Will the US Lose to China on AI? Stock Markets Already Know the Answer, Says Dimitri Zabelin

Key Concepts

• Sovereign AI: The strategic development of AI infrastructure by nations to maintain technological and geopolitical competitiveness.
• Amara’s Law: The principle that we tend to overestimate the impact of a technology in the short run and underestimate it in the long run.
• Jevons Paradox: The observation that as technology increases the efficiency with which a resource is used, the total consumption of that resource may increase rather than decrease.
• Distillation: A process in machine learning where a smaller model is trained to replicate the output of a larger, more complex model.
• Inference: The phase where an AI model is deployed to process data and provide answers (e.g., interacting with ChatGPT).
• Memory on Compute: A hardware architecture where memory units are placed directly on the computational unit to reduce latency and power consumption.

---

1. The US-China AI Hegemony

Dimitri Zabellan argues that the US remains the "indisputable hegemon" in AI, supported by both hardware and software superiority.

• Hardware Gap: The US, in partnership with Taiwan, can mass-produce 2-nanometer chips, whereas China is currently limited to 7-nanometer technology. Zabellan compares this performance gap to the difference between a bicycle and a motorcycle.
• Structural Dependency: China’s recent attempts to compete (e.g., the DeepSeek model) rely on "distillation," effectively training their models on US-generated outputs. Zabellan asserts that because China is structurally dependent on US ingenuity, they will remain perpetually behind, similar to the Soviet Union’s failed attempts to replicate US semiconductor technology.

2. Market Dynamics and the AI Buildout

The current market rally is driven by the "infrastructure play"—the necessity of hardware to support software development.

• The Memory Bottleneck: There is a significant structural lag in memory production. Building a fabrication plant ("fab") takes two years, followed by an 18-month quality inspection cycle. This 3.5-year lead time creates a massive, compounding backlog for memory chips.
• Phased Expansion:
  • Phase 1 (Training): The current phase involves building massive data centers. Approximately $3 trillion in expansion is expected in the US between 2027 and 2030.
  • Phase 2 (Inference): This phase, which has not yet fully begun, involves the deployment of models for end-user interaction.
• Geopolitical Tiering: Access to high-end GPUs is restricted by a three-tier system: Tier 1 (US security architecture/Five Eyes), Tier 2 (General allies), and Tier 3 (Adversaries like Iran, North Korea, and Afghanistan).

3. ROI and Market Sentiment

Zabellan addresses the skepticism regarding AI’s Return on Investment (ROI) and the "hype" cycle.

• Measuring ROI: ROI in AI is often qualitative rather than immediate dollar-denominated. For example, in cybersecurity, AI has been shown to increase analyst "free working hours" by 96%, allowing for higher productivity that is difficult to quantify in traditional accounting.
• Narrative Runway: Zabellan suggests that AI has a "narrative runway"—a period of sustained investor enthusiasm—that is currently strong enough to override macro-economic headwinds like inflation and potential interest rate hikes.
• The "Over-Correction" Fallacy: While some data centers are being delayed or cancelled, Zabellan views this as a "tailwind" for hardware providers like Nvidia and Micron, as it further elongates the backlog and sustains demand.

4. Macro-Economic Headwinds

The conversation highlights the tension between the AI boom and inflationary pressures.

• Inflationary Bottlenecks: The physical construction of data centers requires energy, raw materials, and labor. Geopolitical instability (e.g., the Strait of Hormuz) threatens energy and supply chains, which is inherently inflationary.
• Fed Policy: Despite the market's initial fear that inflation would force the Federal Reserve to hike rates, the market has shown resilience. Zabellan posits that the market has already "internalized" these risks, viewing the current geopolitical tensions as temporary strategic posturing by actors like Iran ahead of US elections.

Synthesis and Conclusion

The AI sector is currently in a transformative "buildout" phase that is structurally supported by a massive hardware backlog and a long-term strategic necessity for nations to maintain technological sovereignty. While short-term market jitters exist due to inflation and questions regarding immediate ROI, the underlying demand for compute and memory remains robust. According to Zabellan, the US maintains a decisive lead, and the "narrative runway" for AI is likely to continue driving market performance despite macro-economic challenges, as the technology’s potential for autonomous reasoning represents a fundamental shift in the global economy.