The core message of the provided transcript excerpt is that venture investing in AI is both extraordinarily expensive and still grounded in real adoption potential: the panelists resist calling the market a simple bubble, but they repeatedly distinguish between validated growth-stage companies and speculative inception rounds. The session opens by framing AI startup investing as one of the “craziest” eras investors have seen, where very large seed and inception rounds are becoming more common. The moderator introduces the panel as a cross-section of European and US venture perspectives: Creandum as a deeply European fund, General Catalyst and Accel as US firms with permanent European presences, and Andreessen Horowitz as a US firm investing in European deals without a European office.

Kathi Wilhelm then brings the investors into the discussion by asking each to state their focus areas. Zhenya Loginov, drawing on operating experience at Miro and Segment before joining Accel, says he is interested in both AI infrastructure and applications, especially “intelligence per watt,” including new chip architectures and efficient models. Zeynep Yavuz-Willson describes General Catalyst as a transformation firm and says her focus is on diffusing AI into the economy, particularly treating AI adoption as a change-management challenge in Europe’s services economy rather than only a technical problem. Chris Schnabel of Andreessen Horowitz says he invests globally in early-stage AI applications and is especially excited about traditional industries that had been underserved by software before AI.

The central debate in the excerpt is whether the AI market is a bubble. Zhenya rates the market roughly seven or eight out of ten on “bubbliness,” but clarifies that he sees it as expensive rather than detached from fundamentals because there is still meaningful innovation and value creation. He says he would be proven too skeptical if one or two new AI labs succeed in becoming the next major foundation-model company, or if applied software companies discover stronger revenue protection and defensible moats. Zeynep separates later-stage growth rounds from inception rounds: she rates growth rounds for companies with product-market fit as only two or three out of ten on bubble risk, because AI adoption remains early, especially in Europe, where she says less than 20% of businesses with more than ten employees adopted AI in the previous year.

For Zeynep, the more speculative area is large inception rounds, especially at the model layer and increasingly at the application layer as companies blend downward into models using open source and custom training. She rates those rounds closer to seven or eight because investors are backing exceptional talent before product-market fit is known. Chris agrees that the market is still very early for both consumer and enterprise adoption, and that most model capabilities have not yet diffused into the economy. The excerpt ends as Kathi shifts to a practical operating question for Zhenya: how to identify real, sticky revenue when AI companies can scale to tens of millions in revenue within months. The transcript provided ends before his answer is available, so later session conclusions cannot be verified from the supplied text.