“The best founders find a way to make it happen”: VC Roundtable with Bryan Kim and David Clark | E2222

What It Covers

VCs Bryan Kim (a16z), David Clark (VenCap), and Jason Calacanis debate AI bubble concerns, evaluate Oboe's $16M Series A, analyze AI startup valuations and margins, discuss power infrastructure challenges, and predict 2025 IPO prospects for mega-cap private companies.

Key Questions Answered

• •AI Valuation Framework: Companies growing to $100M revenue in 12-18 months justify higher valuations than historical benchmarks. VenCap identifies six bubble characteristics and finds AI markets lack speculative excess despite froth, with revenue and adoption supporting current pricing levels unlike past bubbles.
• •Ownership Economics for Returns: Early stage funds target 7% ownership through accelerators, assuming 50-70% dilution. For venture returns to work, managers need enough ownership that one company can return the entire fund. At seed stage, 2% of a decacorn generates meaningful returns when combined with pro rata rights.
• •AI Gross Margin Strategy: Founders accept lower initial margins (due to inference costs) while building tiered pricing models across consumer, enterprise, and creator segments. Margin dollars matter more than percentages when companies reach order-of-magnitude larger scale than SaaS predecessors, with costs declining over time.
• •Exit Timing Philosophy: Best firms hold top companies through compounding growth rather than selling prematurely for DPI. Databricks demonstrates this, reaching $4B run rate growing 50% annually at $125B valuation. Manage liquidity through portfolio construction mixing early and growth stages, not pressuring premature exits.
• •Product Velocity as Moat: Post-ChatGPT era demands shipping products weekly and testing distribution channels rapidly, replacing traditional retention-focused strategies. Founders who combine fast product iteration with innovative go-to-market (hackathons, influencer partnerships, startup alliances) win competitive AI markets where model layers evolve constantly.