01
Capability is real
Vision-language-action models, world models and accelerated simulation have made competent physical behaviour reproducible in the lab. Reliable deployment has not followed.
Thesis
Paretia backs the constraints whose removal unlocks the system.
The position, the test we apply to it, the fund built to finance it, and the things we will not fund — stated plainly enough to be held against us.
Why now
The platform layer arrived before the tooling did.
02
Capital is concentrating
Record funding is flowing into large platform and system rounds. Concentration at the top is precisely what leaves the enabling layer underfunded and mispriced.
03
Evaluation lags capability
The field cannot yet say reliably whether a system is safe, better, or ready. Measurement, adaptation and operations are the constraints a seed cheque can still move.
The test
An investment should score highly across all seven dimensions.
A high score on criticality with a low score on capture is a research project. A high score on capture with a low score on leverage is a consultancy. We need all seven.
| Dimension | Investment question |
|---|---|
| Criticality | Does the constraint materially limit reliability, cost, speed, safety or data efficiency? |
| Frequency | How often does the customer actually encounter the constraint? |
| Leverage | Can one solution serve multiple systems, tasks or verticals? |
| Depth | Is the product integrated deeply enough to resist replacement? |
| Learning | Does usage create proprietary data or a benchmark advantage? |
| Capture | Can the company retain pricing power and strategic importance? |
| Capital fit | Can seed capital prove technical and commercial leverage? |
Fund I
Sized to finance evidence, not spectacle.
Fund I has not closed. Everything below marked aspirational is a plan we intend to be judged against, not a fact. Our ownership needs should never force an oversized round.
| Item | Plan | Status |
|---|---|---|
| Target fund size | $40M | Target |
| Stage | Pre-seed and seed | Fixed |
| Initial cheque — pre-seed | $0.6M – $1.2M | Aspirational |
| Initial cheque — seed | $1.2M – $1.8M | Aspirational |
| Entry ownership | 6% – 9% | Aspirational |
| Retained after modeled dilution | 5% – 7% | Modeled |
| Role | Lead or co-lead 40% – 50% of rounds | Plan |
| Milestone financed | 18 – 24 months to technical superiority, repeatable design-partner use, cross-system validation or an institutional Series A | Plan |
Exclusions
Discipline is mostly saying no.
The tell we avoid: a deck that says “multi-vertical” but cannot name a single urgent customer or a falsifiable performance metric.
Careful with full-stack robots. They qualify only when the enabling asset — the data engine, the evaluation system, the manipulation primitive — becomes reusable across systems rather than locked to one platform.
- Generic AI infrastructure with no physical-world dependency
- Commodity dashboards, labeling services, system integration
- Open-source projects without a credible capture mechanism
- Thin wrappers around one vendor's robotics model
- Undifferentiated robot hardware
- Consumer gadgets and novelty robotics
- Passenger autonomy
- Foundation-model races requiring mega-fund capital
- Multi-vertical stories masking a lack of customer urgency
Own a constraint on this map?
We would rather read one paragraph about the bottleneck you control than a twenty-page deck about the market it sits in.