The Blink Edition No. 7

The European private markets briefing you can't afford to miss. Deals, signals, and moves, decoded every Monday by SeedBlink.

"AI native" was the phrase that opened doors two years ago. Today it closes them.

More and more venture funds now screen out startups that lead with the label, and not because AI stopped mattering. It's because the label stopped saying anything. When every company in the room runs on the same handful of models, "we use AI" describes your tooling, not your business. What investors want to understand is which problem you solve and why your solution creates value. AI is how you deliver it, not what you sell.

There's a second risk I'd watch more closely than most people do: what happens when the platform you depend on decides to ship your product. A company's entire business model can be at risk if it sits on a single AI provider, one that may launch a competing product, change its terms, or simply become your competitor. It has already happened to companies far larger than the ones pitching us, and the pace of change makes it hard to see coming.

None of this is an argument against building with AI. It's an argument for knowing exactly where you stand when the ground moves. Beyond solving a real problem, there's a question worth asking, before a round and after it: if your model provider shipped your product tomorrow, what would you still have?

Roughly nine out of ten companies in Y Combinator's summer 2025 cohort were AI-native. Nearly all of them depended on the same three providers for the models underneath.

That observation comes from Brookings, and the spending behind it is substantial: Menlo Ventures puts enterprise spending on generative AI at $37 billion in 2025, up from $11.5 billion the year before. The dependency is even starker on this side of the Atlantic. The 2026 AVP Transatlantic Founder Index, a survey of more than 100 venture-backed founders across Europe and the US, found 69% of companies running primarily or entirely on US cloud and AI providers, including 62% of European founders.

This changes what diligence has to cover. For years, platform risk meant channel risk: what happens if the app store changes its rules, if the ad algorithm shifts. Model dependency is a sharper version of the same problem, because the provider is not only a supplier. It may also be a direct competitor in the application layer, with privileged visibility into usage patterns and product demand. Brookings documents cases of providers already using that position against downstream competitors.

So the questions worth asking in a data room have changed. Not just "what's your gross margin," but "what happens to it if inference pricing doubles." Not just "what's your moat," but "which part of it survives if your provider ships this natively." A company with proprietary data, workflow ownership, or genuine distribution can absorb a shift at the model layer. A company whose product is a thin interface over someone else's API cannot, and that difference doesn't show up in a growth chart until it's too late to act on.

For investors, model dependency belongs on the risk register next to customer concentration, and for the same reason: it's a single point of failure that looks like an efficiency right up until it isn't.

‍

The festival is over. The business keeps going.

UNTOLD's 11th edition wrapped in Cluj this weekend, four days of one of the largest music events in Europe. What remains once the stages come down is the company behind it.

Ten editions of operating history before this one, revenue up nearly 28% last year to €28.7M while staying profitable, and a #3 ranking worldwide by DJ Mag three years running. The round on SeedBlink is still open, and it's the first time individual investors can back that business directly, with festival access perks for the first 500 investors.

Explore the round.‍
‍

Germany 🇩🇪
‍
NavVis raises $85M Series D
‍
‍Munich-based NavVis raised $85M led by The Jordan Company for spatial twins of factories and construction sites, having scanned over a billion square metres in 2025 alone, the kind of proprietary data physical AI can't be trained without

Read more

‍

Cyprus 🇨🇾

Omilia secures €58.1M Series B

‍Omilia raised €58.1M led by Expedition Growth Capital after growing ARR more than tenfold since its Series A, built on its own voice AI stack rather than third-party models, which lets it price without passing token costs to customers.

Read more

‍

France 🇫🇷  
‍
Chargepoly raises €23M

Aix-en-Provence-based Chargepoly raised €23M led by Meridiam to build depot charging for electric truck fleets, already running hundreds of DC fast charging points across France, the UK and Canada.

Read more

‍

The offline economy: are investors betting on startups that take us off-screen?

2026 was supposed to be the year AI ate everything. Instead, some of Europe's more interesting early-stage activity is in companies that deliberately do less with technology, on the thesis that when software is cheap to build, defensibility moves to the physical world.

Read more

‍

Knowing when to stop: the art of making a loop converge

a16z's Yoko Li on why AI agents have no idea when their work is done. In one test, two-thirds of the total spend bought no improvement at all, with the loop running long after progress had stopped.

Read more

‍

Sweaty, insane, cracked: decoding the language of founder praise

Sifted's field guide to the escalating vocabulary of tech compliments. Like startup valuations, the praise has suffered from inflation: once "genius" would do, now you apparently need to be "insanely cracked."

Read more