The first wave of generative AI — dominated by consumer chatbots and image generators — was the opening act. The real, durable transformation will unfold inside enterprises, where large language models are poised to redefine workflows, automate complex reasoning tasks, and unlock unprecedented productivity across every function.
We are now entering the second act, and it is both more complex and more consequential than what came before. At Nahkoda, we have spent considerable time mapping this landscape, and our conclusions are both optimistic and nuanced.
The "Last Mile" Problem
For enterprises, the critical challenge is not access to powerful models — that is now a commodity. The real difficulty lies in the last mile: connecting those models to proprietary data, legacy software stacks, and stringent security protocols in a way that is reliable, auditable, and actually useful.
We are particularly focused on the infrastructure layer that makes this possible:
- Data Privacy & Security — Solutions that allow enterprises to train and run models against sensitive data without exposure to third-party providers.
- Model Fine-Tuning & Specialization — Platforms that allow companies to adapt general-purpose foundation models for specific vertical use cases — legal, medical, financial, and beyond.
- Orchestration & Integration — Tools that manage the complex flow of context between LLMs and existing enterprise systems, from CRMs to ERPs.
"The next multi-billion dollar AI companies will be those providing full-stack, vertical-specific solutions. A generic chatbot is useful. An AI that understands pharmaceutical compliance is indispensable."
Where We See Opportunity
We believe the highest-value AI companies will be those that achieve both technical depth and domain specificity simultaneously. The vertical AI thesis — purpose-built LLMs for regulated industries — is where we are spending much of our diligence time in 2025 and into 2026.
Healthcare, legal, and financial services present the most compelling opportunities. These industries have the highest data density, the greatest regulatory complexity, and — critically — the highest willingness to pay for accuracy and compliance. An AI system that reduces medical coding errors by 30% is not a nice-to-have; it is a revenue line item.
Risks We Are Watching
The category is not without risk. Commoditization pressure from foundation model providers is real. The moats must be built in data, distribution, and vertical integration — not in model weights. Companies that can aggregate proprietary data pipelines and maintain human expert loops will be the most defensible.
We are also watching the regulatory environment closely. The EU AI Act and emerging US frameworks will create compliance overhead but also competitive advantages for incumbents who get ahead of them.
Our Thesis in Brief
We are looking for founders with deep domain expertise who understand their verticals from the inside — not just AI engineers building tools for markets they don't know. The best enterprise AI companies we have seen are founded by people who have lived the problem they are solving.
If you are building at the intersection of deep domain knowledge and AI-native architecture, we want to talk.