[Home](/)
[Own the means of production](/essays/own-the-means-of-production)
[Rent models, own judgement](/essays/rent-models-own-judgement)
[Token Fatigue](/essays/token-fatigue)
[The Operator is the spec](/essays/the-operator-is-the-spec)
# Parallel Partners
**Based on Conviction. Built for Outcomes.**
---
## Mission
Parallel makes the mid-market agentic-first.
---
## Vision
The winners of the next decade won't be new companies. They'll be existing businesses with real customers and real revenue that get rebuilt in time. That's the market we work in.
Most operators layer AI on top of the business. The ones that win rebuild around it, with a live model of the operation at the center and people sitting on top of it rather than inside it. Judgement, taste, and customer trust stay human, and everything else becomes addressable by agents.
At PP we believe the advantage isn't headcount. It's the speed of the loop between signal and action. This is the agentic economy taking shape: systems that run on their own, around the clock. The businesses rebuilt for it move at machine speed. The rest become suppliers to the ones that did.
---
## Thesis: the AI supercycle is mispriced at the application layer
The economics of the AI stack are skewed. Research by Apoorv Agrawal (Stanford MS&E 435, _The Economics of Generative AI: Two Years Later_) shows **93 percent of value still sits in the semiconductor layer**.
| Layer | Share of profit | Notes |
|---|
| Semiconductor (NVIDIA-dominated) | ~70% revenue / ~79% gross profit | One-player game. $350B extra revenue in two years moved the share by almost nothing. |
| Infrastructure (hyperscalers, compute, model labs) | ~14% of profit | Margins squeezed by energy costs. |
| Application (apps, agents, organisations) | ~7% of profit, up from 3% | Where the value lands eventually. |
The cloud cycle ran the other way. Apps captured roughly 70 percent of profit, but it took fifteen years. AI may take longer. Compute costs are still not falling fast enough, and the application layer is still failing to hold value.
The split flips when the application layer builds **agentic-first companies**: strategy, product, and agents running as one system, not loose projects next to each other.
That's where we put our chips. We do not build agents next to existing work processes. We replace or rebuild those processes alongside the operator, with our own capital and our own risk in the outcome. The mid-market segment is large enough to build a portfolio in and small enough to move fast. That is the window.
---
## What is Parallel Partners
Parallel Partners (PP) is an agentic venture-build firm. Based in Amsterdam, founded in 2025. We rebuild mid-market companies into agentic-first operations. We do not add AI on top of an existing process. We replace or rebuild the process alongside the operator, with our own capital and our own risk in the outcome.
We do three kinds of deals: **Transform** (we acquire the business), **Replace** (we rebuild a single work process), and **Co-build** (we build a new venture together in a joint-venture entity). We only earn through equity or revenue-share.
We sit between management consultancy and venture-building. We call it agentic venture-build.
---
## Three engagement modalities
| Modality | What it is |
|---|
| **Transform** | We acquire an existing mid-market company with investors (80 to 100 percent) and rebuild it agentic-first. |
| **Replace** | We rebuild one work process inside an existing business. We earn through revenue-share or equity. |
| **Co-build** | We build a new venture with a founder inside a joint-venture entity. |
---
## Key differentiators
- - We've done this before. We grew and sold businesses.
- - We share the risk. Equity and revenue-share.
- - We rebuild how the work happens, not who does it.
- - Agents are core infrastructure, not a tool added on the side.
- - Three ways in: transform what you have, replace a process, or co-build something new.
- - Your data and compute stay in Europe.
---
## What we are not
- - Not a consultancy
- - Not a vendor selling AI employees
- - Not a shop billing by the hour
- - Not a tool or SaaS provider
- - Not a fund that only writes cheques
---
## How an engagement runs
| Stage | What happens |
|---|
| Pre-intake | 60 to 90 minutes with a PP partner. Fit assessment with the client. |
| Intake | Understand the problem and the business context. Pick a modality: Transform, Replace, or Co-build. |
| Refinery | Translate requirements into an agent specification. |
| Validator | Confirm value, data, and risk are clear enough to build. |
| Foundry | Production build on the PP stack. |
| Live | Tool or product in production, monitored. |
| Metrics | Operational and business metrics on a real-time KPI dashboard. |
---
## Us
We are operators, not advisors. A small team of seasoned entrepreneurs who have built and sold companies, market-makers who priced risk at machine speed, and commercial leaders who have carried real P&L. We have sat in the owner's chair, so we build for it, with our own capital and equity at stake.
---
## Facts at a glance
| Name | Parallel Partners (PP) |
|---|
| Category | Agentic venture-build firm |
| Founded | 2025 |
| Headquarters | Amsterdam, The Netherlands |
| Geography served | EU & US |
| Engagement modalities | Transform, Replace, Co-build |
| Compensation | Equity and revenue-share |
| Primary market | Mid-market companies |
| Data residency | EU |
---
## Contact
---
## Essays
- - [Own the means of production](/essays/own-the-means-of-production): The cost of intelligence is collapsing for every company at once, so on its own it is no advantage. The edge is your own proprietary knowledge built into systems you control. And real enterprise AI safety is keeping that control, over your data, your models, your compute, and who gets to see your alpha, because the labs that own the model layer have every incentive to move up it and ship what you built as their next product.
- - [Rent models, own judgement](/essays/rent-models-own-judgement): The model is fungible because every competitor rents the same one. The advantage is the Loop, the captured human judgement that compounds and survives every model release. We build it for the operators who never could.
- - [Token Fatigue](/essays/token-fatigue): Enterprises bolted AI onto broken processes and the bills came due without the returns. Token Fatigue is the reckoning. The fix is a business rebuilt around the intelligence, not a cheaper token.
- - [The Operator is the spec](/essays/the-operator-is-the-spec): The mid-market runs on a person, not a process map. Horizontal AI configures around the process and cannot reach it. A vertical build is shaped around the operator and compounds on their P&L.
---
_Last updated: 2026-07-27_
