Index Partner Defends Mega-Fund Bets as Frontier AI Labs Near $1T

Fireside Chat with Jahanvi Sardana, Index and Ben Bergman, Business Insider — Ai4

Ai4

Open-weight models function as commercial tools for lagging rivals rather than altruism, while frontier labs' unprecedented revenue trajectory justifies late-stage mega-fund valuations despite early-stage seed froth.

AI-assisted summary of Fireside Chat with Jahanvi Sardana, Index and Ben Bergman, Business Insider — original source. Check the original for full context.

Source published: 2026-08-22

Detailed notes

## Background Published on YouTube by Ai4 on August 22, 2026, this video features a fireside chat between Jahanvi Sardana, Partner at Index Ventures, and Ben Bergman, Senior Correspondent at Business Insider. According to the page description, the session brings together Sardana and Bergman for an in-depth conversation exploring current artificial intelligence trends and venture investment dynamics.

## TL;DR Index Ventures partner Jahanvi Sardana examines the stark divide between Silicon Valley AI enthusiasm and real-world adoption, identifying high-upside investment opportunities in vertical 'neofirms' and foundational infrastructure. While skeptical of inflated early-stage valuations for lab spinouts, she defends hyperscaler mega-funds and reframes open-weight releases as pragmatic commercial strategies rather than pure altruism. Sardana rejects catastrophic predictions of structural unemployment, concluding with a warning about the rise of 'cognitive debt' as humans increasingly outsource critical thinking to automated models.

## Atmosphere & style The dialogue is crisp, witty, and grounded in sharp economic realism, opening with playful banter about the blistering 113-degree desert heat before diving into unsentimental market realities. Bergman serves as an engaging, probing interviewer, allowing Sardana to deliver candid, fast-paced insights punctuated by humorous self-awareness and frank declarations about industry hype. The conversation transitions smoothly from rigorous venture calculus to a thoughtful, reflective examination of how cognitive automation might permanently alter human purpose.

## Key takeaways - [00:04](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=4s) Only about 0.1% (10 basis points) of the global workforce actively uses AI agents daily, revealing a massive gap between Silicon Valley adoption and broader real-world utility. - [00:04](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=4s) Venture investment is shifting from traditional SaaS tooling toward 'Neofirms'—tech-enabled services delivering autonomous AI professionals directly to clients in fields like law and wealth management. - [02:47](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=167s) While foundational token consumption and frontier lab revenues are scaling rapidly, an early-stage valuation bubble exists around lab spinout founders who lack viable business models. - [04:20](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=260s) Open-sourcing model weights is primarily a commercial and competitive maneuver rather than pure altruism, often employed by market participants trailing behind the frontier. - [06:52](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=412s) Tech giants and geopolitical players strategically open-source AI to commoditize complementary layers, from Nvidia boosting GPU demand to Meta avoiding ecosystem lock-in. - [06:52](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=412s) US chip export restrictions have not halted Chinese AI progress, instead catalyzing domestic innovation such as training frontier models on Huawei silicon. - [10:10](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=610s) Most vertical SaaS startups should bootstrap directly to profitability, as venture capital is fundamentally designed to finance high-risk, capital-intensive R&D. - [10:10](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=610s) AI is unlikely to create mass structural unemployment; technological revolutions historically automate repetitive labor while generating entirely new categories of human work. - [12:53](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=773s) Venture mega-funds are expanding because leading AI hyperscalers (like OpenAI and Anthropic) are growing revenue far faster than the Magnificent Seven did in previous decades. - [16:20](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=980s) Speculative infrastructure overbuilds and financial bubbles are historical necessities that build the foundational rails required for downstream consumer breakthroughs. - [16:20](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=980s) Over-relying on LLMs for synthesis and writing risks creating 'cognitive debt,' causing intellectual capabilities to atrophy unless society develops deliberate mental exercises. - [19:25](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=1165s) As raw intellectual synthesis becomes automated and commoditized by AI, societal prestige and human value may migrate away from raw intelligence toward other human qualities.

## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=4s) AI Adoption Discrepancies and the Rise of "Neofirms"

Ben Bergman kicks off the conversation with a nod to the brutal weather outside, joking that the term "fireside chat" takes on a whole new meaning when it is 113 degrees in the desert and questioning who would join the downstairs run club in such heat. Jahanvi Sardana introduces herself and the philosophy of Index Ventures, explaining their global mandate to become the largest and most valuable shareholder in category-defining tech companies—having backed foundational players like Wiz, Scale AI, Figma, and Robinhood. Sardana pokes fun at her internal reputation among Index partners for being drawn to "complicated businesses," noting that she is currently trying to rebrand herself.

When asked what currently excites her, Sardana highlights the stark dichotomy between the Silicon Valley echo chamber and the rest of the economy. While tech hubs like San Francisco are saturated with developers running "a bajillion agents and Mac minis," the global reality is that only about 10 basis points (0.1%) of the worldwide workforce actively uses AI agents in their day-to-day workflow. Even though ChatGPT has amassed 800 million users, many interact with it merely once a week. Sardana sees the true opportunity in bringing genuine AI utility to regular people—like her mother—which she believes will spawn a completely new generation of consumer and prosumer software.

Beyond consumer applications, Sardana outlines two core investment frontiers: - **Neofirms (Tech-Enabled Services):** Rather than selling traditional SaaS tooling to doctors and lawyers, Neofirms completely reinvent vertical service sectors by delivering autonomous AI lawyers or AI wealth managers directly to clients. - **Cybersecurity and Infrastructure "Plumbing":** Paralleling historical tech shifts where "you build the city first and then plumbing comes along" (such as Wiz scaling 20 years into cloud adoption or LinkedIn creating Kafka), infrastructure and security layers around AI are still in their infancy.

[02:47](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=167s) Token Demand Realities and Early-Stage Valuation Bubbles

Addressing the topic of frothy market valuations, Bergman asks where the real AI bubble is concentrated. Sardana points out that while public markets have already seen significant pullbacks—with several prominent AI-related stocks tumbling 40% to 50%—underlying foundational demand remains exceptionally strong. Across the ecosystem, token consumption continues to surge, GPU pricing remains high, and frontier lab revenues (such as Anthropic and OpenAI) are scaling rapidly past historical SaaS benchmarks.

However, Sardana draws a firm line when it comes to early-stage venture funding. She expresses skepticism toward the rush to fund researchers spinning out of major artificial intelligence labs with massive seed valuations simply because their peers raised significant capital. Emphasizing fundamental economic principles, she remarks that "capitalism is ultimately really boring and that means that you have to like earn more than you spend." For startups raising aggressively without clear business models or revenue plans, a dangerous valuation bubble definitely exists.

[04:20](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=260s) Closed Frontier Models vs. Open-Weight Competition

Bergman pivots to the rise of competitive open-weight models emerging from China, citing Moonshot AI’s Kimi model reaching performance parity with top proprietary models on specific benchmarks, and asks whether this poses an existential threat to Index's investment in Anthropic. Sardana clarifies that Index embraces an "abundance mindset" rather than a zero-sum view, holding stakes in both closed frontier labs like Anthropic and open-weight infrastructure providers like Fireworks AI.

While acknowledging Kimi's technical milestones—such as its context window and parameter scale—Sardana argues that claims around open-weight models overtaking the frontier are often overstated. Benchmark achievements often match older model checkpoints rather than the real-time frontier, and in artificial intelligence development, a lead of even a few months represents an enormous operational advantage. Furthermore, she demystifies the idealism surrounding open source:

> "Nobody's open sourcing stuff from the goodness of their hearts, right? Like these things are really expensive to train. You need a ton of compute. So people are not just giving it away because they love the world."

She explains that open releases are primarily commercial distribution strategies, noting that models like Kimi still enforce licensing agreements once commercial thresholds (such as $20 million in revenue) are met. When Bergman playfully presses her on whether Anthropic is planning an initial public offering in the fall, Sardana laughs it off, stating, "Can't comment. I know as much as you do."

[06:52](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=412s) The Geopolitics and Business of Open Source

Investigating why an overwhelming share of open-weight models originate from China, Sardana frames open-source strategy strictly as a pragmatic commercial maneuver: "You preach open when you're behind... when you're second place." She maps out how key market participants use open source to commoditize their complements: - **Nvidia (Jensen Huang):** Championing open ecosystems commoditizes software layers and prevents enterprise lock-in to proprietary stacks, ultimately driving higher demand for Nvidia GPUs over custom silicon (ASICs). - **Meta (Mark Zuckerberg):** Open-sourcing Llama prevents Meta from becoming structurally dependent on closed frontier labs. - **Chinese Tech Ecosystem:** Strategically open-sourcing weights helps bypass US chip export controls and aligns with national top-down industrial priorities. By establishing the global open software layer, China aims to commoditize foundation models while monetizing its advanced lead in manufacturing and robotics.

When frontier teams actually secure the top spot, their incentives instantly flip toward proprietary lock-in, as seen when Alibaba's Qwen or MiniMax initially restricted weights before public pressure or corporate mandates shifted their posture. Regarding US chip restrictions, Sardana argues they have largely failed to halt progress, observing that "when there is scarcity and creativity that's almost always better than abundance and creativity." As evidence, she points to Chinese delivery giant Meituan successfully training frontier-grade models on domestic Huawei silicon rather than restricted US chips.

[10:10](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=610s) The Myth of Job Extinction and the Two-Person Unicorn

The discussion turns to the popular concept of the "two-person unicorn" and whether ultra-lean, AI-native startups threaten the traditional venture capital model. Sardana embraces the concept, arguing that the majority of modern companies—particularly routine vertical SaaS providers targeting small niches—should never take venture capital in the first place and would be far healthier bootstrapping directly to profitability.

Venture capital, she argues, exists specifically to fund high-risk, capital-intensive research and development before commercial revenue materializes. She points out that the leading AI hyperscalers (Anthropic, OpenAI, SpaceX) are consuming more capital and hiring talent faster than anyone else, directly contradicting their own rhetoric about AI eliminating human labor.

Addressing long-term labor fears, Sardana flatly rejects the idea that AI will cause widespread structural unemployment: - Short-term displacement will automate mundane, repetitive tasks. - Historically, technological revolutions displace labor only to birth entirely new, previously unimaginable industries; the tech conference they are sitting in would have made no sense during the Industrial Revolution. - Sardana jokes that "most jobs are fake jobs—like my job is a fake job, your job is a fake job," and predicts humans will invent countless new roles across hospitality, arts, small businesses, and entertainment.

[12:53](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=773s) Mega-Funds, Artisanal Investing, and Hyperscale Valuation Math

Bergman highlights the massive wave of venture fundraises across the industry—such as Index Ventures raising $3.5 billion, Andreessen Horowitz pulling in $15 billion, Thrive Capital securing $10 billion, and Sequoia Capital raising $7 billion. Sardana positions Index not as a bloated mega-fund, but as "scaled artisans" who maintain rigorous, high-conviction selection while deploying capital at scale.

``` VENTURE CAPITAL FUND SIZE DEBATE ┌──────────────────────────────────────────────┐ │ The Skeptical LP View: │ │ • Most VC funds trail public market returns. │ │ • S&P 500, QQQ, & Mag 7 generated 20%+ IRRs │ │ over the last decade. │ └──────────────────────┬───────────────────────┘ │ ▼ ┌──────────────────────────────────────────────┐ │ The Generational Hyperscaler Thesis: │ │ • SpaceX, OpenAI, Anthropic scaling faster │ │ than any prior tech cohort. │ │ • Projected ~$1T combined revenue in 5 yrs │ │ (Mag 7 took two decades to hit $1T). │ │ • Companies staying private far longer. │ └──────────────────────────────────────────────┘ ```

Sardana walks through the math of why massive fund sizes can be justified today. While most venture firms historically fail to beat baseline index benchmarks like the S&P 500, Nasdaq-100 (QQQ), or the Magnificent Seven (which yielded 20%+ IRRs over the last decade), the current wave of market leaders is scaling at unprecedented speeds. Whereas the Magnificent Seven took two decades to achieve a collective $1 trillion in revenue, the triumvirate of SpaceX, OpenAI, and Anthropic is on trajectory to approach that run-rate in a fraction of the time.

She explains that Index's decision to participate in Anthropic at an unusually high valuation was driven by this reality. Despite Index's strict ethos of early entry and high ownership, the launch and developer traction of Claude Code signaled recursive self-improvement—where AI directly builds better AI—making it a generational, venture-scale bet even at a growth stage.

[16:20](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=980s) Infrastructure Overbuilds and the Threat of "Cognitive Debt"

On the question of whether the current AI market constitutes an overbuild, Sardana argues that speculative infrastructure bubbles have historically been necessary engines of societal progress. Just as the buildouts of the Industrial Revolution, electricity grids, and fiber-optic networks created temporary financial crashes, they ultimately left behind essential rails that unlocked massive downstream consumer value.

Asked for a provocative prediction or "hot take," Sardana turns to the psychological and intellectual toll of generative AI. She observes that every major technological shift causes human capabilities to atrophy when outsourced:

- **The Industrial Revolution:** Physical labor was outsourced to machines, causing human muscles to atrophy and necessitating the invention of gyms and structured exercise. - **The Internet Era:** Information retrieval was outsourced to search engines and algorithmic feeds, degrading human attention spans. - **The AI Era:** Cognitive synthesis and problem-solving are outsourced to large language models, threatening intellectual depth.

Citing academic research on "cognitive debt," Sardana points to control-group studies where students who outsourced essay composition to AI showed quantifiable depletion in neural pathway engagement compared to those who wrote manually. She cautions that society urgently needs to invent "gyms for our brains," admitting that even in her own life, she finds herself having to consciously resist the reflex to outsource critical thinking and everyday writing to automated assistants.

[19:25](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=1165s) The Future Value of Intelligence and Closing Remarks

Reflecting on how people interact with emerging technology and younger generations, the discussion turns to the simple advice: "Why don't you just ask your kids, talk to your kids, talk to your children?" This prompts a philosophical consideration about the long-term impact of artificial intelligence on human cognition and self-worth. It is suggested that perhaps society's current paradigm may shift entirely; right now, an immense amount of societal value is placed strictly on intelligence, but in an AI-dominated future, raw human intelligence might become largely irrelevant as value migrates toward entirely different human traits or capabilities.

Concluding on what is wryly dubbed that "optimistic note," the fireside chat wraps up. The speakers express their mutual gratitude, thanking each other and the audience members for joining the session before bringing the conversation to a close.

## Notable quotes > [00:04](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=4s) fireside chat — Ben Bergman

> [00:04](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=4s) a bajillion agents and Mac minis — Jahanvi Sardana

> [00:04](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=4s) you build the city first and then plumbing comes along — Jahanvi Sardana

> [02:47](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=167s) capitalism is ultimately really boring and that means that you have to like earn more than you spend. — Jahanvi Sardana

> [04:20](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=260s) Nobody's open sourcing stuff from the goodness of their hearts, right? Like these things are really expensive to train. You need a ton of compute. So people are not just giving it away because they love the world. — Jahanvi Sardana

> [04:20](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=260s) Can't comment. I know as much as you do. — Jahanvi Sardana

> [06:52](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=412s) You preach open when you're behind... when you're second place. — Jahanvi Sardana

> [06:52](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=412s) when there is scarcity and creativity that's almost always better than abundance and creativity. — Jahanvi Sardana

> [10:10](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=610s) most jobs are fake jobs—like my job is a fake job, your job is a fake job — Jahanvi Sardana

> [12:53](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=773s) scaled artisans — Jahanvi Sardana

> [16:20](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=980s) gyms for our brains — Jahanvi Sardana

> [19:25](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=1165s) Why don't you just ask your kids, talk to your kids, talk to your children? — Ben Bergman

## Mentioned Jahanvi Sardana, Ben Bergman, Index Ventures, Business Insider, Wiz, Scale AI, Figma, Robinhood, ChatGPT, Kafka, LinkedIn, Anthropic, OpenAI, Moonshot AI, Kimi, Fireworks AI, Nvidia, Jensen Huang, Meta, Mark Zuckerberg, Llama, Alibaba, Qwen, MiniMax, Meituan, Huawei, SpaceX, Andreessen Horowitz, Thrive Capital, Sequoia Capital, Claude Code

## Video description Jahanvi Sardana - Partner, Index Ben Bergman - Senior Correspondent, Business Insider

[Source](https://www.youtube.com/watch?v=sCO0rkCXnUA)

中文简报

前沿AI实验室估值逼近万亿美元，Index合伙人捍卫巨型基金押注开放权重模型是落后竞争对手的商业工具而非出于利他主义；尽管早期种子轮存在泡沫，前沿实验室前所未有的收入增长轨迹证明了后期巨额基金估值的合理性。 Index Ventures合伙人Jahanvi Sardana剖析了硅谷对AI的热情与现实落地应用之间的显著脱节，指出了垂直“新型企业”（neofirms）与底层基础设施中蕴含的高回报投资机遇。尽管她对实验室衍生项目虚高的早期估值持怀疑态度，但她捍卫了超大规模巨型基金的布局，并将开放权重模型的发布重新定义为务实的商业策略而非纯粹的利他主义。Sardana驳斥了关于结构性失业的灾难性预测，并在最后警告称，随着人类越来越将批判性思维外包给自动化模型，“认知负债”（cognitive debt）正日益加剧。全球仅约 0.1%（10 个基点）的劳动力在日常工作流中使用 AI agent，硅谷的狂热与现实世界的大规模实际效用之间仍存在巨大鸿沟。 (0:04) 风险投资正从传统 SaaS 工具转向“Neofirms”（技术赋能型服务），即直接向客户交付自主 AI 律师或财富管理师等专业服务。 (0:04) 底层 Token 消耗与前沿实验室收入正高速增长，但在缺乏可行商业模式、单纯依赖名校实验室背景的早期初创公司中存在危险的估值泡沫。 (2:47) 开源模型权重本质上是落后者的商业与竞争策略，而非纯粹利他；同时美国芯片出口管制未能阻碍中国 AI 研发，反而倒逼出在华为芯片上训练前沿模型的创新。 (6:52) AI 不会导致大规模结构性失业，科技革命通常会自动化重复劳动并催生全新行业，但过度将思考外包给 LLM 可能带来认知能力退化的“认知负债”风险。 (10:10)

Reading summary

Index Ventures partner Jahanvi Sardana argues that despite public market pullbacks and an early-stage venture valuation bubble, underlying demand for artificial intelligence remains robust as foundational companies scale faster than previous tech giants. However, she contends that broad adoption is still in its infancy and warns of emerging intellectual risks such as cognitive atrophy. Adoption Realities and the Rise of Neofirms (0:04) Sardana contrasts the Silicon Valley bubble with the broader global economy, estimating that only roughly 10 basis points (0.1%) of the worldwide workforce actively uses AI agents in daily workflows. Even with ChatGPT amassing 800 million users, many engage with the platform only once a week. Sardana sees major investment potential in building intuitive software for mainstream consumers as well as backing "Neofirms"—tech-enabled vertical service companies that deploy autonomous agents directly to clients instead of selling standard SaaS tools to professionals like doctors or lawyers. Valuation Bubbles Versus Foundational Demand (2:47) While acknowledging that several public AI-related stocks have retreated 40% to 50%, Sardana maintains that core demand remains strong, evidenced by rising token consumption, high GPU prices, and rapidly growing revenues at frontier labs like OpenAI and Anthropic. At the same time, she identifies a dangerous bubble in early-stage seed funding, where venture capital chases spinout researchers at high valuations without clear business models or paths to profitability, running counter to the basic economic imperative of earning more than spending. Strategic Incentives Driving Open-Weight Models (6:52) Sardana demystifies the open-source movement, stating that teams "preach open when you're behind" and use open-weight releases as commercial distribution strategies to commoditize complementary layers. Meta open-sources Llama to avoid reliance on closed frontier labs, Nvidia supports open ecosystems to prevent software lock-in and spur GPU demand over custom chips, and Chinese firms utilize open weights to bypass U.S. export restrictions while pairing software with domestic manufacturing. She notes that U.S. export restrictions have not halted progress, pointing to Meituan training models on Huawei silicon, while maintaining that frontier proprietary labs preserve critical technical leads. Labor Market Evolution and Hyperscale Math (10:10) Dismissing concerns over mass technological unemployment, Sardana argues that AI will automate mundane tasks while creating previously unimaginable industries, noting that top AI firms are rapidly increasing hiring and capital expenditure. She justifies mega-fund sizes by projecting that leading private hyperscalers like SpaceX, OpenAI, and Anthropic could approach a combined $1 trillion in revenue far quicker than the two decades it took the Magnificent Seven, which drove Index's later-stage investment in Anthropic following evidence of recursive self-improvement in Claude Code. Historical Infrastructure Bubbles and Cognitive Debt (16:20) Drawing parallels to the Industrial Revolution, electricity grids, and fiber optics, Sardana views potential infrastructure overbuilding as a necessary prelude to downstream consumer value. Her primary concern centers on the psychological toll of AI, specifically "cognitive debt." Just as mechanization led to muscular atrophy and search engines degraded attention spans, Sardana cites studies showing depleted neural engagement among students who outsource writing to AI, warning that society will need to establish intellectual equivalents of physical gyms to preserve critical thinking. The discussion closes on an unresolved question regarding the long-term societal value of human intelligence, suggesting that if generative models commoditize analytical tasks, human worth and economic rewards may shift away from raw intellect toward entirely different capabilities.

精炼阅读摘要

Index Ventures合伙人Jahanvi Sardana指出，尽管硅谷内部呈现AI过度饱和，但全球实际工作场景中的AI渗透率仅约0.1%，未来的核心商业增量在于重构垂直专业服务的“Neofirms”以及基础设施底层；同时，早期早期种子轮存在脱离经济常识的估值泡沫，而开源权重多为落后竞争者的商业分发策略，未来社会更需警惕过度依赖AI造成的“认知债务”。采纳鸿沟与新型垂直服务业态 (0:04) Sardana强调了硅谷技术圈与实体经济之间的巨大断层。在旧金山等技术中心，开发者已经密集部署了海量智能体和硬件环境，但从全球劳动力市场来看，仅有约0.1%（10个基点）的工作者在日常工作流中真正使用AI Agent。即便是月活达到8亿的ChatGPT，多数用户的实际使用频率也仅为每周一次。因此，她认为真正的机会在于将具备实际效用的AI工具推向大众市场，并催生两类代表性投资方向。其一是被称作“Neofirms”的技术驱动型服务公司。这类企业不再沿袭向医生或律师出售传统SaaS软件的路径，而是直接面向终端客户交付由自主AI驱动的法律或财富管理等专业服务。其二则是网络安全与底层基础设施的“管道工程”。Sardana援引云计算普及二十年后Wiz才实现爆发、以及LinkedIn孵化Kafka的历史经验指出，技术周期通常是“先建城市，后铺管道”，当前围绕AI的安全与基建层仍处于起步阶段。早期估值泡沫与开源权重的商业博弈 (2:47) 谈及市场泡沫时，Sardana对二级市场与一级市场作出了明确区分。尽管部分公开上市的AI概念股已从高点回撤40%至50%，但底层的真实算力与应用需求依然强劲，代币消耗持续攀升，GPU价格高企，头部前沿实验室（如OpenAI和Anthropic）的收入增速显著超越了历史SaaS增长基准。然而，早期风险投资领域正面临严峻的估值泡沫。部分从顶尖实验室分拆出的研究团队在缺乏清晰商业模式或盈利路径的情况下盲目推高种子轮估值，违背了资本主义“收入必须大于支出”的基本常识。针对开源权重对闭源巨头的挑战，Sardana认为宣称开源模型完全超越前沿闭源水平的言论被夸大了，几周或几个月的迭代领先在前沿领域构成了巨大的商业护城河。她直言“没有机构是出于纯粹的善心而开源高昂的模型”，开源实则是第二梯队竞争者为了商品化其互补品而采取的商业分发策略。例如，Nvidia倡导开源软件以防止企业被专有软件栈锁定，从而推动其GPU硬件销售；Meta开源Llama旨在避免受制于闭源巨头；中国企业开源则意在规避芯片出口限制，并利用其在制造与机器人领域的优势实现变现。此外，Sardana观察到美团在华为昇腾芯片上成功训练前沿模型的案例，表明外部技术封锁未能彻底遏制其创新进程。超级巨头扩张逻辑与劳动形态重构 (10:10) 针对大型风投机构近期募集巨额基金的争议，Sardana将Index Ventures定位为兼具严苛筛选标准与规模化资金的“规模化手艺人”。她承认历史上多数风投基金的回报无法击败标普500、纳斯达克100（QQQ）或过去十年年化内部收益率超过20%的“美股七巨头”，但当前以SpaceX、OpenAI和Anthropic为代表的超级巨头正在以史无前例的速度规模化。七巨头耗费了二十年才实现1万亿美元的总营收，而这三家新兴领军企业有望在更短的时间周期内达到该量级，且企业维持私有化状态的时间大幅延长。基于此逻辑，Index参与了Anthropic高估值的成长轮融资，尤其看好Claude Code所展现出的AI自主迭代开发潜力。在劳动力与就业议题上，Sardana驳斥了AI将引发全面结构性失业的悲观预期。她认为，虽然重复性、机械性任务在短期内会面临替代，但从工业革命以来的历史规律看，技术革命必然会催生此前难以想象的全新行业与岗位。未来人类社会将创造出更多分布在艺术、娱乐、酒店服务及个性化商业中的全新角色，技术巨头当前仍在以极高速度扩张并争抢顶尖人才的事实本身也反驳了劳动力迅速消亡的论调。基础设施建设浪潮与认知债务隐忧 (16:20) 针对AI基础设施是否存在过度建设的质疑，Sardana认为投机性基建泡沫往往是推动社会跨越式发展的必然动力。正如工业革命、电网铺设和光纤网络早期的金融泡沫一样，短期的资本过剩和价格回落最终留下了至关重要的基础设施网络，从而在长期释放巨大的下游消费价值。然而，Sardana对AI带来的心理与智力退化表达了深切担忧，并提出了“认知债务”（Cognitive Debt）的概念。回顾技术发展史：工业革命外包了体力劳动导致人类肌肉萎缩，促成了健身房的诞生；互联网外包了信息检索，削弱了人类的注意广度；而生成式AI正在将综合思考与复杂问题解决外包给大语言模型。她引述对照组学术研究指出，使用AI代写文章的学生在神经通路活跃度上呈现量化下降。她呼吁社会亟须建立针对大脑的“思维健身房”，并警惕人类个体在不知不觉中丧失深度批判性思考与书写表达的能力。对话最终探讨了技术对人类价值维度的重构：当智能本身被AI高度平民化与普遍化之后，社会对人类个体价值的衡量标准可能会发生质的转变，纯粹的原始智力或许将不再是决定社会地位的核心指标。然而，估值极高的新兴巨头能否如期兑现其万亿美元收入预期，以及人类社会能否在不可逆的认知退化发生前构建起有效的思维防御机制，依然是留待实践验证的核心悬念。

Detailed notes — zh

## Background 本内容由 Ai4 频道于 2026 年 8 月 22 日发布在 YouTube 平台。视频记录了 Index Ventures 合伙人 Jahanvi Sardana 与《商业内幕》（Business Insider）资深记者 Ben Bergman 之间的一场炉边漫谈。 ## TL;DR 对话深入探讨了 Index Ventures 的最新 AI 投资逻辑，指出尽管旧金山科技圈沉浸于 Agent 热潮，但全球大众与劳动力端的实际渗透率仍处于极早期的 0.1% 水平。Jahanvi 剖析了当前早期种子轮投资的估值泡沫，并从商业博弈视角解析了开源与闭源大模型的生态定位，以及芯片出口管制下中国大模型的突围创新。此外，她证实了机构以 1800 亿美元估值重注 Anthropic 的底层逻辑，并提出了一个反共识警示：随着人类将思考过程全盘外包给 AI，社会正面临陷入“认知负债”的潜在风险。 ## Atmosphere & style 整场对谈在 113 华氏度的极端高温户外展台展开，主持人幽默自嘲的开场瞬间拉近了现场距离。Jahanvi 的表达干练、犀利且极具商业穿透力，既有对万亿级超级巨头与万亿参数大模型的宏大洞察，也有对真实商业本质与生活常识的冷静回归。对谈后半段从硬核投资逻辑自然过渡到对人类“智能贬值”与“认知负债”的哲学反思，在冷峻警示与幽默化解的交织中给观众带来深刻的思想冲击。 ## Key takeaways - [00:04](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=4s) AI 渗透率与行业认知存在巨大鸿沟：尽管旧金山科技圈沉浸在多 Agent 开发中，但全球仅约 0.1% 的劳动力真正使用 Agent，AI 普及仍处于极其早期的阶段。 - [00:04](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=4s) Index Ventures 当前重点关注三大 AI 方向：面向大众的消费级与 Prosumer 应用、颠覆现代服务业的“新形态机构（Neofirms）”，以及落后于应用爆发的网络安全与基础设施。 - [02:47](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=167s) 当前 AI 赛道的终端算力与 Token 调用需求依然强劲，真正的泡沫主要集中在缺乏清晰商业模式与运营能力的早期孵化和种子轮投资阶段。 - [04:23](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=263s) 开源与闭源并非零和博弈：开源本质上是商业策略与分发通道，通常是落后者用于挑战领先者或硬件厂商用于拉动算力消耗的手段，而前沿实验室的数月领先优势构成了决定性的护城河。 - [06:52](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=412s) “当你处于落后位置时才会宣扬开源”：NVIDIA 开源是为了商品化互补品以卖出更多 GPU，Meta 开源是为了防止被闭源巨头卡脖子，而一旦开源团队逼近技术最前沿，同样会趋向闭源。 - [06:52](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=412s) 美国对华芯片出口管制并未达到预期效果，“匮乏加创造力往往胜过丰饶加创造力”，中国企业在算力受限下依然能利用国产与非顶尖芯片攻克前沿模型训练。 - [10:06](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=606s) “两人独角兽”模式的兴起不会消灭风险投资，大部分垂直初创企业本应走自力更生的盈利路线，而 VC 的核心使命是为大规模硬核研发提供前置资金支撑。 - [10:06](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=606s) AI 不会从根本上消灭人类就业，历史上大多数工作都是技术变革创造出来的“人造岗位”，社会将不断演化出全新的商业、服务与人际互动形态。 - [12:52](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=772s) Index Ventures 以 1800 亿美元估值重注 Anthropic，是基于押注下一代超级云厂商/超大规模计算平台（Hyperscaler）以及 AI 递归式自我提升的战略判断。 - [17:16](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=1036s) 全盘外包思考正在让人类陷入“认知负债”：正如机器让肌肉退化、算法降低专注力，过度依赖 AI 导致人类大脑神经通路活跃度下降，社会亟需建立“大脑的健身房”。 - [19:25](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=1165s) 未来人类“智力”的溢价或将被重新评估：当 AI 能够廉价提供高智商输出时，单纯的聪明才智可能变得不再关键，人类社会的核心价值将转向其他专属特质。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=4s) 极热天气的开场与 Index Ventures 的 AI 投资全景活动在室外高达 113 华氏度的极端高温下开场，主持人 Ben Bergman 幽默地调侃在这样的天气里举办“炉边漫谈”（Fireside Chat）别有一番风味，楼下居然还有跑步俱乐部，简直不可思议。随后，Index Ventures 合伙人 Jahanvi Sardana 介绍了自己和机构的投资理念。Index Ventures 是一家全球顶尖的风投机构，曾早期投资了 Wiz、Scale AI、Figma 以及 Robinhood 等定义行业赛道的领军企业。Jahanvi 笑称自己在合伙人当中以“专门投资复杂业务模式”而闻名，目前正试图调整这一定位。谈及当下最令她兴奋的投资方向，Jahanvi 指出当前科技行业存在巨大的认知割裂与反差：身处旧金山的从业者整天参加 AI 会议，看着工程师们摆弄成百上千个 Agent 和 Mac mini，会误以为 AI 已经彻底普及并饱和；但实际上，全球只有约 10 个基点（即 0.1%）的劳动力在真正使用 Agent 并从中获益。即使像 ChatGPT 这样拥有 8 亿用户的全球级现象产品，多数普通人也仅是一周使用一次。真正让她兴奋的是 AI 如何走进普通人的真实生活（比如她的母亲），让大众切实体会到 Agent 的价值。 Jahanvi 重点分享了她当前关注的三大核心领域： ```mermaid flowchart TD accTitle: Index Ventures AI Investment Thesis accDescr: Jahanvi Sardana's core AI investment focus areas at Index Ventures. subgraph CoreThemes["Index Ventures 核心关注赛道"] A["消费级与专业消费级 AI 面向全球 99.9% 尚未渗透的普通大众"] B["新形态机构 (Neofirms) 技术赋能型服务：AI 律师、AI 财富管理师"] C["底层基础设施与网络安全 先建城市后修下水道：新一代 Plumbing 设施"] end ``` - **消费级与专业消费级（Prosumer）应用**：处于极其早期的渗透阶段，蕴藏着催生新一代大众级 AI 产品的巨大机遇。 - **“新形态机构”（Neofirms）**：一种技术赋能型服务的新商业模式。不再仅仅是向律师或医生兜售软件工具，而是利用 AI 彻底重塑整个现代服务业，打造真正的“AI 律师”或“AI 财富管理师”。 - **网络安全与基础设施**：底层设施建设永远滞后于技术浪潮。Jahanvi 打比方说，“通常都是先建好城市，随后才会铺设下水管道”。就像云计算爆发 20 年后才诞生了 Wiz，LinkedIn 发展到一定阶段才催生了 Kafka 一样，AI 基础设施的构建才刚刚拉开序幕。 --- ### [02:47](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=167s) 资本泡沫、算力需求与早期投资的估值乱象针对 Index Ventures 刚刚完成的新一期 35 亿美元基金募集，Ben Bergman 抛出了关于当前市场估值过高与泡沫的疑问。今年二级市场部分 AI 概念股已从高点回撤 40% 至 50%，引发了广泛担忧。Jahanvi 对此回应称，任何具有划时代颠覆性的新技术浪潮都会在局部催生泡沫，但判断是否存在泡沫的最核心标准是“底层真实需求是否出现断崖式下跌”。在她看来，目前全球对 Token 与算力的需求不仅没有破裂，反而呈现出极其强劲的爆发力。例如 Anthropic 的业务规模已经超越了大部分传统软件公司的总和，OpenAI 的 Token 调用量持续飙升，半导体领域的 GPU 价格依然居高不下。只要终端计算需求真实存在，AI 赛道就仍有巨大的发展纵深。然而，Jahanvi 尖锐地指出了真正存在严重泡沫的环节——**早期孵化与种子轮投资**： - 许多来自大厂顶尖实验室的研究员，看到身边同侪轻松拉到巨额融资后便纷纷跟风创业，而市场上充斥着盲目抢夺项目的热钱。 - “资本主义的本质终究是非常无聊且朴素的——你必须赚得比花得多（earn more than you spend）。” - 大量高估值的早期 AI 初创团队既没有清晰的商业模式，也对公司实际运营毫无概念，这正是当下最为危险的泡沫所在。 --- ### [04:23](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=263s) 开源与闭源之争：中国开源大模型与商业博弈谈及 Index Ventures 投资的 Anthropic，Ben Bergman 提及近期中国月之暗面（Moonshot AI）发布的开源模型 Kimi 3（在多项基准测试中逼近顶尖专有模型），并质疑这是否会削弱 Anthropic 等闭源前沿模型的付费壁垒。 Jahanvi 阐述了她的“丰饶思维”（Abundance Mindset）：开源与闭源绝非零和博弈，两者拥有各自的生态位。Index Ventures 既重注了闭源模型巨头 Anthropic，也投资了提供开源权重模型推理托管服务的平台 Fireworks AI。针对 Kimi 3 的表现，Jahanvi 分析道： - **技术基准客观评估**：Kimi 3 凭借 2.8 万亿参数与超大上下文窗口确实令人惊艳，但外界对其评测表现有一定程度的夸大。在具体任务中，它逼近的是 Fable（年初训练的模型 Mythos 的精简版）。在日新月异的 AI 领域，前沿闭源实验室拥有的数月领先优势就是决定性的护城河。 - **开源背后的商业本质**：“没有人是出于纯粹的善意去开源模型的。”大模型预训练需要耗费天文数字的算力与资金，开源本质上是一种商业策略或分发通道。即使是像 Kimi 这样的开源权重模型，在达到特定营收规模（如 2000 万美元）后也会触发商业授权协议，并非毫无保留的慈善行为。当被问及 Anthropic 是否会在今年秋季启动 IPO 时，Jahanvi 笑着表示自己无法置评，“我知道的和你一样多”。 --- ### [06:52](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=412s) 战略开源逻辑与美国芯片出口管制的实际成效讨论深入到为何当前诸多开源权重模型集中涌现于中国及部分特定巨头。Jahanvi 提出了深刻的商业竞争法则：“**当你处于落后或第二名的位置时，你才会极力宣扬开源**（You preach open when you're behind）。” ```mermaid flowchart LR accTitle: Strategic Drivers of Open Source in AI accDescr: Comparison of motivations behind open-sourcing models for market challengers vs leading frontier labs. subgraph Challengers["第二梯队 / 硬件生态巨头"] Nvidia["NVIDIA (黄仁勋) 商品化互补品，拉动 GPU 算力消耗"] Meta["Meta (扎克伯格) 打破前沿实验室垄断，建立开发者标准"] China["中国大模型厂商 规避制裁限制，依托制造/机器人软硬协同"] end subgraph FrontierLeaders["第一梯队 / 领跑者"] Frontier["Anthropic / OpenAI 保护核心技术 IP（如同可口可乐秘方）"] end Challengers -->|采取开源策略| Market["开源生态格局"] FrontierLeaders -->|采取闭源策略| Market ``` - **NVIDIA 的算盘**：黄仁勋倡导开源是为了“将互补品商品化”（Commoditizing complements），模型越开源、软件越繁荣，市场对 NVIDIA 算力芯片的消耗就越大，从而抑制定制 ASIC 芯片的生存空间。 - **Meta 的动机**：马克·扎克伯格重注开源 Llama，是因为 Meta 在大模型底层研发上落后于顶级实验室，绝不希望未来受制于 OpenAI 或 Anthropic。 - **中国的战略考量**：开源既是规避美国先进芯片出口管制的一种软件协同突围手段，也是国家层面的顶层策略——在制造业和具身智能机器人领域实现跨越式领跑，通过开源模型使全球技术栈向其靠拢，再在其他环节变现。 - **领跑者的闭源必然性**：可口可乐绝不会公开配方。一旦原本开源的团队逼近科技最前沿，他们同样会转向闭源（如通义千问 Qwen 的特定前沿迭代）。关于美国对华芯片出口管制是否奏效，Jahanvi 直言“并不认为它成功了”。她指出：“**匮乏加创造力，往往胜过丰饶加创造力**（Scarcity and creativity is almost always better than abundance and creativity）。”比如中国本地生活巨头美团（Meituan）利用华为自研芯片成功训练出了最新的先进大模型，这证明在算力受限的倒逼之下，团队依然能用国产与非顶尖芯片攻克前沿模型训练难题。 --- ### [10:06](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=606s) “两人独角兽”与风投模式的演进：AI 真的会消灭工作吗？针对当下硅谷热议的“AI 时代只需两名员工就能打造一家 10 亿美元独角兽公司”的论调，Ben 询问这是否会颠覆风投模式本身的生存价值。Jahanvi 对此表示非常欢迎“两人独角兽”的出现： - **大部分公司不应拿 VC 的钱**：绝大多数垂直 SaaS 或细分领域的初创公司本就应该走 Bootstrap（自力更生）路线，依靠精简团队快速实现盈利。 - **风投的核心使命不会改变**：VC 的底层逻辑是“在商业收入跟上之前为大规模硬核研发提供资金支撑”。从 Anthropic、OpenAI 到 SpaceX 等新一代超级巨头来看，它们不仅没有减少融资，反而在以更惊人的规模募资和扩张团队。关于“AI 是否会取代人类就业”，Jahanvi 坚决表示“绝不可能”（Absolutely not）： - 短期内确实存在特定繁琐任务被 AI 自动化接管所带来的阵痛与岗位替代。 - 从人类历史长河看，“大多数工作本身就是被创造出来的‘人造岗位’（Most jobs are fake jobs）”。工业革命时期根本不存在现代意义上的会议产业或风投行业。 - 人类社会将不断创造出全新形态的工作——更多的餐饮、个性化小商业、新型娱乐消费以及难以预料的互动方式。长远来看，人类完全不必担忧失去工作。 --- ### [12:52](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=772s) 巨型风投基金的合理性与以 1800 亿美元重注 Anthropic 随着今年顶级风投基金规模不断膨胀（A16Z 募集 150 亿美元、Thrive Capital 100 亿美元、Sequoia 70 亿美元），相比之下 Index Ventures 的 35 亿美元显得较为克制。Jahanvi 将 Index 定位为“规模化手艺人”（Scaled Artisans），坚持精细化的投资节奏。谈及超级基金存在的合理性，Jahanvi 指出行业目前存在两种针锋相对的思潮： | 行业视角 | 核心论点与支撑数据 | |:--- |:--- | | **质疑视角** | 多数 VC 机构根本没有存在的必要，因为其中位数回报甚至跑不赢美股大盘（过去十年标普 500、QQQ 与美股七巨头 Mag 7 带来了 20%+ 的内部收益率 IRR）。 | | **巨型基金成立视角** | 新一代超级巨头的体量将远超以往。SpaceX、OpenAI、Anthropic 在未来 5 年内的总营收有望逼近 1 万亿美元，而科技七巨头当年实现这一体量花了整整 20 年。加之优质企业在私有市场停留的时间显著拉长，必须有海量资本支撑其成长。 | Jahanvi 证实了 Index Ventures 以 **1800 亿美元**的投后估值重注了 Anthropic 的成长期轮次。尽管这一估值体量对于传统追求极高持股比例和早期进入的 Index 来说是前所未有的破例，但如今在二级市场该估值已大幅攀升： - 这是一场押注“下一代超级云厂商/超大规模计算平台（Hyperscaler）”的极为罕见的世纪机遇。 - 当时 Claude Code 刚开始爆发，如果相信“AI 的递归式自我提升（Recursive Self-Improvement）”以及“代码生成是 AI 训练下一代 AI 最核心的抓手”，那么在那个关键时间节点重注 Anthropic 具有极佳的风险收益比。 - 历史上的工业革命、电力革命和互联网泡沫都经历了大建与过度建设，正是这种超前建设最终沉淀为惠及全社会与终端消费者的长期创新土壤。 --- ### [17:16](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=1036s) 深度思考与反共识预测：AI 是否正在让人类陷入“认知负债”？在漫谈的尾声，Ben Bergman 邀请 Jahanvi 抛出一个打破常规的反共识预测（Hot Take）。Jahanvi 提出了一个发人深省的警示：**AI 正在让我们变得越来越愚蠢**。她从技术进化对人类生理与心理机能的外包历史进行了纵向对比： - **工业革命**：人类将体力劳动外包给机器，导致肌肉机能退化，人类因而发明了“健身房”来主动锻炼体魄。 - **移动互联网时代**：信息过载与算法投喂将注意力外包，直接导致了人类专注力与耐心的大幅滑坡。 - **生成式 AI 时代**：人类正在把“智能本身”和思考过程全盘外包。 Jahanvi 援引了一项学术对照研究：研究人员对比了使用 AI 代写论文与独立撰写论文的对照组，发现过度依赖 AI 会导致大脑神经通路的活跃度下降，学者们将这种现象定义为“**认知负债**（Cognitive Debt）”。她呼吁业界深入思考人类是否需要建立“大脑的健身房”，并在日常工作与生活中保持警惕，强迫自己不要仅仅因为 AI 能做就将一切思考都外包给模型。Ben Bergman 也对此深有感触，承认自己如今遇到任何问题第一反应就是询问 GPT，甚至用《纽约时报》推荐的方法让 AI 剖析自己性格中的最大弱点，这种对算法的情感与认知依赖正在重塑人类的生活方式。 ### [19:25](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=1165s) 重新审视人类“智力”的价值与炉边谈话总结对话在尾声阶段触及了关于人际真实沟通以及人类核心价值的深刻反思。针对日常中人们常有的疑问——“为什么不直接去问问你的孩子，多和孩子们聊聊天呢？”（"Why don't you just ask your kids, talk to your kids, talk to your children?"），对话者顺势引申出了对技术浪潮下人类智力定位的哲思。一直以来，社会体制与文化体系都赋予了“智力”（intelligence）极高的衡量权重与溢价；然而随着人工智能的演进，未来单纯具备高智商或聪明才智可能变得不再那么至关重要，甚至会变得无关紧要（irrelevant），人类社会真正认可的核心价值或许将彻底转移到其他的维度与特质之上。面对这种打破传统认知、既深刻又略带冷峻的推论，主持人以展台特有的幽默感调侃道，就在这样一个“如此乐观的基调”（"on that optimistic note"）下结束今天的分享。随后，主持人向 Index Ventures 合伙人 Jahanvi Sardana 以及全场参会的听众致以谢意，正式为本场关于 AI 发展、投资与未来形态的炉边对谈画上了圆满的句号。 - **对智力定义的重构**：当今社会过度推崇和定价“智力”，但未来的价值创造重心可能会向智能之外的其他人类专属特质转移。 - **对谈闭幕**：主持人在幽默化解深层哲学思考的同时宣布对话结束，并对嘉宾与观众致以感谢。 ## Notable quotes > [00:04](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=4s) 通常都是先建好城市，随后才会铺设下水管道。 — Jahanvi Sardana > [02:47](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=167s) 资本主义的本质终究是非常无聊且朴素的——你必须赚得比花得多（earn more than you spend）。 — Jahanvi Sardana > [04:23](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=263s) 没有人是出于纯粹的善意去开源模型的。 — Jahanvi Sardana > [06:52](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=412s) 当你处于落后或第二名的位置时，你才会极力宣扬开源（You preach open when you're behind）。 — Jahanvi Sardana > [06:52](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=412s) 匮乏加创造力，往往胜过丰饶加创造力（Scarcity and creativity is almost always better than abundance and creativity）。 — Jahanvi Sardana > [10:06](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=606s) 大多数工作本身就是被创造出来的‘人造岗位’（Most jobs are fake jobs）。 — Jahanvi Sardana > [17:16](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=1036s) AI 正在让我们变得越来越愚蠢。 — Jahanvi Sardana > [19:25](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=1165s) Why don't you just ask your kids, talk to your kids, talk to your children? — Jahanvi Sardana > [19:25](https://www.youtube.com/watch?v=sCO0rkCXnUA&t=1165s) on that optimistic note — Ben Bergman ## Mentioned Jahanvi Sardana, Ben Bergman, Index Ventures, Business Insider, Anthropic, OpenAI, ChatGPT, Claude Code, Wiz, Scale AI, Figma, Robinhood, Fireworks AI, 月之暗面 (Moonshot AI), Kimi 3, NVIDIA, 黄仁勋, Meta, 马克·扎克伯格, Llama, 美团, 华为, SpaceX, 通义千问 (Qwen), Andreessen Horowitz (a16z), Thrive Capital, Sequoia Capital, LinkedIn, Apache Kafka, 纽约时报 (The New York Times) ## Video description Jahanvi Sardana - Partner, Index Ben Bergman - Senior Correspondent, Business Insider [Source](https://www.youtube.com/watch?v=sCO0rkCXnUA)

Source quotes

[0:04] Jahanvi Sardana: 通常都是先建好城市，随后才会铺设下水管道。 [2:47] Jahanvi Sardana: 资本主义的本质终究是非常无聊且朴素的——你必须赚得比花得多（earn more than you spend）。 [4:23] Jahanvi Sardana: 没有人是出于纯粹的善意去开源模型的。 [6:52] Jahanvi Sardana: 当你处于落后或第二名的位置时，你才会极力宣扬开源（You preach open when you're behind）。 [6:52] Jahanvi Sardana: 匮乏加创造力，往往胜过丰饶加创造力（Scarcity and creativity is almost always better than abundance and creativity）。 [10:06] Jahanvi Sardana: 大多数工作本身就是被创造出来的‘人造岗位’（Most jobs are fake jobs）。 [17:16] Jahanvi Sardana: AI 正在让我们变得越来越愚蠢。 [19:25] Jahanvi Sardana: Why don't you just ask your kids, talk to your kids, talk to your children? [19:25] Ben Bergman: on that optimistic note

Source quotes

[0:04] Ben Bergman: fireside chat [0:04] Jahanvi Sardana: a bajillion agents and Mac minis [0:04] Jahanvi Sardana: you build the city first and then plumbing comes along [2:47] Jahanvi Sardana: capitalism is ultimately really boring and that means that you have to like earn more than you spend. [4:20] Jahanvi Sardana: Nobody's open sourcing stuff from the goodness of their hearts, right? Like these things are really expensive to train. You need a ton of compute. So people are not just giving it away because they love the world. [4:20] Jahanvi Sardana: Can't comment. I know as much as you do. [6:52] Jahanvi Sardana: You preach open when you're behind... when you're second place. [6:52] Jahanvi Sardana: when there is scarcity and creativity that's almost always better than abundance and creativity. [10:10] Jahanvi Sardana: most jobs are fake jobs—like my job is a fake job, your job is a fake job [12:53] Jahanvi Sardana: scaled artisans [16:20] Jahanvi Sardana: gyms for our brains [19:25] Ben Bergman: Why don't you just ask your kids, talk to your kids, talk to your children?

Original transcript

[0:04] Thank you so much for joining me. Fireside chat has a different meaning when it's 113 [0:09] degrees outside. Also said there was a run club downstairs. I don't know if anyone did. Yeah. [0:14] Who's trying to run outside? No, that not a good idea. But uh we are excited to have you here and [0:20] um if you could just start out, you know, by saying what you invest in and what you are [0:25] most excited about right now. Yeah. Um really happy to be here. Thank you for having me. Um [0:31] I'm Janvi. I'm a partner at Index Ventures. We're a global venture capital firm investing [0:38] uh you know with a immodest goal of being the largest and most valuable shareholder in category [0:43] defining companies who were early investors in whiz scale Figma Robin Hood. Um and I'm [0:51] infamously known among my partners for better for worse for investing in complicated businesses. So [0:55] I'm trying to rebrand myself these days. Uh but to your question on what I'm excited about, I'm I'm [1:03] really excited about uh the dichotomy that we're living in right now. So, you know, you and I are [1:09] in San Francisco often. We go to AI conferences. We've got engineers with their bajillion agents [1:15] and Mac minis. So, we just feel like, oh my god, AI has saturated the world. And then you look at [1:20] this stat which is that around 10 basis points so.1% of the global workforce uses agents in any [1:29] shape or form and they could get a lot of value from it. So you know obviously chat GPT has um [1:34] has become like a global phenomenon but it's like 800 million users people use it once a week. So, [1:39] I'm excited about just the re normal people and real people like my mom getting value out of AI [1:46] agents. And I think we're really early innings of that. Um, and that could lead to maybe a whole new [1:51] generation of consumer and proumer companies. I'm also investing in this new business model which [1:58] is called Neoirms, which I'm calling Neoirms. It's essentially tech- enabled services. instead [2:03] of selling software to lawyers and doctors, can you really reinvent a whole sector and services [2:10] sector? So, an AI lawyer, an AI wealth manager and how you diffuse that um in society. And um lastly, [2:18] I also think we're early innings of like cyber security and and infrastructure. So, you know, [2:23] usually you build the city first and then plumbing comes along. So um whiz happened 20 years after [2:30] cloud and you needed LinkedIn for Kafka and um and so on. So I think we're just scratching the [2:37] surface there. And you guys just raised a new fund. That's right. Three billion. That's right. [2:42] New set of funds. Congratulations. All right. We'll get that shout out in. And uh yeah, so [2:47] we've seen a lot of um crazy valuations this year. I feel like I write about them every week. What do [2:53] you feel like is is most overvalued right now? It's a good question. Um, I think every wave and [3:05] new piece of technology that's as transformational leads to some bubbles forming somewhere. We're [3:11] actually seeing some of that reckoning in the public markets, right? Like all the AI stocks [3:15] down 40 50%. Um I think about bubbles in a very maybe like too much of a simplistic way like is [3:22] there demand right? Um and I don't see demand for tokens cracking anywhere. Like you look at [3:29] anthropic revenue like it's bigger than you know it's crossed 60 billion and it's bigger than most [3:34] of software companies combined. Open AI um just token demand even at the semis like GPU prices up. [3:42] So so I think as long as the demand is there you know there's there's just so much more that's left [3:49] to do in AI. But to your point on valuations are sky-high and where I do think we're in a bubble is [3:55] is really um at the early stage investing stage where you you're a researcher coming out of a [4:00] big lab you want to raise money because your friends raised a lot of money and everyone's [4:05] um you know ready to fund you and um look I think capitalism is ultimately really boring and [4:12] that means that you have to like earn more than you spend and these companies don't really have [4:16] business models or sense of what they're doing. So, I do think in in that sense we're we're in [4:20] a bubble. And you mentioned Anthropic, which you guys are investors in. Um, and you know, [4:26] I'm wondering how you feel about um you know, these Chinese models coming out cuz that's been [4:30] so interesting this summer. Of course, Moonshot AI released Kimmy 3, an open model that came close [4:36] to the leading proprietary models that happened a couple weeks ago. Does that concern you as far as [4:43] being an investor in Anthropic? Who's going to pay for Anthropic when they can just get something so [4:47] much cheaper? So, we're investors in Enthropic. We're also investors in fireworks, which serves [4:53] these openweight models. So, look, I'll tell you um and it's not us hedging one way the other. My [5:00] um biggest lesson in this wave is this abundance mindset and one is not happening at the expense of [5:07] the other, right? Like there's room for closed and frontier models. there's room for cheaper [5:14] open source models. So, it's not one or the other. Um, I think Kimmy 3 was a really exciting [5:21] um announcement and really impressive on on a lot of fronts. You know, the 2.8 trillion parameters, [5:26] the context window. Um, and but I I do think it was a little bit overblown in terms of benchmarks. [5:34] It came close to in certain tasks close to fable which is um a watered down version of mythos which [5:43] was trained uh earlier this year. So the frontier is still you know few months ahead and in an AI [5:50] like that those few months are are everything right. So um I think there's just room for for [5:55] all of it. um there's always going to be frontier tasks that are performed by the frontier models [6:00] and there's going to be um you know room for like cheaper models not just from open source but [6:05] also from enthropic. Um, I'll also say with open source like it gets uh it gets loosely termed like [6:15] nobody's open sourcing stuff from the goodness of their hearts, right? Like these things are really [6:19] expensive to train. You need a ton of compute. So people are not just giving it away because they [6:24] love the world. Um it's it's essentially like a commercial uh it's it's a commercial model, [6:30] right? Like you uh give open weights, you have a commercial agreement. Even with Kimmy, [6:35] you have this 20 million in revenue and then you have a commercial agreement with Kimmy. So, it's [6:39] not it's not so black and white. And you think Anthropic's going to go public this fall? Can't [6:46] comment. I hope so. Can't comment. I I know as much as you do. Oh, okay. Well, let me know. Um, [6:52] why are all or why are most of the openweight models coming out of of China and and not here? [6:59] I know there's a few here, but mostly China. I I think the philosophy of it's it's all business [7:07] decisions. So you preach open when you're behind like when you're second place and that's just the [7:15] reality and it makes a lot of business sense. So if if you look at open models here, Nvidia, [7:21] why is you know would Jensen be preaching open source if that was not good for his business? [7:27] Like we know it makes sense for his business because he's commoditizing his compliments and [7:31] you know that leads to more consumption of his chips versus custom chips. You've got um Zuck [7:37] talking about uh you know has been a big champion of open source because he knows he's behind and [7:41] doesn't want to be uh beholden to the Frontier Labs. I think the the China piece is interesting. [7:49] There's probably we can only hypothesize. There's um one is side s side s side s side s side s side [7:54] s side s side s side s side s side sidest stepping export controls of US chips right like so that's [7:59] one way of of sidest stepping that it's probably also like a top- down decision by you know because [8:05] that countries run more top down so um they are ahead they're leaprogging in manufacturing and [8:12] robotics so again commoditizing your compliment so the world um sort of consolidates on their stack [8:18] and they can monetize on on all the others But yeah, you mentioned um Jensen and Mark Zuckerberg, [8:25] also Elon Musk all want, you know, these more open models, but um the companies, you know, [8:31] on leading do not. Um is there is that just for competitive reasons or are there other reasons [8:38] that they don't want that? I I think again we can hypothesize um and there could be two two reasons. [8:47] Why would you why would you open when you're at the frontier again when you're number one you [8:52] there's no incentive for you to open source right like why would you if you're Coca-Cola why would [8:56] you give your IP away so I think that's that's just the reality even and then even the open [9:01] source ones when they reach the frontier they're incentivized to close it so um Quen their recent [9:07] model was closed source till you know there was like a more of a president statement or mpark was [9:13] closed so so I think that's the reality there's also probably another hypothesis is is there real [9:18] demand on the enterprise side for open source and I'm assuming once the Frontier Labs feel that they [9:24] will also open source. Yeah. And and you mentioned the restrictions on on chips. Do you think that [9:28] the export controls that the US has had have had uh have been successful? Um I don't think so. Uh [9:38] because when there is scarcity and creativity that's almost always better than abundance and [9:45] creativity. So, we saw there was Muay Than the Door Dash of China that trained their recent model [9:53] with Huawei chips and um that's pretty jarring. So, now they're able to train state-of-the-art [10:00] model models with um worse silicons and homemade silicons. Yeah, not a lot of scarcity in Silicon [10:06] Valley these days. That's true. I I wish there was more scarcity in Silicon Valley. I'm sure you [10:10] do. Um but you know another thing about that is that um in a way you know startups should be more [10:17] scarce um because you know they always talk about like the twoperson unicorn startup. You don't need [10:23] as many people anymore theoretically. Um does that threaten you know the venture capital model [10:30] your job if startups can only need you know two founders two employees? I would love for there to [10:37] be more twoerson unicorns. I think look I think majority of businesses in the world should not [10:42] be venture funded. You should have a bootstrap company you know if you're going after a smaller [10:48] niche and you want to get to profitability you can do that really fast. Um majority of vertical SAS [10:52] companies I don't think should be venture funded right because by definition they're they're not [10:56] going after like the most ambitious ideas. So I would welcome and love nothing more than twoperson [11:03] unicorns. Do I think the room for venture capital is basically funding R&D before revenue catches [11:09] up? Is that always going to exist for sure? And then hopefully with that ambition, we're just [11:15] only moving humanity forward. Like if you look at, you know, look at the biggest companies right now, [11:20] right? Like the new hyperscalers, you've got Enthropic, OpenAI, SpaceX. Are they raising less [11:25] money or or more money? And also, these guys love talking about how AI is going to take over jobs [11:32] and all this all this fear. They're hiring faster than anyone. Like, why why are they not stopping [11:38] hiring and and being more efficient? So, um but overall, I' I'd love more twoerson unicorns. So, [11:44] so you I don't have to run them. Yeah. So, you but you don't I take it by your answer. You [11:49] don't agree that AI is going to replace jobs? Absolutely not. Why not? I um I think I have a [12:00] lot of I I think AI in the short term might cause some displacement where there's some tasks that [12:06] are mundane tasks that are taken over by AI. If you if you just think about jobs like most jobs [12:12] are fake jobs like my job is a fake job, your job is a fake like most jobs. Yeah, maybe yours is [12:18] more real. Um but you know if you like there's no re the conferences didn't exist in the industrial [12:25] revolution. So I think humanity we constantly just create new jobs and we're going to continue to do [12:31] that and there will be and it's it's even hard for us to imagine what those new jobs look like. [12:36] Maybe there'll be a lot more small businesses there'll be more restaurants. There'll be you [12:40] know other ways of how we entertain ourselves and keep ourselves busy. And um I have I have zero [12:46] fear that in the long term AI will take over jobs. Well, let's let's hope so. Um you know, this year [12:53] I I mentioned you guys raising it was actually 3.5 billion, not three. Uh but then A16Z 15 billion, [13:01] Thrive 10 billion, Sequoia 7 billion. These are some big fund sizes. A lot of driver ours seems [13:07] modest compared to a competitor. Yes, I know. Only three and a half billion. Got to find some [13:11] more whizzes, I guess. But what um what effect do you think these giant funds, yours included, [13:18] um have on on the market? First of all, I wouldn't call us the giant funds. I think [13:24] we fall in this more we we've always had a more of a artisal approach to investing and we kind of [13:30] call ourselves like the scaled artisans. So to the extent that fund sizes let us invest in the best [13:37] companies and at the same time and support them for the long haul and at the same time generate [13:42] um outsiz returns um you know we're very excited about that and we think the size allows us to do [13:50] that. Um what I think is interesting there there there are two schools of thought. But [13:54] one is venture capital most venture firms should not exist because most most venture firms don't [14:02] outperform like median public markets return. So if you just look at S&P QQQ [14:09] um MAG 7 in the last 10 years they have delivered 20% plus IRRs. Most venture firms are not able to [14:17] do that. So why LPS continue to invest them? Now that's that you know there's a lot of different [14:21] reasons because they allocate a certain pool of capital. So there's there's there's one [14:26] that's one school of thought. The other is now on the other hand how do you justify such big [14:32] funds? Um I you know you have to believe that this generation of outcomes is larger than the [14:38] last generation of outcomes. So between SpaceX, OpenAI, Anthropic, all these three companies are [14:45] approaching probably a trillion in revenue in the next 5 years if and if you compare that to Max 7, [14:51] it took them two decades to get combined trillion in revenue. So um do does that mean more capital [14:58] companies are staying private for longer? So I think there's a there's a reason for larger [15:03] funds to exist. As much as I'd love to dunk on my competitors for raising Yes. Um you're welcome to [15:08] do that as well. But and what what valuation did you guys invest in anthropic at 180 billion? Okay. [15:15] Which was unusual for Yeah. Yeah. So now it's you know like 1.3 trillion in secondary markets [15:21] primaries about a billion. Um so but that's that's big I mean and it kind of unusually hard to get a [15:29] venture size return from that big of a valuation. Case in point, like in the past investing, we [15:36] could not even think about that because we obsess over high ownership. We obsess over coming into [15:41] companies early and that's always going to be the ethos of index. But then you have these one ina- [15:48] million scenarios where you're like, okay, this is the next hyperscaler. And even at 180 billion, [15:53] which is again unprecedented for us to enter at that valuation, we're talking about venture [15:58] scale returns fully at at at the growth stage. Um, and I do think that and in that moment it [16:05] it's an interesting riskreward because claude code had just started to take off and that was [16:10] like and you know if you believe in recursive self-improvement and coding being the most [16:14] important thing because then AI trains more AI um it was like a really interesting point in time um [16:20] to have invested and do you think that we are in a bubble right now or and does it even matter? [16:30] We probably are in a bubble in the private markets, early stage markets. Um, does it really [16:34] matter? Yeah, because people are going to lose a lot of money. Do I think it's good sometimes when [16:41] bubbles create an overbuild? Do I think ultimately that's good for society and end consumers? For [16:46] sure. Um, you know, industrial revolution, you have electricity, you had internet. it always [16:54] um there was an overbuild and that led to a lot more innovation in the future and [16:58] most importantly value to um and consumers. So um yeah in some shape or form we are in a bubble but [17:06] um in in other ways like to my earlier point I'm not seeing a demand slowdown. So it's not the most [17:12] obvious bubble the way people would yeah state. Well, I want to end by uh asking since you are [17:19] in the business of predicting the future, I want you to give a hot take. Yeah. Something that would [17:25] uh get I try to predict the future. I'm almost always wrong. Yeah. Well, um yeah. So, give us a [17:30] a hot take uh prediction for something that would surprise us. I don't know if this is a hot take [17:38] or if this would surprise you, but it's something that I've been thinking about and I don't think [17:43] we talk about enough is how um AI is making us really dumb. Um like if you think about just in [17:52] every generation new technology leads to when we outsource something it leads to some atrophy in [17:59] in humanity. So like in industrial revolution um we we our muscles atrophied and that's why [18:05] we created gyms and we kind of have to exercise that with the internet it really led to a decline [18:10] in our attention span and now we're talking about intelligence itself and we're we're outsourcing [18:15] that and that by the way they're like researchers research about it where there was a control group [18:20] where people were outsourcing essay writing to AI versus not and what was like you know and the [18:25] researchers called that cognitive debt where there was like a dep depletion and neural pathways. So, [18:31] um I think we need to think about that more of how do we need gyms for our brains and even just, [18:38] you know, day-to-day sometimes I feel like I have to force myself to not outsource everything just [18:43] because I can to AI. Um what do do you think it's making you dumb? Uh well, yeah. I mean, [18:51] I hopefully not dumb. I but it's definitely like my first reflex to just like ask you [18:56] know GPT everything or with anything is wrong but hopefully it also you know it I did this [19:03] thing also where that I read about in the New York Times where I asked it to you know kind [19:08] of like tell me my biggest weaknesses um and I thought like that was really interesting. Um, [19:14] but you know, there was also the whole Sam Alman thing last week where he tweeted, "Oh, here's a [19:19] great idea. Like, if your kids like something like put on an AI podcast, somebody was like, [19:25] "Why don't you just ask your kids, talk to your kids, talk to your children?" So, um, yes. U, [19:30] but hope, yeah, hopefully it's not making Maybe it doesn't matter. Maybe intelligence, right? Now, [19:35] we've given so much value to intelligence and maybe in the future being intelligent is [19:40] irrelevant. Um and there's there's value across somewhere else. Yes. Well, on that optimistic [19:46] note, we will end. So, thank you so much and thank you all of you for joining us. Thank you.