How Cheap AI Could Derail OpenAI And Anthropic's IPOs

Key Concepts

• Frontier Models: Large-scale, high-capability AI models (e.g., OpenAI’s GPT, Anthropic’s Claude Opus).
• Open Source AI: Models with publicly available weights, increasingly dominated by Chinese labs (e.g., DeepSeek, Moonshot).
• Inference Efficiency: The ability to run AI models with minimal compute resources (GPUs).
• On-Prem/Air-gapped Deployment: Running AI models within a client’s private infrastructure to ensure data security and sovereignty.
• Distillation: A technique where smaller models learn from the outputs of larger, more powerful models, often used by Chinese labs to close the capability gap.
• Democratically Aligned Technology: AI developed by companies in Western, democratic nations, prioritized by regulated industries for trust and security.

---

1. The Valuation and Market Reality

Wall Street is currently evaluating OpenAI and Anthropic for IPOs with valuations exceeding $800 billion. The investment thesis relies on these companies being the "next Google or Microsoft," possessing long-term pricing power. However, this narrative is being challenged by:

• Cracking Pricing Power: The cost of using top-tier models like Claude Opus is significantly higher—up to 9x more expensive—than performant Chinese open-source alternatives.
• Shrinking Moats: The performance gap between American frontier models and Chinese open-source models is closing rapidly. Data from Open Router shows Chinese models grew from 1% of usage in 2024 to over 40% in 2025.
• Compute Constraints: American labs are burning massive capital on infrastructure and power, costs which are passed to consumers. Conversely, Chinese labs, restricted by export controls on Nvidia chips, have been forced to innovate through algorithmic efficiency and smaller, more effective models.

2. The "Trust" Stronghold

Despite the cost advantages of Chinese models, American Frontier Labs retain a critical advantage in regulated industries (defense, banking, healthcare, grid operators).

• Security Concerns: Organizations in these sectors cannot risk using models that may contain "backdoors" or expose sensitive data to foreign entities.
• The Premium Justification: These industries are willing to pay a premium for "democratically aligned" technology that can be deployed in private, air-gapped, or on-premise environments.

3. The Rise of Specialized Competitors

The market is shifting toward a "middle ground"—capable models that are efficient and secure.

• Cohere: Founded by Aiden Gomez, the company focuses on enterprise-grade, secure deployments that run on limited compute (2–4 GPUs). Their revenue grew 6x last year, proving demand for specialized, private AI.
• Nvidia (Neotron): Nvidia is positioning its own open-source models as a neutral, trusted alternative to both Chinese labs and closed-source frontier labs.
• Reflection AI: A startup building American-made open-source frontier models to compete directly with DeepSeek.

4. Strategic Shifts and Industry Perspectives

• The "Efficiency" Wave: Aiden Gomez argues that the next phase of AI adoption will be driven by CFOs optimizing for cost. The industry is moving toward "right-sized" models that deliver high performance without requiring massive, unsustainable infrastructure.
• Elon Musk’s Hedge: Musk merged xAI into SpaceX to backstop the massive capital burn of AI development with the stable, multi-billion dollar revenue of a rocket company, signaling that standalone AI labs may be too risky for long-term sustainability.
• The Role of Open Source: While hyperscalers (AWS, etc.) promote Chinese open-source models for low-sensitivity tasks, the industry is increasingly looking for Western-aligned open-source contributions to maintain a competitive edge.

5. Notable Quotes

• Aiden Gomez on the China threat: "The gap between the two countries... is closing rapidly."
• Aiden Gomez on security: "If you're going to be replacing your entire software engineering team with models... you probably don't want that coming from China because it might be subtly introducing vulnerabilities."
• Aiden Gomez on the future of infrastructure: "There will be much more AI in the world, but those AIs will be using less GPUs to do the same work as they were doing six months ago."

6. Synthesis and Conclusion

The "trillion-dollar" valuation of frontier AI labs is predicated on the assumption of sustained, unchallenged dominance. However, the market is bifurcating:

1. Commoditized AI: Low-sensitivity tasks are shifting toward cheap, efficient, and increasingly capable Chinese open-source models.
2. Secure/Regulated AI: High-trust, critical infrastructure tasks are moving toward private, on-premise deployments provided by companies like Cohere and Nvidia.

The primary takeaway is that the "frontier" is no longer just about model size; it is about efficiency, security, and trust. OpenAI and Anthropic face a dual threat: Chinese labs are winning on cost and efficiency, while specialized American competitors are winning on security and enterprise integration. The future of AI economics will likely favor those who can deliver high-utility models with a smaller, more efficient compute footprint.