Çelik here.

This week Europe quietly changed how it places its bets. France committed €655 million to put a French model on a million government desks, Poland's development bank backed a startup tracking objects in orbit, and the same French lab lined up money at a valuation that doubled in nine months. The thread is the same in every case: public demand, public money, and private capital are converging on a short list of national champions.

If you raise or build in the region, the Poland story shows where state money is actually flowing, and the capital roundup sets the valuation bar you will be measured against. And the AI Act vote at the bottom explains why the runway just got longer for whoever is already on the list.

On June 16, PM Sébastien Lecornu committed €655 million to a single sovereign AI assistant for the French civil service, a chatbot built on Mistral that will reach roughly a million of France's 2.6 million public workers. The same week, France moved to replace Palantir inside its domestic intelligence service with local software. The state is not subsidizing a champion anymore, it is becoming its largest customer.

The positional shift is guaranteed demand. A frontier lab's hardest problem is durable revenue, and a government that wires your model into a million desks solves it with one signature. Mistral now holds the thing capital cannot manufacture, a captive market the size of a mid-cap company, before it has to win a single enterprise deal on merit.

If you sell software into European public sectors, the buying test is shifting from "best tool" to "made here," and a US logo is becoming a liability in the room. Map which of your public-sector pipelines now carry a sovereignty requirement, and line up a European-hosted or open-weight option before an RFP demands one.

Small teams don't have time to stitch together five tools and hope it works.

Apollo gives you everything you need to find leads, reach them, and close deals — all in one place:

Stop juggling tools and start building pipeline that scales.

With Apollo, the AI revenue engine powering 4M+ users.

Sign up for free

On June 12, Polish space-tracking startup Sybilla Technologies raised over €8 million, and the notable name on the cap table was public: Vinci, the investment arm of Poland's state development bank BGK, alongside TCEE Fund IV. Sybilla's optical-sensor network tracks objects in orbit, the kind of dual-use capability a government wants to own outright. Days earlier, Czech AI commissioner Jan Kavalírek opened talks in Warsaw on a joint Czech-Poland AI Gigafactory bid.

The shift is that CEE states have stopped waiting for Brussels and started deploying their own capital into their own champions. A state bank co-investing in space-defense is the clearest sign yet that public money in the region is steering toward sovereign deep tech, not generic startups. Poland is running the playbook France just ran, demand plus capital behind a national winner, on a smaller budget.

For founders in the Balkans, the lesson is the backer, not the satellite. Public development banks and EU-program money are turning into early-stage investors for dual-use and sovereignty-adjacent tech. Map which state-linked funds in your country can co-invest, and frame your next round around a capability a government would want to own, not just a market you want to sell into.

In the same week, the money chasing Europe's security and sovereignty stack repriced. Mistral, the lab France just adopted, is in talks to raise about €3 billion at roughly €20 billion, nearly double its September mark. Israeli cyber firm Dream, which sells sovereign AI defense to European governments, raised $260 million at a $3 billion valuation on about $300 million of sales. Both are rising toward the $18 billion benchmark defense-AI builder Helsing set this spring.

Read them next to the France deal. The money is not spreading across a field of contenders, it is compounding into a few names that already have governments as customers, the same dynamic that just handed Mistral a million public desks. Private capital is following public demand into the same short list.

For regional founders and investors, this is the valuation bar you will be measured against, and the category signal is loud: defense, cyber, and sovereign infrastructure are where the largest European checks land. If you are raising nearby, anchor your comps to these rounds, and decide whether you are building something a champion acquires or something that has to outrun one.

On June 16, the European Parliament voted 423 to 57 to adopt the Digital Omnibus, pushing the AI Act's high-risk obligations from August 2026 to December 2027 and content-labelling rules to December 2026. The same package put hard teeth on consumer harm: a full ban on non-consensual "nudifier" systems, fines up to €35 million, a December 2026 deadline. It still needs Council sign-off.

The deadline every operator wrote into a 2026 plan just moved sixteen months. For the champions raising and selling this week, that is room to scale before the heaviest compliance load lands. The reprieve is selective, though. The consumer-harm bans got sharper, not softer, so the relief flows to infrastructure and enterprise AI, not to anything pointed at the public.

If you build high-risk AI for the EU market, do not unwind your compliance work, re-time it. The obligation is still coming, just in late 2027, which frees up engineering you had earmarked for this summer. But if your product touches synthetic media or anything a regulator would call consumer harm, your clock got shorter, not longer: December 2026, with fines that bite.

Five tools to install or test this week. Productivity and sales were quiet, so design and dev each get two.

Marketing: Google Analytics now counts visitors sent by AI assistants. Google added reporting that tracks real people arriving from ChatGPT and other assistants, traffic your dashboards have been missing. If you cannot see AI referrals, you cannot defend the budget that earns them. Turn it on and baseline how much of your traffic already comes from chatbots.

Design: xAI's Grok Imagine Video 1.5 hits general availability at 86% below Sora 2 Pro. xAI moved its top-ranked image-to-video model to GA across the API and apps at $4.20 a minute on June 16. For a small team that collapses the cost of ad creative and product demos. Generate one launch clip before you brief an agency.

Design: Anthropic rebuilt Claude Design and fixed the token burn. The June 17 update imports your design system from a GitHub repo or raw files, edits the canvas directly, and stops the runaway spend that ate Pro allowances. If you bounced off Claude Design over cost, it earns a second look this week.

Dev: GitHub's Copilot now finds the right agent for you. Agent Finder lets you describe a task in plain language, then searches a catalog of agents, ranks the matches, and pulls the right one in. It removes the config tax of juggling specialized agents by hand. Point it at a multi-step task and let it route.

Dev: Databricks Agent Bricks adds Grok natively. At its Data + AI Summit, Databricks made xAI's Grok available inside Agent Bricks on June 18, so you can build agents on governed data without separate API plumbing. If your stack already runs on Databricks, prototype one agent on it this week.

Next edition soon, Çelik