The artificial intelligence landscape is rapidly evolving, with large language models becoming the foundation for countless applications. As someone who closely follows tech developments, I believe we’re witnessing just the beginning of how AI will reshape business models and create new opportunities for entrepreneurs.
Jensen Huang, the visionary co-founder of Nvidia, recently shared insights about where the real value in AI businesses will emerge. His perspective offers a compelling roadmap for entrepreneurs and investors looking to build lasting AI companies.
Beyond the Foundation Models
The current AI ecosystem typically follows a straightforward pattern: large companies create foundation models and offer them through APIs, while smaller businesses build applications on top. This structure seems simple, but I think it misses the crucial middle layer that Huang highlights.
What’s becoming clear is that we’ll likely see “a small handful of fundamental large models” that serve as the base infrastructure. These models will be massive, expensive to train, and provide general capabilities. But the real differentiation won’t happen at this layer.
Instead, the most significant opportunities lie in what I call the “transformation layer” – companies that take these foundation models and adapt them for specific domains. This isn’t just about simple fine-tuning but involves deeper specialization.
The Transformation Layer: Where Value Lives
The middle layer Huang describes will consist of companies that:
- Take existing large models and transform them for specific industries
- Develop specialized tuning techniques beyond basic fine-tuning
- Create domain-specific versions optimized for particular use cases
These specialized models will serve distinct purposes – from medical applications to computer interactions to personalized assistants. The companies that master this specialization process will create lasting value because they’ll own the critical link between raw AI capability and practical applications.
I predict this middle layer will become the most competitive and innovative space in the AI economy. While foundation models may become commoditized over time, the expertise to transform them for specific domains will remain valuable.
Building an Enduring AI Business
For entrepreneurs looking to build lasting AI companies, focusing on this transformation layer offers several advantages:
- Lower capital requirements than building foundation models from scratch
- Ability to create true differentiation through domain expertise
- Potential for strong defensibility through specialized data and tuning methods
- Direct alignment with specific customer needs and use cases
The companies that succeed will combine deep understanding of specific domains with technical expertise in model adaptation. They’ll create value not by owning the base models but by transforming them into something uniquely suited for particular applications.
This approach also addresses one of the biggest challenges in AI today: how to build something defensible when the underlying models are increasingly accessible. The answer lies in specialization and adaptation rather than trying to compete with tech giants on general-purpose AI.
The Path Forward
As AI continues to mature, I expect we’ll see a flourishing ecosystem of these middle-layer companies. They’ll become the connective tissue between raw AI capabilities and practical applications, creating enormous value in the process.
For investors, identifying the teams that can execute this strategy effectively will be key. The winners won’t necessarily be those with the most advanced AI research but those who can best adapt existing models to solve real problems in specific domains.
The future of AI business isn’t just about building bigger models – it’s about making them more useful, more specialized, and more aligned with specific needs. That’s where the enduring value will be created.
Frequently Asked Questions
Q: Why won’t most companies build their own foundation models?
Building foundation models requires enormous computational resources, specialized talent, and massive datasets. Most companies will find it more practical and cost-effective to leverage existing models and focus on adapting them for specific applications rather than competing directly with tech giants who have already invested billions in this area.
Q: What skills will be most valuable for entrepreneurs in the AI “middle layer”?
Success in the middle layer will require a hybrid skill set: deep domain expertise in a specific field (like healthcare, finance, or education), technical knowledge of model tuning and adaptation techniques, and business acumen to identify valuable use cases. Teams that combine these capabilities will be best positioned to create lasting value.
Q: How will companies in this middle layer protect their competitive advantage?
Competitive advantage will come from proprietary tuning techniques, specialized datasets for adaptation, domain-specific optimizations, and the accumulated knowledge of how to transform general models for particular applications. The intellectual property won’t be in the base models themselves but in the transformation process and resulting specialized models.
Q: What types of specialized AI models might we see emerge in this middle layer?
We’ll likely see models specifically optimized for medical diagnosis and treatment planning, legal document analysis and contract creation, educational content personalization, financial analysis and risk assessment, scientific research in specific disciplines, and many more. Each will build on foundation models but add layers of specialization that make them uniquely valuable for their target applications.
