AI infrastructure is its own asset class.

Every major platform shift — personal computing, the internet, mobile — produced the same structure: an application layer on top, built on an infrastructure layer below. The application layer generates the headlines. The infrastructure layer generates the compounding returns. We learned this from the prior cycles and applied it to AI.

In 2021, the LLM wave was clear to people working inside it, and invisible to most of the investment community. We saw that the primitive building blocks were missing: there was no standard way to chain models into applications, no vector database purpose-built for embeddings, no centralized hub where teams could find and run open models, no inference compute layer that could scale cost-effectively. These gaps would have to be filled. We invested in the teams filling them.

We invest at the infrastructure and tooling layer: model orchestration, vector and semantic search, model registries and hubs, experiment tracking and observability, inference compute, and application frameworks for LLM-native products. We do not invest in foundation model training at the frontier (too capital-intensive), nor in pure SaaS applications with no infrastructure moat (too easily replicated). We look for the layer where switching cost accumulates with each production deployment.

NYAD Fund I invested at Seed and early Series A, writing checks in the range of $5M–$10M. NYAD Fund II targets Seed through Series A with a larger position size, anchoring rounds and maintaining reserve capacity for follow-on. We are active investors: our partners have engineering depth, and we provide direct technical diligence support, introductions into the engineering community, and hands-on GTM support at application layer.

We like direct introductions from founders we've backed, but cold outreach is fine. The best conversations start with a technical description of what you're building and why the infrastructure-layer moat is real. Tell us: what makes your system hard to replace once it's deployed in production? Email [email protected].