Summary
Matt Murphy, partner at Menlo Ventures and lead investor in Anthropic, joins Harry Stebbings on 20VC for a candid conversation about the current AI investment landscape. Murphy recounts the origin story of Menlo’s Anthropic bet — introduced to Dario Amodei and Tom Brown by former associate Anjen Mida, he saw a technically superior team spending a fraction of OpenAI’s compute while matching benchmark performance. The $4 billion pre-revenue valuation was a stretch for a typical venture fund, but conviction in the founding team and the benchmarks won out.
The conversation covers strategic terrain that matters for anyone tracking the AI industry: whether open-source models can meaningfully displace frontier proprietary systems like Claude, the economics of full-stack AI development including custom silicon, and how model routing and inference services create or destroy value. Murphy argues that Anthropic’s tiered model family — spanning Haiku, Sonnet, Opus, and Fable — positions it for a world where enterprises blend models based on task and cost rather than committing wholesale to one provider.
Murphy also addresses why AI labs are moving into custom chips despite the difficulty of competing with Nvidia: at sufficient revenue scale, even marginal inference cost reductions on specific workloads justify the engineering investment. The episode is a useful window into how top-tier AI venture investors think about moats, margins, and the evolving battle between open and closed AI systems.
📺 Source: 20VC with Harry Stebbings · Published July 27, 2026
🏷️ Format: Podcast
