
Deep-tech software that controls systems that cannot fail
Software that acts, decides or enforces in the systems the world depends on — and the hardware that serves it.
Where we invest
Agents are repricing the application layer. Value is moving to software that runs machines, grids, vehicles and sensors — often on hardware it was designed with — and to the controls that let AI act safely. Regulation is moving the same way: audit and safety evidence is now something customers have to buy.
The control layer around AI
Authorisation, reliability, audit and human oversight for AI that moves money, treats patients or runs plants.
The application layer
Horizontal SaaS and AI wrappers, the layer agents are commoditising.
Where software meets the physical world
Edge, firmware, on-device inference, digital twins and vertical data, now including the hardware that serves the software.
Seven areas
Where software controls systems that cannot fail, and where the Critical ecosystem gives our founders a genuine, unfair advantage.
Automation & Industrial
Software that runs plants, robots and production lines
Defence & Dual-Use
Software for sovereignty, resilience and national security
AI Control Plane & Assurance
Authorisation, reliability and audit for AI that acts in the real world
Cybersecurity
Securing software, infrastructure, devices and now autonomous agents
Energy & Critical Infrastructure
Software that controls grids, utilities and the infrastructure behind them
Mobility & Aerospace
Autonomy and the systems that move people and goods
Medical & Bio-Integrated
Where regulation is hard and silent failure is unacceptable
What we look for
It starts with exceptional, domain-expert founders building at the edge of science and engineering, who fit the Critical ecosystem. Then every company has to pass four tests.
Deep-tech software at the core
A software moat, above or below the application layer.
Software that controls
It acts, decides or enforces. Software that only visualises, reports or predicts is out.
Critical to the customer
When it fails, the customer loses safety, the mission or money.
Scales on software economics
€0.5–3M takes it to scale, while others build the hardware.
How far hardware can go
Control, sensing and edge intelligence increasingly depend on hardware, so we no longer reject it by category. What matters is that the software stays the scalable core and the hardware makes its moat stronger.
Fit
- Pure software
- Software on off-the-shelf hardware
Fit if all four tests hold
- Hardware-enabled software
- Co-designed hardware/software systems
- Autonomy platforms, by exception
No fit
- Hardware is the product
- Silicon and novel physics
- Production is the business
In any area, we also pass on services- and integration-led models, consumer apps, marketplaces and token models, and direct competitors of our portfolio companies.

We commit early, write meaningful cheques, and stay close for the whole journey.
Why it lasts: seven moats
We back businesses whose advantage compounds rather than erodes as AI accelerates. Every company we back holds at least one of these moats, with evidence — and the more it holds, the deeper our conviction.
Mission-criticality
Buyers reject good-enough and don't switch: replacing the product would put safety, uptime, the mission or money at risk.
Proprietary data
Unique operational, clinical and sensor data an agentic model can't scrape, synthesise or copy — and that only deepens with every deployment.
Hardware / software co-design
Software co-designed with the hardware it runs on — at the edge, in firmware, on the device — the kind of depth you can't fake with an API call.
Regulatory certification
Safety and compliance certifications, and the audit trail behind them, that took years to earn and that rivals must reproduce.
AI control plane
The same pressure forcing enterprises to deploy AI forces them to govern it. Control over what agents may do becomes worth more as agents scale.
Closed-loop control
The software acts inside the customer's process — it corrects, stops or authorises — so removing it means re-engineering the operation.
Owned sensing & field base
Sensors the company owns in the field give it data no rival can scrape, and an installed base that drives recurring revenue.
Our process
Screening
Initial fit against the thesis: the four tests and at least one moat.
Deep dive
Technical and business diligence — we validate the technology, not just the deck.
Strategic plan
We shape the plan and the round alongside the founders before we commit.
Investment Thesis