门罗创投（Menlo Ventures）合伙人马特·墨菲（Matt Murphy）做客20VC节目，与哈里·斯特宾斯（Harry Stebbings）讨论了门罗创投的重大投资，尤其是在人工智能领域的投资，以及更广泛的风险投资趋势。 墨菲讲述了门罗创投对Anthropic进行锚定投资的缘由。经安贾娜·米塔（Anjanae Mita）引荐，他立即被达里奥·阿莫迪（Dario Amodei）的愿景（阿莫迪是OpenAI的创始人之一，后因理念不合离职）和技术实力所折服。尽管Anthropic当时尚未实现营收，估值却高达40多亿美元——对于门罗创投6亿美元的风险基金来说，这无疑是一项巨大挑战——但基准测试显示Anthropic的模型在性能上与ChatGPT不相上下，同时花费的资本显著减少，这进一步坚定了墨菲的信念。他赞扬了合伙人在支持这项非传统交易时所展现出的灵活性，承认最初小额投资，随后通过SPV（特殊目的载体）扩大规模的决定令人忐忑不安，因为这是门罗创投的首次尝试，但最终证明令人振奋。这一经历凸显了风险投资领域的一个转变：鉴于当前可能出现的巨额回报，投资于那些“异类”公司远比追求更高的所有权比例更为关键。 谈到稀释和SPV，墨菲指出，由于公司需要资本来进行进攻性扩张以及持续融资的信号效应，现在公司接受更高稀释的程度已成为常态。SPV是突破基金投资上限的工具，能让有限合伙人（LPs）参与到后期投资机会中。他相信有限合伙人正日益理解这种新的动态。 门罗创投对Lovable的投资，该公司在一年内从零增长到3亿美元的年经常性收入（ARR），这体现了他们对“异类”公司的追逐。墨菲强调了创始人的愿景，以及其成为“有史以来最有价值公司之一”的潜力。他谈到了人工智能领域的利润率挑战，指出虽然许多公司目前利润率为20-30%，但要实现长期成功，尤其是考虑到推理成本，必须有可信的途径达到60-70%的利润率。 关于开源模型和前沿模型之间的争论，墨菲认为这不是“二选一”的问题，而是“两者兼具”的局面。尽管开源模型功能齐全，但像Anthropic这样的基础模型能提供卓越的智能和性能，从而提高客户留存率和收入。公司将采用多模型策略，以优化成本、推理能力和延迟。他特别提到了Open Router，一家门罗创投投资的AI路由公司，称其为“猛兽”，能帮助应用程序智能地选择最佳模型。他还认为，公司自研定制芯片（如谷歌、Meta、Anthropic）是为了优化成本并在规模化应用中获得特定的性能优势。 另一家由马克斯（Max）领导的重要投资组合公司Lagora，因其对基础模型的防御性而受到讨论。墨菲认为，法律、税务和会计等复杂的多参与者工作流程，需要基于模型构建的专业应用程序，这使得通用模型难以取代它们。 关于投资阶段，墨菲认为A轮融资充满挑战，因为许多公司在产品市场契合度有限的情况下，估值却很高。门罗创投现在采用“杠铃策略”，要么更早投资（种子轮、NeoLabs），要么投资于后期阶段的“异类”公司（年经常性收入超过1000万美元），这些公司通常市场领导地位更为明确。门罗创投30亿美元的基金规模虽然庞大，但却反映了他们刻意选择维持一个专注的团队和文化，将协作和一致性置于指数级增长之上。 当被问及错失的交易时，墨菲指出“入场太晚”或缺乏既有关系是常见原因。他还强调，在公司可以在一年内从零增长到1亿美元ARR的环境下，过去“三三翻倍，双双翻倍”（triple, triple, double, double）这样的增长指标已不再足够令人兴奋。他最大的“失误”是Plaid，这教会了他韧性。他表示很欣赏来自11 Labs的马蒂（Matty），称其是自己乐意支持的创始人。 展望未来，墨菲最看好人工智能在医疗突破和医疗服务交付方面的潜力。他还指出，AI基础设施和开发者工具是目前投资不足但潜力巨大的领域。他认为当前的人工智能浪潮是他职业生涯中最大的一次，并预测未来5-10年社会将发生令人难以置信的变革。

Matt Murphy, a partner at Menlo Ventures, joined Harry Stebbings on 20VC to discuss Menlo's significant investments, particularly in AI, and broader venture capital trends. Murphy recounted the origin of Menlo's anchor investment in Anthropic. Introduced by Anjanae Mita, he was immediately convinced by Dario Amodei's vision (Amodei, an OpenAI creator, left due to differing priorities) and technical prowess. Despite Anthropic being pre-revenue with a $4B+ valuation—a stretch for Menlo's $600M venture fund—Murphy's conviction was bolstered by benchmarks showing Anthropic's models on par with ChatGPT while spending significantly less capital. He praised his partners for their flexibility in backing the unconventional deal, acknowledging that the decision to invest smaller initially and then scale up via an SPV (Special Purpose Vehicle) was nerve-wracking, being Menlo's first of its kind, but ultimately exhilarating. This experience highlighted a shift in VC: ownership percentage is less critical than being *in* outlier companies, given today's massive outcome scenarios. Discussing dilution and SPVs, Murphy noted that companies now normalize higher dilution due to the need for capital to play offense and the signaling effect of constant fundraising. SPVs are tools to exceed fund limits, bringing LPs into later-stage opportunities. He believes LPs are increasingly understanding this new dynamic. Menlo's investment in Lovable, a company growing from zero to $300M ARR in a year, exemplified their pursuit of outliers. Murphy emphasized the founder's vision and the potential for it to be "one of the most valuable companies of all time." He addressed the challenge of margins in AI, noting that while many companies currently have 20-30% margins, a credible path to 60-70% is crucial for long-term success, especially with inference costs. On the debate between open-source and frontier models, Murphy believes it's not an "either/or" but a "both/and." While open-source is functional, foundation models like Anthropic's offer superior intelligence and performance that boost customer retention and revenue. Companies will adopt a multi-model approach, optimizing for cost, reasoning, and latency. He highlighted Open Router, an AI routing company Menlo has invested in, as a "beast" that helps applications intelligently select the best model. He also sees companies building custom chips (like Google, Meta, Anthropic) as an optimization play to manage costs and gain specific performance advantages at scale. Another significant portfolio company, Lagora, led by Max, was discussed for its defensibility against foundation models. Murphy argued that complex, multi-constituent workflows like legal, tax, and accounting require specialized applications built on top of models, making them difficult for a general model to displace. Regarding investment stages, Murphy views Series A as challenging due to high valuations for often limited product-market fit. Menlo now employs a "barbell strategy," investing earlier (seed, NeoLabs) or in later-stage outliers (> $10M ARR) where market leadership is clearer. Menlo's $3B fund size, while substantial, reflects a deliberate choice to maintain a focused team and culture, prioritizing collaboration and alignment over exponential growth. When asked about lost deals, Murphy identified being "late to the party" or lacking pre-existing relationships as common reasons. He also stressed that past growth metrics like "triple, triple, double, double" are no longer exciting enough in an environment where companies can go from zero to $100M ARR in a year. His biggest "miss" was Plaid, which taught him resilience. He expressed admiration for Matty from 11 Labs as a founder he'd like to back. Looking ahead, Murphy is most excited about AI's potential for medical breakthroughs and healthcare delivery. He also highlighted AI infrastructure and developer tools as an underinvested area with significant potential. He sees the current AI wave as the biggest of his career, predicting mind-boggling societal transformations in the next 5-10 years.