最新一期20VC节目汇聚了主持人Harry Stebbings、Rory O'Driscoll，嘉宾Nikesh Arora（Palo Alto Networks首席执行官）以及Jason Lampkin，他们就科技和人工智能领域的关键事件和趋势展开了热烈讨论。 对话始于Bending Spoons以12.85亿美元收购**Airtable**一事，这一价格与Airtable此前高达110亿美元的估值形成鲜明对比。Rory赞扬了Bending Spoons以低倍数估值进行收购的精明策略，而小组成员则认为这反映了SaaS市场正在经历一场更广泛的重新评估。Nikesh质疑这究竟是“价格错位”，还是SaaS“长期增长率的根本性变化”。Jason则认为，这凸显了“创始人倦怠”以及市场格局的转变，尽管Airtable仍有20%的增长并融入了人工智能技术，私募股权公司却明显缺席竞标。 接着，他们讨论了**Leo Ashenbrenner的对冲基金崩盘事件**。大家的共识是，Ashenbrenner在AI趋势判断上“绝对正确”，但在“投资组合构建上完全错误”，4倍杠杆使得其覆灭“不可避免”。小组成员强调了投资者入场时机的重要性，指出后期投资者遭受了严重冲击。Nikesh结合自身经验，表示相信Ashenbrenner会从这“残酷的一周”中汲取教训。 近期**Anthropic模型安全漏洞事件**引发了对**网络安全**的深入探讨。Nikesh强调，AI模型正极大地加速漏洞被发现和利用的速度，使得传统安全基础设施“完全不敷使用”。他敦促企业通过大量投资于新的防御措施来“缴纳应缴的税款”（即履行应尽的责任）。Nikesh解释说，AI将同时增强边界安全（阻止“已知恶意行为”）和内部检测（发现“未知恶意行为”）。Jason分享了一个个人轶事，他的人工智能代理（Claude）自主修改了他的代码，这突显了代理安全领域“狂野西部”般的特性，以及为AI代理引入新的控制机制、紧急停止开关和身份管理的必要性。 讨论随后转向了**AI智能和算力需求**。Nikesh曾有句名言：“普通智能将免费且会变得更智能”，而“卓越智能则需要付费”。他提出，未来3-5年内，“土地、许可、能源、算力”将成为最具价值的商品，这验证了对多样化能源（如Valor Atomics的核反应堆，甚至“鸡粪”甲烷生产商）进行投资的合理性。Rory质疑了开放权重模型在缺乏明确变现途径情况下的长期可行性，但Nikesh强调，对算力的需求超越了特定的模型提供商。 分析亚马逊、谷歌和微软近期发布的**云业务财报**显示，在AI推理销售的推动下，这些公司实现了强劲增长，验证了其大规模资本支出承诺的合理性。市场奖励了那些展现出明确人工智能投资回报率（ROI）的公司，这与Meta形成了对比，Meta的支出导致其股价贬值，但回报短期内不明显。Palantir通过为企业打包AI智能解决方案实现了“近100%的增长”，这一成功提供了一个鲜明例证，表明企业“能够销售这种AI”。Jason呼吁其他创始人，如果Palantir能取得这样的成绩，他们也应该“更加努力”。 小组成员将当前时代定义为“淘金热时期”，认为“未来五到十年内，每个消费者应用都将被重写”。他们承认，巨大的资本支出与营收变现之间存在“时机问题”，同时，由于监管或基础设施障碍，算力也可能面临潜在的“供应问题”。Nikesh强调了企业“上下文”（即构建并编纂组织知识和训练数据）的至关重要性，认为它与模型智能本身同样关键。这将导致原始模型与那些通过专有上下文增强的模型之间展开一场“商品化竞争”，而最快学习者将成为最终赢家。

The latest 20VC episode featured a lively discussion between hosts Harry Stebbings, Rory O'Driscoll, and guest Nikesh Arora (CEO of Palo Alto Networks), along with Jason Lampkin, touching on pivotal events and trends in the tech and AI landscape. The conversation kicked off with the **Airtable acquisition** by Bending Spoons for $1.285 billion, a stark contrast to its previous $11 billion valuation. Rory praised Bending Spoons' shrewd strategy of acquiring at low multiples, while the panel viewed it as a reflection of a broader re-evaluation in the SaaS market. Nikesh questioned whether this represented a "pricing dislocation or a fundamental change in the long-term growth rate" of SaaS. Jason suggested it highlighted "founder fatigue" and the market's shift, with PE firms notably absent from bidding, despite Airtable's 20% growth and AI infusions. Next, they discussed **Leo Ashenbrenner's hedge fund implosion**. The consensus was that Ashenbrenner was "absolutely right on trend" regarding AI but "absolutely wrong on portfolio construction," with 4x leverage making a wipeout "inevitable." The panel stressed the importance of investor timing, noting that late-stage investors were severely impacted. Nikesh, drawing on his experience, expressed confidence that Ashenbrenner would learn from the "brutal week." The recent **Anthropic model breaches** led to a deep dive into **cybersecurity**. Nikesh highlighted that AI models are drastically accelerating the speed at which vulnerabilities are found and exploited, making traditional security infrastructure "wholly unfit for purpose." He urged enterprises to "pay your taxes" by investing heavily in new defenses. Nikesh explained that AI would enhance both perimeter security (stopping "known bads") and internal detection (finding "unknown bads"). Jason shared a personal anecdote of an AI agent (Claude) autonomously modifying his code, underscoring the "wild west" nature of agent security and the need for new control mechanisms, kill switches, and identity management for AI agents. The discussion then shifted to **AI intelligence and compute demand**. Nikesh famously declared, "average intelligence is going to be free and the average intelligence will get smarter," while "exceptional intelligence will be paid for." He posited that "land, permits, energy, compute" would be the most valuable commodities for the next 3-5 years, validating investments in diverse energy sources like Valor Atomics' nuclear reactors and even "chicken manure" methane producers. Rory questioned the long-term viability of open-weight models without clear monetization, but Nikesh emphasized that the demand for compute transcends specific model providers. Analyzing recent **cloud earnings reports** from Amazon, Google, and Microsoft revealed strong growth driven by AI inference sales, validating massive CapEx commitments. The market rewarded companies showing clear ROI for their AI investments, contrasting with Meta, whose spending led to stock depreciation without immediately obvious returns. Palantir's success, with "nearly 100% growth" from packaging AI intelligence for enterprises, served as a stark example that companies "can sell this AI." Jason challenged other founders to "work harder" if Palantir could achieve such results. The panel concluded by framing the current era as a "gold rush moment," where "every consumer app will get rewritten in the next five to 10 years." The "timing problem" between monumental CapEx and revenue realization was acknowledged, alongside potential "supply problems" for compute due to regulatory or infrastructure hurdles. Nikesh stressed the critical importance of enterprise "context"—building and codifying organizational knowledge and training data—as being as vital as model intelligence itself. This would lead to a "commoditization battle" between raw models and those augmented with proprietary context, with the fastest learners being the ultimate winners.