20VC: Leading Anthropic's First Ever Round | Will Open Source Threaten Anthropic's Business | Do Margins Matter in a World of AI | Why Triple, Triple, Double, Double is Not Good Enough Today | Why Series A is Hard Today with Matt Murphy @ Menlo

Summary

Matt Murphy lays out Menlo’s Anthropic decision as a deliberate exception: the company’s models looked technically special, had strong early benchmarks, and appeared to have achieved that with dramatically less capital than peers. He argues that in today’s AI market, ownership percentage matters less than being positioned in the extreme outliers, because the winners will be much larger and more capital-intensive than prior venture norms. Murphy also says open source will matter in multi-model systems, but it is unlikely to displace frontier foundation models like Anthropic, which can win on intelligence, retention, and willingness to pay. On the venture side, he describes a barbell strategy, compressed seed-to-A timelines, the fading of old signaling rules, and why Series A is especially awkward right now. He closes with where Menlo sees opportunity next: AI infrastructure, observability, routing layers like OpenRouter, selective chip economics, and long-term healthcare and drug-discovery applications.

Notes

• Menlo’s Anthropic investment was a rules-breaking exception, not a standard fund fit.
• The initial Anthropic check was a little over $10M, followed by a $500M+ SPV in the next round.
• Menlo’s internal constraint was a narrow average check size, which Anthropic exceeded.
• Anthropic’s technical edge was framed as strong performance on far less capital.
• Pre-launch, Anthropic was already comparing favorably to ChatGPT on benchmarks.
• The Dario story mattered because he was seen as the technical creator behind the relevant OpenAI work.
• Ownership matters less in AI when returns are dominated by extreme outliers.
• More capital is flowing into top AI companies, making ownership harder to preserve.
• AI companies are raising more often for growth, offense, signaling, and liquidity.
• SPVs are a practical way to break fund-level concentration limits for exceptional rounds.
• Secondary-market SPVs are increasingly problematic for founders.
• Margins still matter, but the right AI gross-margin target is often 60% to 70%, not 80% to 90%.
• Many current AI companies start around 20% to 30% gross margin because inference costs are structurally high.
• Open source will be useful in a mixed stack, but not enough to displace frontier foundation models.
• Anthropic’s models were described as special, performant, and intelligent.
• The next wave of AI is about smarter routing and model selection, not one model everywhere.
• OpenRouter is positioned as an inference marketplace where developers choose models by price, reasoning, performance, and latency.
• Custom chip efforts are a cost-optimization play at very large scale.
• Google TPUs and Amazon Trainium were cited as examples of scale-based vertical integration.
• Application defensibility can come from workflow complexity across multiple stakeholders.
• Ligora’s long-term strategy extends beyond legal into adjacent workflows like tax and accounting.
• Menlo sees the seed-to-Series-A window as dramatically compressed.
• Menlo now runs a barbell strategy: later-stage outliers on one end, earlier seed investing on the other.
• Seed investing has become more prominent because speed and early access matter more.
• The old venture swim-lane model is fading as firms go full stack and collaborate more.
• Negative signaling from institutional participation is much less of a concern for strong companies.
• Pattern recognition and sector density matter more in seed, growth, and semiconductors.
• Chip development cycles can take 2 to 3 years, and design wins can evaporate.
• San Francisco’s AI momentum has returned, and Bay Area context remains a big advantage.
• Europe is harder mode for founders, which can produce stronger entrepreneurs.
• Menlo loses many deals by arriving too late to relationship-building.
• Trust and relationship depth often matter more than valuation in competitive rounds.
• Triple-triple-double-double is no longer the right benchmark for venture-scale growth.
• Companies doing 0-to-100 in a year now define the top end of the market.
• Menlo’s Anthropic position is materially large, with carry economics well above a $10B reference point.
• Menlo describes itself as a challenger firm trying to compound its AI advantage.
• The next big opportunity set is AI infrastructure, observability, and tooling above the foundation-model layer.
• Robotics, Neo Labs, and some defense-tech areas look overheated to Menlo.
• Menlo is in about seven Neo Labs.
• Healthcare and drug discovery are the most exciting long-term AI opportunities for Menlo.