Investors at TechCrunch Disrupt 2025 framed the current AI funding environment as a market moving from hype to homework. On the conference’s AI Stage, partners from leading venture firms emphasized that founders must show measurable economics and durable advantages rather than simply promising to “apply an LLM” to a problem.

“Everybody can call themselves an AI company today,” said a partner from a major Silicon Valley firm on the panel. “What gets you to the next check is repeatable revenue, defensible data, and margins that survive API pricing shocks.” That triad—revenue, data, margins—reappeared across comments and audience Q&A, reflecting investors’ response to two years of rapid base-model commoditization and cloud concentration.

Panelists argued that defensibility increasingly hinges on proprietary datasets, vertical specialization and deep integrations that create switching costs. Startups that focus on industry-specific workflows—healthcare coding, legal discovery, industrial automation—were cited as more investible than horizontal consumer-facing apps, which face low barriers to replication and thin monetization. “Verticals give you context, labels and customer lock-in,” one investor said, noting that enterprise pilots that convert to contract revenue within 6–12 months are now a critical signal for early-stage checks.

Infrastructure plays drew a contrasting kind of enthusiasm. Several VCs said they are allocating to tooling and model-ops companies that help other startups lower inference costs, manage compliance, or deploy private models. The logic is fiscal: margin compression from API costs and cloud bills has made unit economics fragile for app-layer startups, while firms that capture recurring platform fees can scale with healthier gross margins.

Market-level implications are already visible. Valuation discipline has returned in many late-stage rounds, investors said, and capital is concentrating more tightly on teams that can demonstrate metrics—customer acquisition cost, lifetime value, churn—rather than growth narratives alone. That shift has pushed some founders to pursue earlier enterprise deals or to extend seed runs while they refine defensibility.

Policy and antitrust considerations also threaded through the discussion. With a handful of large players dominating base models and cloud infrastructure, panelists warned of regulatory headwinds and potential interventions that could reshape competitive dynamics. “Concentration at the foundation layer creates both business risk and public policy risk,” one panelist observed, arguing that investors now evaluate how startups would fare if access to models or cloud capacity became constrained by regulation or price changes.

For founders seeking term sheets, the practical takeaway from Disrupt was concrete: show credible unit economics, provide evidence of proprietary data or integration-driven lock-in, and have a plan to manage model and cloud costs. TechCrunch’s event, drawing more than 10,000 startup and VC leaders to Moscone West from October 27–29, reflected a broader market moment in which capital is still available but more selectively deployed. Organizers noted an early-bird registration incentive—save up to $668 before prices rise after September 26—underscoring the urgency investors and founders alike feel as they position for the next funding cycle.