A once-in-a-generation shift is underway. The way large companies run their technology — the plumbing that keeps banks, retailers, hospitals, and logistics giants operating — is being redesigned around AI. The businesses that supply the new plumbing will be among the best investments of this decade.
When a customer service agent answers a simple question today, it doesn't run one piece of software. It triggers 15 to 40 internal operations — checking identity, fetching data, verifying permissions, logging, monitoring.
When AI traffic grows 6×, the hidden infrastructure load grows 30 to 60×. That explosion is not served by new AI chips or new AI models. It is served by the layer of software that has always run underneath — and must now run far harder than before.
Nearly every large company has started using AI. Almost none have finished rebuilding their infrastructure to support it at scale.
That 82-point gap represents years of infrastructure investment — and the companies that supply the tools will grow with every enterprise deployment.
In a single month, public markets separated technology companies into two categories. Those AI runs on. And those AI replaces. Our proprietary framework predicted this separation with 89% accuracy across 70+ companies.
Project management, dashboards, helpdesks — these are applications. AI is learning to do the same job for a fraction of the cost.
Networking, data streaming, search — these are infrastructure. Every AI agent runs on layers like these. More AI means more demand.
When a city booms, the landlords do well regardless of which businesses come and go. We invest in the landlords of enterprise technology — the software infrastructure that every AI application must run on, regardless of which AI wins.
We don't bet on which AI model wins. We invest in what all of them run on — and what every enterprise must buy to deploy any of them.
These are the tenants. When AI advances, they face existential disruption.
80%+ gross margins. Near-zero marginal cost. The most durable business model in software.
Six layers of software sit beneath every enterprise AI deployment. The figures below show how demand at each layer scales when overall AI traffic increases sixfold.
| Layer | Role in Plain English | At 6× AI Traffic | Demand Multiplier |
|---|---|---|---|
| Identity & Authorization | The security guard. Every agent action must ask "am I permitted to do this?" | 8–15 permission checks per request | ×8–15 |
| Data & Search | The library. AI agents query and retrieve knowledge before every response. | Every AI response = multiple lookups | ×12–30 |
| Observability & Monitoring | The flight recorder. Logs every event so teams can diagnose and improve. | Fan-out creates exponential log volume | ×20–50 |
| API Gateway | The front door. Routes every inbound request to the correct internal service. | Each request fans into dozens of sub-calls | ×6–8 |
| Agent Execution Environments | Isolated rooms where AI agents run tasks safely, separated from production. | Every agent session = isolated compute | ×6–10 |
| Networking & Service Mesh | The road system. Moves data between services reliably and securely. | Service-to-service traffic scales non-linearly | ×6–8 |
20+ years building enterprise-grade production infrastructure. We have operated the same tools we evaluate — at scale, under real business pressure.
An eight-factor composite scorecard, validated externally: it predicted 89% of positive outcomes in the February 2026 public market separation.
Every company receives a 1–10 AI-replaceability score. We invest exclusively in companies scoring 1–3: infrastructure AI runs on, not infrastructure AI displaces.
We source from open-source communities and major industry events, seeing companies 12–18 months before institutional investors — when valuations are lowest and upside is greatest.
Sourcing, scoring, diligence, and portfolio monitoring are automated. We cover 10× more ground than a traditional team of equivalent size.
Dual lens: Every company is evaluated through two simultaneous lenses — how enterprises actually procure and deploy software, and what runs under the hood technically. Most investors have one or the other. We require both.
Three companies holding the highest composite scores. Immediate candidates.
14 companies across identity, monitoring, storage, and AI deployment layers.
Every time an AI agent does anything — reads a file, sends a message, books a meeting — it must confirm it has permission. Cerbos is the service that answers that question. In a high-volume AI deployment, this check runs 5–15 times per user request.
Every AI response that draws on company knowledge — customer records, documents, product data — requires a fast search query first. Typesense handles that query. Bootstrapped and profitable since 2015 with a 25,000-strong developer following and no enterprise tier yet.
When AI agents write or run code, enterprises require an isolated environment — so a mistake cannot damage live systems. Daytona provides that isolation. A new category with no dominant incumbent, priced per session hour.
Usage-based revenue grows 5–8× over three years as AI traffic multiplies. Infrastructure software exits at 15–25× revenue versus 8–12× for application software. Recent comparables: HashiCorp $6.4B (IBM), Wiz $32B (Google), Kong $1.4B.
Even without AI tailwinds, the pipeline is driven by secular forces: cloud adoption still rising, legacy modernisation ($22B → $52B by 2030), and enterprise software refresh cycles. Either way, we win.
| Question | Our Position |
|---|---|
| Won't Amazon or Microsoft simply build this? | 91% of enterprises self-host for portability and regulatory compliance. Vendor-agnostic, open-source tools consistently win over proprietary cloud alternatives. |
| Can open-source companies generate meaningful revenue? | Every first- and second-tier company has an enterprise revenue path or has already started it. Datadog ($18B), Elastic ($8B), and HashiCorp ($6.4B) are the established proof. |
| What if AI simplifies infrastructure over time? | Our investment horizon covers the current expansion phase. The floor case still delivers 3–5×. Infrastructure simplification at the depth we target takes decades. |
| What if the underlying technology platform changes? | Over 70% of our pipeline is platform-agnostic — running across all major enterprise deployment models. We invest in layers that sit below any specific platform. |
We are actively raising Fund II and evaluating co-investment structures. Focus: Seed and Series A infrastructure software companies — the earliest point of maximum leverage, before traditional investors have the domain depth to assess the opportunity.