Descriptions:
Martin Casado and Sarah Wang, general partners at Andreessen Horowitz, join Latent Space live from the a16z offices to break down the mechanics of AI’s unprecedented capital formation cycle — a flywheel where frontier lab fundraises, compute procurement, and strategic equity stakes have become so intertwined and large-scale that they require hybrid venture/growth deal analysis even for pre-revenue companies.
The conversation is anchored by concrete economic arguments rather than abstraction. Casado revisits his earlier thesis that a $1 billion training run economically justifies a custom ASIC: if a purpose-built chip saves even 20% on compute costs, that’s $200 million — enough to fund a tape-out — and realistic efficiency gains are substantially higher. He notes that OpenAI’s confirmed Broadcom ASIC deals validate what was theoretical two years ago. Wang adds nuance on deal mechanics: at today’s scale, seed-stage AI companies require enterprise BD, complex compute contracts, and go-to-market partnerships that historically wouldn’t appear until Series C.
On the circular funding question — whether strategic investors providing compute in exchange for equity creates bubble dynamics — Casado distinguishes the current moment from the fiber-optic overbuild of the 1990s: there are no dark GPUs, demand is absorbing supply, and the structural imbalance looks different from prior cycles. The pair also touch on a16z’s American Dynamism team’s role in defense and hardware-adjacent AI, and why robotics investments require vertical domain expertise that differs from their horizontal infrastructure approach.
📺 Source: Latent Space · Published February 19, 2026
🏷️ Format: Interview
