
上周,顶尖创始人和投资人齐聚 Cerebral Valley AI 峰会,正值该行业经历最动荡的周之一,期间发生了全球性网络攻击,且黄仁勋被迫辞职。
Last week, top founders and investors gathered for the Cerebral Valley AI Summit during one of the most turbulent weeks the industry has seen, marked by global cyberattacks and Jensen Huang’s forced resignation.
River AI 的 Igor Babuschkin、Vercel 的 Guillermo Rauch、Databricks 的 Ali Ghodsi、Palo Alto Networks 的 Nikesh Arora、Crosby 的 Ryan Daniels、Etched 的 Robert Wachen、Mercor 的 Brendan Foody、Fireworks 的 Lin Qiao 等领袖向我们分享了他们对 AI 未来走向的看法。
Leaders like River AI’s Igor Babuschkin, Vercel’s Guillermo Rauch, Databricks’ Ali Ghodsi, Palo Alto Networks’ Nikesh Arora, Crosby’s Ryan Daniels, Etched’s Robert Wachen, Mercor’s Brendan Foody, Fireworks’ Lin Qiao, and more told us their thoughts about where AI is headed next.
我们将在通讯和 Newcomer AI Summits YouTube 频道发布所有 13 场炉边对谈。
We’re publishing all 13 onstage conversations here in the newsletter and on our Newcomer AI Summits YouTube channel.
感谢赞助商:Fireworks、Oracle、IREN、Index Ventures、Exa、Nebius、WorkOS、ElevenLabs、Higgsfield、Felicis、SambaNova、BessemerVenturePartners、Weekend、LiquidAI、QuantumBlack、SapphireVentures、OutshiftbyCisco 和 OnePay。
Many thanks to our sponsors: Fireworks, Oracle, IREN, Index Ventures, Exa, Nebius, WorkOS, ElevenLabs, Higgsfield, Felicis, SambaNova, BessemerVenturePartners, Weekend, LiquidAI, QuantumBlack, SapphireVentures, OutshiftbyCisco, and OnePay.
Nikesh Arora(Palo Alto Networks) Palo Alto Networks 首席执行官 Nikesh Arora 表示,防御 AI 智能体攻击者需要多个模型,而非单一模型。他说,最好的防御是让安全系统学习数据在特定行业中的正常流动方式,然后为每个客户量身定制,逐一进行。
Nikesh Arora (Palo Alto Networks) Palo Alto Networks CEO Nikesh Arora said that defending against AI agent attackers will take many models, not one. The best defense is to train security systems on how data normally moves through a given industry, then tailor them to each customer, one at a time, he said.
对于一些科技领袖建议他接替黄仁勋在 Nvidia 的职位,Arora 拒绝置评。他预计黄仁勋的辞职会导致短期动荡,但不认为会严重损害公司。
Arora wouldn’t say whether he’d take Huang’s spot at Nvidia, as some tech leaders have suggested. He expects Huang’s resignation to cause short-term instability but doesn’t think it will seriously hurt the company.
AI 100 与 Byron Deeter 及 Talia Goldberg(Bessemer Venture Partners) Bessemer 的 Byron Deeter 和 Talia Goldberg 在 Cerebral Valley AI 峰会现场介绍了与 Newcomer 合作发布的 AI 100 榜单。榜单上公司的总估值达 3.6 万亿美元,是 2016 年首期 Cloud 100 总估值的 36 倍。
AI 100 with Byron Deeter & Talia Goldberg (Bessemer Venture Partners) Bessemer’s Byron Deeter and Talia Goldberg gave a rundown on the AI 100, made in partnership with Newcomer, onstage at the Cerebral Valley AI Summit. The companies on the list are worth a combined $3.6 trillion, 36 times the value of the first Cloud 100 in 2016.
Deeter 表示,这些价值很大程度上集中在最大的基础模型实验室。然而,大多数大型实验室的客户公司并不局限于单一模型类型,而是针对不同任务混合搭配不同模型。
A lot of this value is concentrated in the largest foundation model labs. However, most companies that are customers of the big labs aren’t sticking to a single model type, but are mixing and matching different models for different tasks, Deeter said.
随后,Deeter 与 Eric Newcomer 共同探讨了 Anthropic 为何领先于 OpenAI、Databricks 和 Stripe 如何融入 AI 经济,以及为何 2027 年可能迎来一波 AI 首次公开募股(IPO)。
Afterward, Deeter joined Eric Newcomer to discuss why Anthropic ranks ahead of OpenAI, how Databricks and Stripe fit into the AI economy, and why 2027 could bring a wave of AI IPOs.
Hemant Taneja(General Catalyst) General Catalyst 首席执行官 Hemant Taneja 表示,Nvidia 的动荡引发了关于 AI 领导层连续性的“巨大担忧”,但也可能为更多芯片初创公司打开大门。
Hemant Taneja (General Catalyst) General Catalyst CEO Hemant Taneja said the Nvidia shakeup raises “huge concerns” about continuity in AI leadership, but could also open the door for more chip startups.
Taneja 称中国构成严重风险,因为其拥有能源和制造能力,但表示美国的“自下而上的创业生态系统”仍遥遥领先。他认为,为了赢得公众支持,行业应少谈失业,多谈 AI 能带来的“丰裕”与“能动性”。
Taneja called China a serious risk because of its energy and manufacturing capacity, but said America’s “bottom-up entrepreneurial ecosystem” is still far ahead. To win public support, he argued the industry should talk less about job losses and more about the “abundance” and “agency” AI can bring.
Guillermo Rauch(Vercel) Vercel 首席执行官 Guillermo Rauch 表示,企业正在摆脱昂贵的“极限追求 Token”,转而混合使用小模型、大模型、微调模型和开放权重模型以降低成本。
Guillermo Rauch (Vercel) Vercel CEO Guillermo Rauch said that companies are moving past expensive “token-maxxing” and mixing small, large, fine-tuned, and open-weight models to cut costs.
他表示,为了保持竞争力,企业将不得不让 AI 智能体访问专有数据,并在其上层部署身份协议,以便智能体能在安全权限下代表用户行动。
To stay competitive, businesses will have to give AI agents access to proprietary data, with identity protocols on top so agents can act for users with secure permissions, he said.
Igor Babuschkin(River AI) xAI 联合创始人 Igor Babuschkin 认为,AI 集中在少数实验室手中是当今行业面临的最大风险。他的新公司 River AI 希望为企业,最终也为个人,提供可自主拥有和改进的模型。其首款产品允许企业在自己的硬件上对模型进行后训练和运行。
Igor Babuschkin (River AI) xAI co-founder Igor Babuschkin argued that concentrating AI in a few labs is the biggest risk in the industry today. His new company, River AI, wants to give businesses, and eventually individuals, models they can own and improve themselves. Its first product lets companies post-train and run models on their own hardware.
Babuschkin 表示,一个由许多中等能力模型处理保险理赔和会计等工作的世界也更安全,因为没有单一公司或个人能造成过大的危害。
Babuschkin said a world with many moderately capable models handling work like insurance claims and accounting is also safer, because no single company or person could do outsized harm.
Robert Wachen(Etched) AI 芯片与机架制造商 Etched 联合创始人兼总裁 Robert Wachen 表示,黄仁勋的辞职对这家雇佣了 100 多名前 Nvidia 员工的公司打击巨大。
Robert Wachen (Etched) Robert Wachen, co-founder and president of AI chip and rack maker Etched, said Huang’s resignation hit hard at a company that employs more than 100 former Nvidia workers.
Wachen 依然看好 AI 基础设施,并预测将出现历史上最大规模的白领工作增长,驱动力不是人,而是具备能动性的 AI 智能体。他预期高能力智能体将降低芯片设计的价值,而物理层面的难题 —— 如将芯片和互连集成到数据中心机柜中 —— 依然难以解决。
Wachen is still optimistic about AI infrastructure and predicted the largest growth in white-collar work in history, driven not by people but by agentic workers. He expects highly capable agents to make chip design less valuable, while physical problems, like integrating chips and interconnects into data center racks, stay hard to solve.
Crosby CEO Ryan Daniels 将其初创公司描述为一家 “具备能动性的律师事务所”:它拥有律师、职业责任保险和律所实体,但据他所知,这是第一家直接向非律师销售法律 AI 的公司。
Ryan Daniels (Crosby) Crosby CEO Ryan Daniels has described his startup as an “agentic law firm”: it has lawyers, liability insurance, and a law firm entity, but it’s the first company he knows of that sells legal AI directly to people who aren’t lawyers.
他曾考虑收购律所,但决定致力于消除计时收费模式,方法是雇佣愿意将工作移交给 AI 智能体的律师。
He considered buying law firms but decided to work on getting rid of the billable hour instead, by hiring lawyers who want to hand their work off to AI agents.
Factory CEO Matan Grinberg 与 Fireworks CEO Lin Qiao 探讨了开放权重模型、安全性和 AI 智能体。Grinberg 表示,开放模型在 Factory 运行的 token 中所占比例,已从 2026 年初的几乎为零增长到远超 10%,部分客户使用开放模型的比例甚至超过一半。
Lin Qiao (Fireworks) & Matan Grinberg (Factory) Factory CEO Matan Grinberg and Fireworks CEO Lin Qiao discussed open-weight models, security, and AI agents. Grinberg said open models have grown from almost nothing at the start of 2026 to well over 10% of the tokens running through Factory, with some customers using them more than half of the time.
Qiao 谈到了 Fireworks 近期披露的安全事件,称这是一起旨在窃取算力和 token 以转售的攻击。Fireworks 与 Palo Alto Networks 合作开展取证,并使用了 CrowdStrike 和 Wiz 的检测工具。
Qiao discussed the security incident Fireworks disclosed recently, describing it as an attack aimed at stealing compute and tokens to resell. Fireworks worked with Palo Alto Networks on forensics and used detection tools from CrowdStrike and Wiz.
Applied Intuition 创始人 Qasar Younis 告诉在场观众,美国是他见过的最反 AI 的国家 —— 他将这种情绪归因于不平等和工资停滞。
Qasar Younis (Applied Intuition) Applied Intuition founder Qasar Younis told the crowd that America is the most anti-AI country he’s seen so far — a sentiment he attributed to inequality and stagnant wages.
Younis 表示,他看到对其公司自主技术的需求,来自无人机制造商的比来自自动驾驶汽车制造商的更多。他对人形机器人持怀疑态度,称目前尚无人弄清如何从中获利,且仅靠规模无法解决机器人技术难题。
Younis said he’s seeing more demand for his company’s autonomous technology from drone manufacturers than from self-driving carmakers. He’s skeptical of humanoid robots, saying no one has figured out how to make money from them yet and that scale alone won’t solve robotics.
Mercor首席执行官Brendan Foody解释了AI智能体如何改变了模型训练者的构成。早期的聊天机器人依赖大量工人撰写通顺的句子,但如今的智能体需要顶尖工程师、律师、医生和银行家来设计评估体系和强化学习环境。
Brendan Foody (Mercor) Mercor CEO Brendan Foody explained how AI agents have changed who trains the models. Early chatbots relied on large numbers of workers writing clean sentences, but today’s agents need top engineers, lawyers, doctors, and bankers to design evaluations and reinforcement learning environments.
Foody表示,这项工作的市场规模受限于人类仍能比模型做得更好的任务数量。
Foody said the market for this work is limited by how many tasks humans can still do better than models.
Flapping Airplanes联合创始人Asher Spector解释了他如何运营一个专注于提高AI效率的研究实验室。他不担心OpenAI和Anthropic等前沿实验室进入同一领域。他表示,该实验室的优势不在于更聪明或更努力,而在于拥有不同的激励机制,且没有创新者的窘境。
Asher Spector (Flapping Airplanes) Flapping Airplanes co-founder Asher Spector explained how he’s running a research lab focused on making AI more efficient. He isn’t worried about frontier labs like OpenAI and Anthropic moving into the same space. He said the lab’s edge isn’t being smarter or working harder, but having different incentives and no innovator’s dilemma.
Spector说,该实验室在全栈范围内工作,最专注于训练环节。其框架使芯片既更高效又更强大。英伟达是投资方之一,团队计划在黄仁勋辞职后继续与该公司合作。
Spector said the lab works across the stack and focuses most on training. Its framework makes chips both more efficient and more capable. Nvidia is an investor, and the team plans to keep working with the company after Huang’s resignation.
Databricks首席执行官Ali Ghodsi表示,黄仁勋的离去移除了AI行业的一道财务后盾,因为过去每当投资者犹豫时,黄仁勋都会迅速支持相关公司。Ghodsi预计这种损失将开始伤害许多公司,尽管不会立竿见影。他仍认为2026年是上市的糟糕时机,但表示Databricks不会永远保持私有。
Ali Ghodsi (Databricks) Databricks CEO Ali Ghodsi said Huang’s exit removes a financial backstop for the AI industry, since Huang was quick to support companies whenever investors hesitated. Ghodsi expects that loss to start hurting many companies, though not right away. He still thinks 2026 is a bad year to go public, but said Databricks won’t stay private forever.
Ghodsi说,公司们正专注于削减成本以让AI实现回报。Databricks在OpenAI、Anthropic、Google和开源模型之间路由了超过四千万亿(quadrillion)个token,其新路由器将更简单的请求发送给更便宜的模型。
Ghodsi said companies are focused on cutting costs to make AI pay off. Databricks routes more than four quadrillion tokens across OpenAI, Anthropic, Google, and open-source models, and its new router sends simpler requests to cheaper models.
去年,Cerebral Valley的观众投票选Perplexity为他们最想做空的公司。今年,首席执行官Aravind Srinivas表示那些批评者错了,现在押注Perplexity会失败的人将再次亏损。
Aravind Srinivas (Perplexity) Last year, the Cerebral Valley audience voted Perplexity the company they’d most like to short. This year, CEO Aravind Srinivas said those critics were wrong, and that anyone who bets against Perplexity now will lose money again.
他表示,公司已转向主要服务能够承担令牌成本的企业客户,并更多地定位为一种工具和基础设施平台,而非聊天机器人。
His company has shifted to serving mostly enterprise customers who can afford the token costs, he said, and has leaned into becoming more of a harness and infrastructure play than a chatbot.