Elisabeth Ziv on Backing Frontier Tech and Building Europe's Next Infrastructure

Elisabeth Ziv is an early-stage investor on the investment team at High-Tech Gründerfonds (HTGF), Germany's largest and most active pre-seed investor, with over €3 billion under management and more than 800 companies backed across AI & Software, Industrial Tech, and Life Sciences.

Before HTGF, Elisabeth worked across a range of industries and companies – starting out at an early-stage startup before moving into strategy consulting and the tech M&A team at Accenture, then on to Delivery Hero, Allianz, and Redstone VC. But her real thread has always been building communities and working with early-stage founders. Back at university she founded Imperial Elevate, a student-run organisation of 1,500 members that supported and match-made founding teams and spin-offs at Imperial. Later, she started She's Got CapTable, working to get more women onto European cap tables. Wherever she went, her love of bringing people together and organising events followed, which is exactly how we ended up hosting an event together with Elisabeth and her team at TechBBQ this year, on European infrastructure and frontier technology. It felt like the right moment to sit down and go deeper on where she thinks the next generation of European hard-tech companies gets built.

We talked about backing frontier tech at the earliest stage, what Europe still gets wrong about building its own infrastructure, and what she is watching most closely right now.

Q&A

1. You sit at Germany's most active seed investor, writing some of the earliest cheques into deep tech and frontier companies. What does a frontier tech company have to show you at pre-seed, when there is often little more than a team and a hard technical bet?

A tier-1, triple-A* team and market. Founder-market fit has to be strong: I want founders who've spent real time on one specific problem, actively trying to disprove their own bet, and becoming experts in that market. At pre-seed in the foundation/ infra layer, we underwrite the research team first, people who can pull in a second world-class researcher and strong/industry-specific angels, because there the team is the real investment.

Furthermore, the technical claim still has to be defensible: something I can reproduce or benchmark in diligence. Especially in Europe, I look for a compute-light edge, differentiation through method, not GPU budget, which makes a European frontier bet winnable, not a losing race against better-funded US labs. Beyond that, I always look for investor-founder fit: not every frontier company is a match for us, and not every founder will want us and that's totally fair. In the end it should just click and be an enjoyable experience to work together.

2. We are hosting an event together at TechBBQ on European infrastructure and frontier tech. When you say European infrastructure, what do you actually mean, and which parts of it do you think are most underbuilt right now?

In the past months we've made exciting investments in this space, from an AI foundation model for causal reasoning to bringing memory to AI agents.

I think that sovereignty is shifting from compliance cost to buying criterion. Core model-serving commoditises as open-source matures, so value moves to the layers that change the economics of inference, and to who controls the data feeding the model.

Europe is the one market where strict regulation, industrial data owners, and political will for EU-controlled compute exist at once and that is not a handicap, but much rather it is demand. Open-weight models are here to stay and keep improving, letting industrial groups run capable models on-premise, own their data, and swap the model as it improves. And an underrated second-order effect: build in Europe first, especially in physical and applied AI, under the hardest safety and regulatory bar and you've built something trusted enough to travel anywhere, the US included.

3. Europe keeps producing world-class research and then watching the companies scale elsewhere. From your seat at HTGF, where does that leakage actually happen, and what would it take to keep more of the frontier tech being built here?

The leakage is mostly structural, not compositional and that distinction matters. I recently looked a two transatlantic twins (one was US-based and the other in Europe): same stage, same sector, comparable teams. The European one raised a significantly smaller round. When you hold the company constant and the gap persists, it isn't that Europe builds worse startups or that the team is less strong, the same startup simply gets less capital here, especially in later rounds and at exit, where our market stays thin relative to the US. So it won't fix itself by producing better founders; it closes only when the capital market deepens. Nonetheless, Talent is the second related force: the most promising teams increasingly spin out of the big frontier labs, and that operator experience sits in San Francisco. But that's the narrower gap, Europe has the deep research talent, the inventors; what we lack is the ex-frontier-lab operators to pair them with. Fix formation, deepen late-stage capital, and pair research depth with operator experience and far more of this stays and scales here.

4. Frontier tech takes years and a lot of capital to pay off, which has historically made it a hard fit for venture timelines. How do you underwrite that patience, and where do you draw the line between a deep tech company worth waiting for and a science project?

Venture is a bet, especially the earlier you invest in. Maybe one in ten flies, and I love the infrastructure layer precisely because it's so unpredictable, yet moving unbelievably fast. The old "deep tech needs a decade" rule is breaking: Prior Labs was acquired by SAP roughly 15 months after pre-seed, and players like Cohere and Aleph Alpha are forming sovereign-AI alliances at real scale. The why-now is that the next GenAI wave is finally reaching critical, regulated use cases, predictive maintenance, diagnostics, decisions LLMs can't touch because they aren't causal or deterministic by design.

5. Germany has real strength in industrial tech, energy, and hard engineering. Is the next wave of European frontier companies going to come from that industrial base, or from somewhere less obvious?

Both, but the less obvious answer excites me more. Germany's strength in energy, robotics and hard engineering will anchor a wave. The layer I'd watch most is physical and applied AI, because it comes with a data moat you cannot scrape from the internet: real-world sensor data from brownfield industrial environments is proprietary by nature and compounds with every deployment. And that's precisely where Europe can still win, because we sit on decades of industrial data, from engineering to machine logs, that are real but underused advantages. The core skill is translating that domain knowledge into machine-readable training data. The why-now is structural: roughly 98% of industrial sensor data still goes unused, manufacturing labour shortages are worsening, and NVIDIA's physical-AI push is pulling the category forward. Europe won't out-scale the US, but here it can out-specialise it, because we own the data, and data sovereignty is a strategic question, not a compliance one.

6. If we sit down again in two years, what part of European infrastructure or frontier tech do you think will look most different, and what do you hope people are finally taking seriously by then?

Two things. First, just a prediction: an AI FinOps function becomes standard in most companies. Inference economics and token spend are real budget lines now, and someone has to own them, that role barely exists today and will be everywhere in two years. Second, more of a hope: that we stop treating all AI as one thing. "AI equals AI" is how bad capital gets allocated; the sooner people distinguish LLMs from causal, physical or verifiable systems, the healthier this gets. And I hope the technology finally leaves our bubble, I mean around 2.4 billion people actively use generative AI today, but roughly 71%, never knowingly have. The real infrastructure story isn't the next model; it's what happens when that 71% comes online, and who owns the data and the layers underneath when they do.