在一场关于人工智能现状与未来的广泛讨论中，Insight创始人杰瑞·默多克（Jerry Murdoch）结合其25年来在科技周期方面的经验积累，分享了他对市场稳定性、技术变革和投资策略的深刻见解。 **AI泡沫与市场调整风险** 默多克对“人工智能泡沫”表达了担忧，他指出目前市场“泡沫达到顶峰”且“价格失控”。他预测市场将在2026年10月至2027年3月期间出现回调，尤其是在“伊朗战争”等地缘政治冲突升级的情况下。他最主要关注的是信贷市场中断，这可能会影响那些积累了大量债务的“超大规模公司”（hyperscalers）。默多克认为，自满是最大的警示信号，他将其与2008年金融危机期间风险被低估的情况相提并论。日本经济的不稳定及其持有的美国国债也构成全球风险。 **超大规模公司 vs. 新云提供商** 尽管超大规模公司由于其业务稳定，最有可能在市场错位时幸存，但默多克认为“新云提供商”（neoclouds，即专门的人工智能基础设施提供商）则非常脆弱。他估计，其中至少一半将在36个月内消失，尤其是在经济出现动荡的情况下。新云提供商能否生存的关键区别在于资本效率和强大的领导力。他以Fireworks为例，认为其比Base10等公司拥有更高的资本效率和更盈利的模式。 **前沿模型 vs. 开源AI** 默多克挑战了“一个代币就是一个代币”的观念。他认为，定制模型会改变代币的价值和效率。前沿模型（例如OpenAI、Anthropic）价格昂贵且可定制性较低，而开源模型则提供了显著更便宜的选择（每代币0.10-0.11美元，而前沿模型为两位数美分），并允许进行专业化定制。他预测开源模型将被大规模采用，尤其是在客户服务或编码等高度专业化的任务中。尽管前沿模型可能继续引领复杂创新，但由于成本和定制需求，开源模型将占据全球需求的很大一部分。 **企业AI战略与网络安全** 企业在与前沿模型提供商共享专有数据方面变得越来越谨慎，这与Alex Karp的担忧不谋而合。默多克建议公司审慎行事，并将敏感数据保留在防火墙内，这为开源解决方案带来了更多机会。他警告称，缺乏深思熟虑的人工智能战略或产品的SaaS公司，尤其是那些未能适应代理系统“协同工作时代”的公司，将面临巨大困境。 安全是一个被严重低估的威胁。默多克强调需要沙盒（如Docker、E2B）来隔离AI代理，因为它们的概率性质使其具有风险。他认为，由于这些新漏洞，一个“网络安全的黄金时代”即将到来。 **投资理念与市场动态** 默多克的投资建议核心是寻找那些出于绝对必要性而创立公司、并专注于影响力的创始人。他提倡注重资本效率，并在扩张前主导利基市场。他批评那些仅仅为了抢占市场而奉行“低利润文化”的公司，指出虽然这对于亚马逊有效，但大多数公司需要优先考虑利润以实现长期可持续发展。 关于估值，他承认当前存在炒作周期，但告诫不要认为每家公司都能达到“万亿美元”的市值。虽然像Fireworks这样的基础设施公司，由于其在AI生态系统中的重要作用，可以证明其快速增长的合理性，但应用层公司则不然。他指出，IPO市场充满挑战，许多公司如果缺乏引人注目的AI整合，可能会面临“巨大折扣”。私募股权公司杠杆率高，特别容易受到金融动荡的影响。 **AI和区块链的未来** 默多克认为最令人兴奋的发展将是“持续学习模型”（预计2-10年内实现），随后是“终身学习”。这些模型将与当前静态模型截然不同，并最终取而代之，迎来新一代的AI能力。 他还预测，区块链技术目前正处于“幻灭的低谷”，但将在五年内找到真正的应用价值。具体来说，他认为它对于“代理支付”、金融基础设施以及AI推理的去中心化交易至关重要，将超越投机性加密货币的范畴。 **快速点评科技巨头** 在“娶、杀”游戏（在此指对公司前景的判断）中： * **娶（长期持有）：** Meta、谷歌和微软。尽管面临挑战，它们庞大的用户基础提供了稳定性、收入和适应时间。 * **杀（短期风险）：** Meta，并非因为它会崩溃，而是因为它可能成为一家“无趣的”公用事业公司，派发股息而非实现爆发式增长。 * **苹果的AI战略：** 其长期影响尚不确定，取决于其幕后操作是“天才般的耐心”还是“愚蠢的消费”。 * **英伟达（NVIDIA）：** 他预测五年内英伟达将“突破”10万亿美元，尽管近期有所停滞。他将当前的市场犹豫归因于市场加速的普遍放缓和潜在的经济衰退。 默多克总结道，尽管AI领域激动人心，但区分真正的创新、可持续的商业模式和短暂的炒作至关重要。

In a wide-ranging discussion on the state and future of AI, Jerry Murdoch, founder of Insight, shared his insights on market stability, technological shifts, and investment strategies, drawing on 25 years of tech cycle experience. **AI Bubble and Market Correction Risks** Murdoch expressed concern about an "AI bubble," highlighting current "peak froth" and "out of control prices." He predicts a market correction between October '26 and March '27, particularly if geopolitical conflicts, like the "Iran war," escalate. The primary concern is a credit market disruption, which could impact "hyperscalers" who have accumulated significant debt. Complacency is the biggest warning sign, drawing parallels to the 2008 financial crisis where risks were underestimated. Japan's economic instability and its holding of US treasuries also pose a global risk. **Hyperscalers vs. Neoclouds** While hyperscalers are best prepared to survive a dislocation due to their stable businesses, Murdoch believes "neoclouds" (specialized AI infrastructure providers) are highly vulnerable. He estimates at least half of them will disappear within 36 months, particularly if an economic disruption occurs. The key differentiators for survival among neoclouds will be capital efficiency and strong leadership, citing Fireworks as a more capital-efficient and profitable model compared to others like Base10. **Frontier Models vs. Open-Source AI** Murdoch challenges the notion that "a token is a token." He argues that customizing models changes the token's value and efficiency. While frontier models (e.g., OpenAI, Anthropic) are expensive and less customizable, open-source models offer a significantly cheaper alternative ($0.10-$0.11 per token vs. double-digit cents) and allow for specialization. He foresees massive adoption of open-source models, especially for highly specialized tasks like customer service or coding. Though frontier models will likely continue to lead in complex innovation, open-source will capture a vast portion of global demand due to cost and customization needs. **Enterprise AI Strategy and Cybersecurity** Enterprises are increasingly cautious about sharing proprietary data with frontier model providers, echoing Alex Karp's concerns. Murdoch advises companies to practice discernment and keep sensitive data behind their firewalls, driving further opportunity for open-source solutions. He warns that SaaS companies without a thoughtful AI strategy or product, especially those not adapting to the "co-work era" of agentic systems, will struggle significantly. Security is a major underestimated threat. Murdoch emphasizes the need for sandboxes (like Docker, E2B) to contain AI agents, whose probabilistic nature makes them risky. He believes there's a "golden age of cyber" coming due to these new vulnerabilities. **Investment Philosophy and Market Dynamics** Murdoch's investment advice centers on finding founders driven by an absolute necessity to build their business, focusing on impact. He advocates for capital efficiency and dominating a niche before expanding. He criticizes companies that embrace a "low-margin culture" solely for land grabs, noting that while it worked for Amazon, most companies need to prioritize margin for long-term sustainability. Regarding valuations, he acknowledges the current hype cycle, but cautions against assuming every company can achieve "trillion-dollar" status. While some infrastructure companies like Fireworks can justify rapid growth due to their essential role in the AI ecosystem, application-layer companies may not. He notes that the IPO market is challenging, and many companies may face "mega discounts" if they lack compelling AI integration. PE firms, highly leveraged, are particularly vulnerable to financial dislocations. **The Future of AI and Blockchain** Murdoch believes the most exciting development will be "continuous learning models" (estimated 2-10 years away), followed by "lifelong learning." These models will fundamentally differ from current static models and will eventually replace them, ushering in a new generation of AI capabilities. He also predicts that blockchain technology, currently in a "valley of disillusionment," will find true utility within five years. Specifically, he sees it as crucial for "agent payments," financial infrastructure, and decentralized exchanges for AI inference, moving beyond speculative cryptocurrencies. **Quick Takes on Tech Giants** In a "Shag, Marry, Kill" exercise: * **Marry (long-term hold):** Meta, Google, and Microsoft. Despite their challenges, their immense user bases provide stability, revenue, and time to adapt. * **Kill (short-term risk):** Meta, not for collapse, but for potentially becoming a "boring" utility company, yielding dividends rather than explosive growth. * **Apple's AI strategy:** Its long-term impact is uncertain, depending on whether it's "genius patience" or "dumb consumption" behind the scenes. * **NVIDIA:** He predicts it will go "over" $10 trillion in five years, despite recent plateaus, attributing current market hesitation to a general slowing of market acceleration and potential economic downturns. Murdoch concludes that while the AI landscape is incredibly exciting, it's crucial to distinguish between genuine innovation, sustainable business models, and temporary hype.