Legora CEO Max Junestrand said Europe should put no effort into building its own frontier AI lab, calling the idea wishful thinking, on the same day Bloomberg reported the company is in talks to raise at least $300m at a pre-money valuation of about $8.5bn. That combination matters because it pairs a strategic position with a live funding number.
What was said in Amsterdam and what the funding talks show
Junestrand spoke at the HumanX conference in Amsterdam on Tuesday, addressing the summer's run of incidents in which AI agents hacked other companies. He told Politico that the warnings that followed deserve a pinch of salt, arguing that much of the alarm is marketing that overemphasizes proposed capabilities. Hours after the interview, Bloomberg reported that Legora is negotiating a round of at least $300m at roughly $8.5bn pre-money, which would close above $8.8bn if the terms hold. Talks are continuing and Legora declined to comment.
Legora runs entirely on models it buys from American labs, so a European frontier lab would be a supplier rather than a competitor to it. Junestrand argued Europe should build on top of the models it receives, because the technology arrives as an input and the value sits in what gets built with it. That position is close to Ursula von der Leyen's State of the Union line, in which she backed an AI slowdown and said Europe need not build frontier models to draw value from them. He was less comfortable on what dependence means in practice: asked whether repeated delays in getting the newest OpenAI and Anthropic models put European customers at a disadvantage, he answered that of course they do.
Legora's own numbers show how fast the application layer is moving. The company reached $100m in revenue inside 18 months, and CFO David Eckstein posted that it hit $200m in annual recurring revenue last week, with the second $100m taking under six months. He added that the company is on track to add more ARR this quarter than its entire revenue base at the end of 2025, with a pilot win rate of 75% so far this quarter. More than 40% of new business in the third quarter came from in-house legal teams rather than law firms. The Series D opened in March at $550m, a $5.55bn valuation, then a $50m extension with Nvidia and Atlassian took it to $600m and $5.6bn.
What this means for companies building on purchased models
For a company buying models rather than training them, the practical consequence is that vendor choice and access timing become core product decisions. Junestrand said securing Legora against AI-driven attacks ranks above shipping new features, because the company holds legal data, and that reaching the newest models quickly is part of how it does that. The reach is wider than the funding round suggests: 130,000 lawyers use Legora each month across 2,100 firms and legal teams in more than 80 countries, with more than half of the AmLaw 50 as customers, including Salesforce, Palo Alto Networks, Baker McKenzie, and White & Case. For a small vendor, the same dependency shows up as a procurement question; for a large one, it shows up as a product roadmap.
The limits are visible in the same interview. Politico describes the model delays as repeated rather than occasional and attributes the priority to plugging cybersecurity gaps, but neither the article nor Junestrand puts a length on them. The $8.5bn figure is pre-money, talks could change, and Legora declined to comment, so the valuation is not settled. Gross margin is positive and improved while revenue doubled, but that is the company's own figure rather than an audited one. Meanwhile Harvey, the closest American competitor, announced a $550m round at a $15.6bn valuation earlier this month, so a European company growing at Legora's rate still carries roughly half its rival's valuation while both buy from the same suppliers.
The marker to watch is how the round closes. If investors price the application layer near $9bn on revenue doubled in under six months, built on other people's models, that confirms the division of labor Junestrand described. If the terms move down or the round stalls, the market is telling European AI companies that buying models is not enough to command that multiple.
