activity translates into lower operating costs, higher margins and recurring enterprise revenue.

On October 29, ByteDance opened Doubao Work to third-party Chinese models. Enterprise customers can now access Moonshot's Kimi K3.1 Pro, Z.AI's GLM-5.3 and DeepSeek's V4.1-Flash through a common interface, using existing AI credits, with administrators retaining control over model access.

Instead of competing solely to produce the best model, ByteDance is positioning itself to control how enterprises select, combine and consume models. As models become interchangeable for standardized workloads, pricing power may migrate from developers toward platforms controlling distribution and execution — the ability to route workloads toward the most efficient model while owning the customer relationship. China's advantage is no longer simply cheaper inference, but the combination of low-cost models, extensive distribution and integrated agentic execution.

AWS reported quarterly revenue of approximately $46 billion, representing 39% annual growth, while its Bedrock inference platform recorded nearly 17-fold year-on-year growth in token volume. Operating cash flow rose 29% over the trailing twelve months, yet free cash flow deteriorated from positive $14.8 billion to negative $42.4 billion, primarily reflecting AI infrastructure investment. The company reaffirmed roughly $220 billion in 2026 cash capex.

This is not evidence that American AI infrastructure is unprofitable — AWS remains strongly profitable at the operating level. It demonstrates a growing distinction between commercial demand and the free cash flow available after investment. Notably, Bedrock now includes Chinese models from Moonshot and Z.AI alongside American systems: Chinese model diffusion can generate additional business for American cloud providers. America may retain substantial infrastructure revenues even when customers choose Chinese-origin intelligence — the question is whether those revenues justify the capital required.

These developments point toward a redistribution of value across the AI Supercycle. Frontier developers face pressure to justify premium pricing; infrastructure providers must demonstrate utilization and free-cash-flow conversion; enterprise platforms gain control of model selection and workflow execution; industrial adopters capture productivity as intelligence becomes cheaper.

The transmission runs: frontier innovation → lower inference costs → enterprise integration → expanding workloads → new distribution economics → reallocation of profits. This is not necessarily zero-sum — American infrastructure can benefit from Chinese adoption, and Chinese enterprises from American innovation. But competitive advantage increasingly depends on who captures the profits that expansion generates.

Our original proposition remains intact: America leads the frontier; China challenges the economics and diffusion of intelligence. China's advantage must now be proven through measurable productivity, not adoption statistics; America's through sustainable infrastructure returns, not capex and revenue growth alone. No verified October 30 Chinese frontier release or new semiconductor export restriction establishes a reversal of American leadership.

Five Key Takeaways for Investors and Boards

demonstrated value over announcements

• Tencent: WorkBuddy's adoption warrants attention; the next evidence required is recurring revenue and measurable productivity improvement.
• ByteDance: Multi-model enterprise platforms could become a source of distribution power, reducing the leverage of individual developers.
• Amazon: AWS growth confirms demand, but negative group free cash flow reinforces the need to distinguish expansion from sustainable returns.
• U.S.–China interdependence: Chinese model adoption can strengthen American cloud revenues; competition and commercial interdependence can coexist.
• The next rotation: Favor demonstrated economic value — revenue retention, lower cost per completed task, margin improvement and ROIC — over model rankings alone.

America leads in creating frontier intelligence. China is accelerating its diffusion. The next AI Supercycle will be decided by who turns intelligence into productivity — and productivity into profits.