For the past several years, the AI industry has operated under a simple assumption that the company with the best large language model (LLM) wins. Every major release has been judged by benchmark scores, reasoning capabilities, context windows, and multimodal performance. I believe this assumption will eventually break down.
My prediction is that frontier AI models will become increasingly commoditized. Distillation techniques, open-weight releases, algorithmic improvements, and the rapid replication of architectural breakthroughs will make it increasingly difficult for any single company to maintain a durable technical advantage. As model capabilities converge, the competitive moat will shift away from the model itself toward everything surrounding it. More specifically, I believe frontier AI companies will increasingly pursue regulation as a competitive advantage. Not simply because AI requires oversight, but because regulatory frameworks can create barriers to entry, restrict competitors, and preserve market access for trusted providers.
Regulation will become the primary moat.
Why Models Will Become Commodities
The history of technology suggests that foundational capabilities eventually become commoditized. Computing power, cloud infrastructure, databases, and telecommunications networks all followed a similar pattern: early leaders gained advantages through technical superiority, but competition eventually shifted toward ecosystems, distribution, integration, and cost efficiency. AI is likely to follow the same trajectory. Open-weight models, distillation techniques, and algorithmic improvements are rapidly reducing the gap between frontier and smaller models, allowing competitors to reproduce much of the capability of leading systems at a fraction of the cost. This does not mean all models will become identical, but the advantage of having the absolute best model will likely diminish as more systems become capable enough for the majority of commercial use cases. As intelligence becomes increasingly available as a commodity, the competitive advantage will shift toward the products, platforms, and ecosystems built around it.
The Competitive Landscape
Model Routing and the Rise of Model-Agnostic Applications
One of the strongest signals that models are becoming infrastructure is the rise of model routing. Rather than sending every request to the most capable/expensive LLM, enterprises are increasingly routing tasks to the model that provides the best balance of performance, latency, reliability, and cost. A simple summarization task may be handled by a low-cost open-weight model, while complex reasoning or coding requests may be directed toward frontier models.
This changes the economics of the AI industry.
Applications increasingly become model agnostic. Developers can abstract away the underlying model provider and dynamically select whichever model performs best for a specific task. Companies such as Cursor demonstrate this shift with Composer 2 using Kimi K2.5. Users are primarily loyal to the application experience, not necessarily the underlying model powering it. As CNBC has reported, model routing encourages customers to view models as interchangeable infrastructure rather than differentiated products.
The more successful model routing becomes, the more it accelerates commoditization.
Pricing Pressure
If multiple models provide comparable performance, price becomes a powerful differentiator.
Open-weight models can often be deployed at significantly lower cost than proprietary APIs. Distillation reduces the infrastructure required to achieve high performance, allowing smaller teams to compete with organizations that have invested billions in training large models.
The premium pricing model of frontier companies depends on maintaining a meaningful capability advantage. If customers can achieve similar results using cheaper alternatives, the economic value of proprietary models declines.
Capital Becomes a Weaker Moat
One of the strongest advantages frontier labs have today is access to enormous amounts of capital. Training frontier models requires billions of dollars in compute, infrastructure, and research talent. However, this advantage is weakening over time. Algorithmic improvements and distillation techniques allow smaller teams to reproduce much of the capability of frontier models at a fraction of the cost, making capital alone a less durable competitive advantage.
The largest training runs may continue producing the most advanced models, but they may not automatically produce the strongest businesses.
Intellectual Property Becomes Less Reliable
Intellectual property protection is another potential moat that may weaken. Patents and copyright protections can provide advantages in jurisdictions that recognize and enforce them. However, AI development is increasingly global, and intellectual property protections have historically had limited influence across geopolitical boundaries.
If competing countries can replicate leading techniques outside Western legal frameworks, IP is not a reliable source of long-term advantage.
Where the Remaining Moats Exist
If technical superiority, capital, and intellectual property become weaker advantages, AI companies will compete through other forms of differentiation.
Product and User Experience
Users rarely buy intelligence itself. They buy software that solves problems.
The winners may not be companies with the smartest model, but companies that create the best user experiences, workflows, integrations, enterprise controls, and automation capabilities.
The application layer may capture much of the economic value created by AI.
Network Effects and Ecosystems
The companies with the largest ecosystems of users, developers, enterprise integrations, and applications will have significant advantages. Once organizations build workflows around a platform, switching costs become substantial. This is the same dynamic that benefited operating systems, cloud providers, and productivity software.
Vertical Integration
AI companies are increasingly moving beyond simply selling APIs. The goal is not just to provide the intelligence layer, but to own the customer relationship by building complete products that combine models, applications, workflows, and user experiences.
Anthropic’s launch of Claude Design is an example of this shift. Rather than simply providing Claude as a foundation model, Anthropic is building a product experience where users collaborate with Claude to create designs, prototypes, presentations, and other visual assets directly within its ecosystem.
This represents a broader industry trend: frontier AI companies are attempting to capture more of the application layer instead of allowing third-party companies to capture all of the value built on top of their models.
The strategic objective is to become the destination where users work and not simply the infrastructure powering someone else’s product.
Regulation as the Ultimate Moat
My prediction is that regulation will become the primary moat protecting U.S. frontier AI companies. As models converge and lower-cost competitors close the technical gap, incumbent AI companies will increasingly seek advantages that cannot be replicated through engineering. Regulation provides that advantage. The public justification will focus on national security, data protection, and foreign influence. However, many of these risks can already be addressed through deployment controls such as self-hosting or using trusted hyperscalers with enterprise security controls.
The underlying incentive is economic. With Chinese open-weight models achieve comparable performance at significantly lower costs, they undermine the pricing power and market share of U.S. frontier companies. Supporting regulatory frameworks that restrict foreign models and increase compliance barriers creates a powerful competitive advantage.
Politicians also benefit from framing domestic AI companies as strategic national assets. The result is regulatory capture: policies justified by security concerns that protect incumbent companies from lower-cost competition.
AI will follow the same pattern seen in other strategic industries, where regulation determines which companies can participate.
A Smaller Addressable Market Than Investors Expect
None of this suggests U.S. frontier AI companies will fail. Companies such as OpenAI, Anthropic, and Google will likely remain enormously valuable. They possess exceptional research talent, strong brands, enterprise relationships, and strategic positions within regulated markets. However, I believe their total addressable market will be smaller than current valuations assume. Many investors appear to be pricing in a future where proprietary frontier models dominate global AI usage. But if open-weight models capture much of the commercial market because they provide sufficient capability at dramatically lower cost, the opportunity for proprietary model providers becomes more concentrated.
The future will landscape will be a segmented market where different ecosystems dominate different regions and industries.
Conclusion
As model routing, distillation, and open-weight models make AI increasingly commoditized, the competitive advantage will shift away from raw intelligence toward ecosystems, distribution, and regulatory capture.