本期20VC节目由Harry Stebbings主持，Rory O'Driscoll、Jason Lemkin以及来自Benchmark的特邀嘉宾Jack Altman一同深入探讨了风险投资和人工智能领域最新、常常是爆炸性的新闻。 讨论首先从Anthropic泄露的S1草案开始，该草案显示其2025年预计营收为46亿美元，但同时运营亏损80亿美元，并有5180亿美元的计算承诺。Rory将大部分泄露的数据斥为“糟糕的解读”，称唯一有用的洞察是，两家客户贡献了25%的营收。他认为只有Anthropic的第三季度数据才能真正说明问题。Jack Altman表示，公开市场投资者可能比私募投资者更具长期导向，他们关注市场结构而非月度波动。然而，Jason Lemkin担心这种负面定性，尤其是“对人类生存构成风险”的披露，可能会劝退散户投资者，并可能导致股价跌破IPO价格，他将其与Facebook的IPO相提并论。Rory反驳说，机构买家与普通大众不同，但他承认可能存在潜在的间接政治反弹。 讨论随后转向了个人AI助手的竞争格局，特别是Instinct和Muse。Instinct最近在由Benchmark领投的一轮C轮融资中，以100亿美元的估值完成了10亿美元的融资。Jack Altman将Instinct描述为“聚合器中的聚合器”，代表了一种与整个第三方互联网进行交互的消费级代理的重要新范式。他将其比作编码工具的兴起，认为在这一关键的新兴领域中，多个参与者都能成功。Jason质疑这些代理是否能像成功的编码或法律AI工具那样实现“全天候不间断”的使用。Harry指出Instinct有潜力蚕食OpenAI等平台的聊天使用量。Rory指出OpenClaw最初也瞄准了这个领域，但Instinct通过提供必要的虚拟CPU基础设施“做到了”。 讨论了Benchmark对Instinct在100亿美元（尚未产生营收）估值下的重大投资，Jack承认尽管价格很高，但由于该技术的新生性质，这感觉像是一项“早期投资”。小组讨论了此类高风险高回报押注的基金规模分配，Harry建议将基金的5%用于50倍上行/1倍下行的投资，而Rory开玩笑地提议将30%用于真正的凯利赌注。Jack承认当前市场处于“最高波动期”，这导致投资回报出现偏差，即“许多人将失败”。 一条主要新闻是AMD以82亿美元的股票收购了Fei-Fei Li的World Lab。Rory称这是对Li的“滑稽的正义”，她作为通过ImageNet成为AI领域的先驱，在短短两年半内成功创立并退出了她的Neolab。Jason认为这是AMD的一项战略举措，AMD的股价今年已飙升279%，市值达到万亿美元，旨在与NVIDIA竞争。Harry表示，这可能是机器人基础模型领域许多“大型收购”中的第一例，并指出潜在收购方众多。小组还谈到了不断上升的资金需求，从Neolabs的5亿美元到推高薪资的“疯狂”人才争夺战，例如Meta聘请MongoDB的CEO负责Muse Enterprise。Jack解释说，这种人才流失是极具吸引力的财务报价与身处创新“白热化中心”的“时代精神”共同作用的结果。 关于OpenAI，讨论转向了其计算能力限制，该公司将其200美元套餐所包含的服务减半。Rory强调AI成本是不可持续的，这使得像Jev（便宜70倍，快100倍）这样的替代方案变得有吸引力。Jack强调，最终“都将归结为谁拥有计算能力”。Jason认为OpenAI和Anthropic可以变得更具成本竞争力，而且由于安全问题，企业客户对开源模型，特别是来自中国的模型，越来越警惕。他认为，由于这些因素，“开源作为市场份额已经达到上限”。 其他值得关注的新闻包括Aura尽管最初计划强劲，却撤回了其160亿美元的IPO，这让小组成员对强劲的市场感到惊讶。作为Aura股东的Rory推测，大量的二级市场份额导致了投资者的价格敏感性，而Jason则鉴于标普的表现，质疑“市场状况”这一理由。NewBank可能以80-120亿美元收购英国数字银行Monzo的事件也引发了讨论，Harry对NewBank对英国资产的兴趣感到惊讶，并暗示Monzo内部董事会的不稳定可能促成了此次出售。最后，Bessemer Venture Partners宣布推出一支高达57.5亿美元的新基金（其中包括17.5亿美元的种子基金），而NFX则选择仅投资GP自有资金，这标志着当前市场中基金管理的两种截然不同的方式。

This episode of 20VC, hosted by Harry Stebbings with Rory O'Driscoll, Jason Lemkin, and special guest Jack Altman from Benchmark, delved into the latest, often blistering, news in the venture and AI world. The discussion opened with the leaked draft S1 from Anthropic, revealing projected 2025 revenues of $4.6 billion against an $8 billion operating loss and $518 billion in compute commitments. Rory dismissed most of the leaked data as "shitty bad takes," stating the only useful insight was two customers accounting for 25% of revenue. He argued that only Anthropic's Q3 numbers would be truly indicative. Jack Altman suggested public market investors might be more long-term oriented than private ones, focusing on market structure rather than monthly fluctuations. Jason Lemkin, however, worried the negative framing, especially the "existential risk to humanity" disclosure, could deter retail investors and potentially lead to a stock drift below IPO price, drawing parallels to the Facebook IPO. Rory countered that institutional buyers are distinct from the general public, but acknowledged a potential indirect political backlash. The conversation then shifted to the competitive landscape of personal AI assistants, particularly Instinct and Muse. Instinct recently closed a $1 billion Series C at a $10 billion valuation, led by Benchmark. Jack Altman described Instinct as an "aggregator of aggregators," representing an important new paradigm of consumer agents interacting with the entire third-party internet. He compared it to the rise of coding tools, believing that multiple players can succeed in this crucial new space. Jason questioned if these agents would achieve the "all effing day long" usage seen in successful coding or legal AI tools. Harry noted Instinct's potential to cannibalize chat usage from platforms like OpenAI. Rory pointed out that OpenClaw initially aimed for this space, but Instinct "did it" by providing the necessary virtual CPU infrastructure. Benchmark's significant investment in Instinct at a $10 billion pre-revenue valuation was discussed, with Jack admitting it felt like an "early-stage investment" despite the price, due to the nascent nature of the technology. The panel debated the appropriate fund sizing for such high-variance bets, with Harry suggesting 5% of a fund for 50x upside/1x downside, while Rory jokingly proposed 30% for a true Kelly bet. Jack acknowledged the current market's "highest variance time," leading to skewed returns where "many will fail." A major news item was AMD's acquisition of Fei-Fei Li's World Lab for $8.2 billion in stock. Rory hailed it as "comic justice" for Li, a pioneer in AI through ImageNet, who successfully launched and exited her Neolab in just two and a half years. Jason saw it as a strategic move for AMD, which has seen its stock soar 279% this year to a trillion-dollar valuation, to compete with NVIDIA. Harry suggested this could be the first of many such "big-ass acquisitions" for robotic foundation models, noting the abundance of potential acquirers. The panel also touched on the escalating capital requirements, from $500 million for Neolabs to the "insane" talent war driving up salaries, exemplified by Meta hiring MongoDB's CEO for Muse Enterprise. Jack explained this talent drain as a mix of compelling financial offers and the "zeitgeist" of being at the "white-hot center" of innovation. Regarding OpenAI, the discussion turned to its compute constraints, with the company halving what its $200 plan buys. Rory highlighted that AI costs are unsustainable, making alternatives like Jev (70 times cheaper, 100 times faster) attractive. Jack emphasized that ultimately, "it's all going to come down to like who's got the compute." Jason argued that OpenAI and Anthropic can become more cost-competitive, and enterprise clients are increasingly wary of open-source models, especially those from China, due to security concerns. He believes "open source has reached its maximum as a market share" due to these factors. Other notable news included Aura pulling its $16 billion IPO despite strong initial plans, surprising the panel given the robust market. Rory, an Aura shareholder, speculated that a significant secondary component led to investor price sensitivity, while Jason questioned the "market conditions" rationale given the S&P's performance. The potential acquisition of British neobank Monzo by NewBank for $8-12 billion also sparked debate, with Harry surprised by NewBank's interest in a UK asset, suggesting Monzo's internal board instability may have spurred the sale. Finally, Bessemer Venture Partners announced a massive new $5.75 billion fund (including a $1.75 billion seed fund), while NFX chose to invest solely GP capital, marking two contrasting approaches to fund management in the current market.