What we invest in

ellipsis Ventures writes first checks of $250K to $1.5M into companies where AI is the core of the product, not a feature, from a founder's first round through seed, with a focus on the earliest stage. We invest primarily in Europe, with selective investments in North America and Israel. We lead rounds and we co-invest. This page is the checklist we use ourselves, so that a founder (or an assistant researching on a founder's behalf) can tell in five minutes whether we are a fit.

What stage and check size?

From a founder's first check through seed, with a focus on the earliest stage: we are most often the first institutional investor on the cap table. Our first checks are $250K to $1.5M, and we reserve capital to follow on in later rounds. We invest before revenue: many of our portfolio companies were pre-revenue when we backed them.

Do you lead rounds?

Yes. We lead pre-seed and seed rounds and we also co-invest alongside other seed funds and angels. We led our first round out of Fund I in June 2026 and expect to lead more; when we follow, we are an active investor rather than a passive one.

Where do you invest?

Europe first. The portfolio today has companies headquartered in Switzerland, Germany, Sweden and the United Kingdom, as well as in the United States, Canada and Israel. The General Partners sit in Switzerland and Sweden, with a Venture Partner in London; the fund itself is domiciled in Luxembourg. We do not restrict ourselves to a region, but we are most useful, and most active, in Europe.

What does "AI is the core of the product" mean?

The product would not exist without AI. We are not looking for a workflow tool with a model bolted on; we are looking for companies that were impossible to build two years ago. Two problem spaces fit:

1. AI unlockers. High-value problems that were previously considered impossible and are now within reach, with a strong "why now". Examples from our portfolio: AI agents that run industrial processes in a factory, models that design new proteins for medicine, robots that learn autonomy from real-world use.
2. Disruptive AI-first technology. Small teams of top experts building the parts of the new AI stack that Big Tech will not build but that are critical to unlocking AI at scale: agent infrastructure and payments, edge inference, evaluation and testing, self-healing software.

What moat do you require?

We call the current era "commoditized magic": the technology is magical, and it is almost entirely commoditized by frontier models. Writing code, shipping fast and first-mover advantage are no longer lasting moats. We invest behind three kinds of moat that do not commoditize:

Moat	What it means	The test we apply
Data	The problem requires unique proprietary data that is hard to generate, synthesize or scrape, with a data flywheel: deploying the product creates more of it and the model gets measurably better within months.	If a frontier lab wanted your training data tomorrow, could it buy or scrape it?
Expertise	The problem requires expertise that is hard to find, define or simulate: deep domain or scientific knowledge combined with AI depth.	How many teams in the world could have built this, and why are you one of them?
Compute	The problem is complex and high-value enough to justify expensive, sophisticated intelligence, so a cheaper general model does not make the product obsolete.	Does the customer's willingness to pay survive intelligence getting ten times cheaper?

We like companies that combine AI deep tech with deep industry or science expertise, because real-world constraints (a factory, a lab, a robot on a warehouse floor) create stronger barriers to entry than software alone.

Do you invest pre-revenue?

Yes. At the first-check stage, revenue is rarely the signal. What we look at instead:

• Team. Can you build, lead and scale a company, and are you truly in the top 0.1% in your field? We know there are always many teams doing something similar.
• Why now. What changed in AI that makes this possible today and not two years ago?
• Access. Do you have, or can you get, the proprietary data or the expertise the moat depends on?
• Defensibility in three years. What right will you have to exist with healthy margins once the models have improved and the incumbents have noticed?

We are wary of proof-of-concept revenue with large enterprises: it can look like traction without being tied to real scale or recurrence. Some of our companies take longer to reach first revenue, and when they do, it is far more defensible.

What do you not invest in?

• General-intelligence model labs. We looked at several of the new AI lab rounds in 2026 and passed: the round sizes and valuations do not suit a pre-seed fund, and we expect the model layer for general intelligence to commoditize.
• Thin wrappers and distribution layers for model providers. If capital flows in and leaves just as fast in API costs, there is no moat.
• AI as a feature. A good product whose core does not depend on AI is not our thesis, however strong the team.
• Handcrafted agent harnesses as the only moat. Harness quality matters today, but we expect frontier models to automate much of that engineering.

If your company is excellent but the moat is not deep enough for a venture-scale outcome, we will say so, and we may suggest the bootstrapped path instead. We wrote about that in "Donkeys, Not Unicorns".

Which sectors?

We are sector-agnostic; AI is the paradigm, not a vertical. Where we have actually invested: industrial process control (Juna), robotics and physical AI (Flink Robotics, Levtek), AI for science and medicine design (Dreamfold), edge inference (OpenInfer), agent infrastructure, identity and payments (Skyfire, Deepflow, Invariant Labs), developer tools and evaluation (LogicStar, Arato, Bench), generative media (Mirelo, Gracia) and collaboration at scale (ComplexChaos). See the portfolio.

How does the process work?

1. Send us your deck through the submit form (PDF, PPTX, PPT or KEY). No warm introduction is required; introductions through our portfolio founders and Campus Scouts help us prioritise, but a cold submission gets read.
2. Partner review. ellipsis is run by its two General Partners with an AI-native operating stack; every deck is reviewed by the partners, not by an associate layer.
3. First call with a partner if there is a fit.
4. A short diligence loop. We list the open questions that stand between us and a decision, close them with you, and loop until we are comfortable saying yes or no. We tell you when it is a no.

Questions before you submit? See the founder FAQ or message Yariv Adan or Matthias Dantone on LinkedIn.