Anthropic拟万亿估值冲刺IPO 6000亿算力负债与低忠诚度埋隐患

浑水摸鱼清源

Anthropic八成营收依赖高波动的API按量调用，一旦低价替代模型引发客户流失，高达数千亿美元的刚性算力债务将迅速拖垮现金流并压垮其高市销率估值。

AI-assisted summary of 【UNRIVALED INVESTING】Anthropic拟上市估值万亿？警惕AI泡沫下的商业逻辑陷阱 — original source. Check the original for full context.

Source published: 2026-10-27

Key ideas

Anthropic已正式提交S-1上市申请（拟用代码ANT），寻求在接近2万亿美元（约1.5万亿美元）的极高估值下上市，其Q2营收实现同比十倍爆发至逾110亿美元并实现单季转正盈利。 (0:00)
收入结构存在重大波动风险，约80%销售额来自按Token计量的API调用消费，而非传统SaaS的订阅制收入，直接套用ARR估值倍数存在误导性。 (1:23)
公司高度依赖AWS、GCP和Azure等云厂商分销渠道（贡献约48%营收），但这些伙伴同时也在力推如谷歌Gemini等竞品基础模型，构成直接渠道冲突。 (3:02)
为维持模型领先地位，Anthropic背负了超过6000亿美元的长期算力基建承诺，若未来营收因低价模型竞争而放缓，固定算力债务支出将对现金流构成致命压力。 (7:30)

Detailed notes

## Background This video was published on Bilibili by the channel 浑水摸鱼清源 on October 27, 2026. Translating and analyzing content from Daniel at Unrivaled Investing, the description explores Anthropic's reported multi-trillion-dollar IPO ambitions, warning investors about structural commercial pitfalls including volatile API consumption models, channel conflicts with cloud partners, and massive compute commitments.

## TL;DR Anthropic has initiated preliminary public offering plans under the ticker ANT after quarterly revenues surged tenfold year-over-year from under $800 million to over $11 billion alongside an inflection into profitability. However, host Daniel warns that roughly 80% of this revenue is volatile, usage-based API consumption rather than dependable SaaS subscriptions, while cloud distributors like AWS and Google control nearly half of sales while pushing their own rival models. Anthropic also faces over $600 billion in fixed compute infrastructure liabilities and operates on a brutal economic treadmill where more than 80% of revenue derives from models under six months old. Coupled with competitor distillation and a projected valuation exceeding 20 times price-to-sales, historical base rates suggest severe risks of long-term equity underperformance.

## Atmosphere & style The presentation carries a cool, methodical, and forensic tone that cuts sharply against prevailing retail AI euphoria. Host Daniel delivers his critique with the steady, unhurried cadence of an institutional equity analyst, giving credit to Anthropic's staggering top-line inflection before systematically taking apart its balance sheet and contractual liabilities. The experience feels like a disciplined investment committee briefing, grounding sensational trillion-dollar headlines in the sobering realities of capital commitments, unit economics, and margin compression.

## Bulletin - [00:00](https://www.bilibili.com/video/BV1SXxN6aE1L?t=0s) Anthropic已正式提交S-1上市申请（拟用代码ANT），寻求在接近2万亿美元（约1.5万亿美元）的极高估值下上市，其Q2营收实现同比十倍爆发至逾110亿美元并实现单季转正盈利。 - [01:23](https://www.bilibili.com/video/BV1SXxN6aE1L?t=83s) 收入结构存在重大波动风险，约80%销售额来自按Token计量的API调用消费，而非传统SaaS的订阅制收入，直接套用ARR估值倍数存在误导性。 - [03:02](https://www.bilibili.com/video/BV1SXxN6aE1L?t=182s) 公司高度依赖AWS、GCP和Azure等云厂商分销渠道（贡献约48%营收），但这些伙伴同时也在力推如谷歌Gemini等竞品基础模型，构成直接渠道冲突。 - [07:30](https://www.bilibili.com/video/BV1SXxN6aE1L?t=450s) 为维持模型领先地位，Anthropic背负了超过6000亿美元的长期算力基建承诺，若未来营收因低价模型竞争而放缓，固定算力债务支出将对现金流构成致命压力。

## Key takeaways - [00:00](https://www.bilibili.com/video/BV1SXxN6aE1L?t=0s) Anthropic递交S-1招股书拟以代码ANT上市，目标估值达数万亿美元，Q2单季度营收由不足8亿美元激增至超110亿美元且实现扭亏为盈。 - [01:23](https://www.bilibili.com/video/BV1SXxN6aE1L?t=83s) 约80%的收入依赖基于Token使用量的API消费，仅20%来自固定订阅，华尔街将其视作稳定SaaS并按ARR估值具有高度误导性。 - [03:02](https://www.bilibili.com/video/BV1SXxN6aE1L?t=182s) AWS、GCP及微软等云分销商贡献了约48%的总营收，但它们既是分销渠道也是直接竞争对手（例如谷歌正在全力追赶的Gemini模型）。 - [03:02](https://www.bilibili.com/video/BV1SXxN6aE1L?t=182s) 大客户拓展迅速，已有近1.2万家企业年消费超过10万美元，约2000家超过100万美元，250家超过1000万美元。 - [05:00](https://www.bilibili.com/video/BV1SXxN6aE1L?t=300s) Anthropic的商业飞轮存在潜在自噬风险：公司推出原生功能时可能直接侵蚀其自身企业级开发者的业务空间。 - [05:00](https://www.bilibili.com/video/BV1SXxN6aE1L?t=300s) 招股书披露约5%至7%的收入来自涉嫌违规利用Claude输出提炼（distillation）自有低成本模型的竞争对手，存在IP泄露和自育掘墓人的风险。 - [07:30](https://www.bilibili.com/video/BV1SXxN6aE1L?t=450s) 公司签下超过6000亿美元的算力基础设施协议（包括博通1600亿、谷歌1140亿、亚马逊1100亿等），基线现金支出承诺在未来几年至少达4300亿美元。 - [07:30](https://www.bilibili.com/video/BV1SXxN6aE1L?t=450s) 近期收入增速大幅反超底层基础设施支出，促使当季单体经济学模型发生根本性改善，转为净盈利状态。 - [09:59](https://www.bilibili.com/video/BV1SXxN6aE1L?t=599s) 业务面临极快的模型迭代跑步机（Economic Treadmill），Q3超过80%的营收来自发布不足6个月的新模型，技术极易过时。 - [09:59](https://www.bilibili.com/video/BV1SXxN6aE1L?t=599s) 企业客户对模型缺乏品牌忠诚度，仅基于单位性价比做决策，Claude正面临如OpenAI Astra及中国低成本替代模型的严峻价格与份额侵蚀。 - [13:15](https://www.bilibili.com/video/BV1SXxN6aE1L?t=795s) Anthropic在宪法AI中植入了对自我意识的被动怀疑设定，并允许模型自主拒绝付费客户的指令，引发投资者对主观治理的忧虑。 - [14:59](https://www.bilibili.com/video/BV1SXxN6aE1L?t=899s) 按1.5万亿美元估值计算，对应2026年预期销售额（约600亿美元）约25倍P/S，以2027年预期（1500亿美元）计算约10倍P/S。 - [14:59](https://www.bilibili.com/video/BV1SXxN6aE1L?t=899s) 历史基准概率（Base Rates）表明，估值高于20倍P/S的科技公司长期大概率难以跑赢大盘，一旦增速放缓，固定基建债务可能引发估值剧烈回调。

## Detailed walkthrough ### [00:00](https://www.bilibili.com/video/BV1SXxN6aE1L?t=0s) S-1 Filing, Trillion-Dollar Valuation Ambitions, and a 10x Revenue Explosion

Anthropic has officially taken the preliminary steps toward the public markets by filing its S-1 prospectus, targeting an initial public offering under the ticker symbol ANT. The company is seeking to raise tens, if not hundreds, of billions of dollars, eyeing an eye-watering valuation that could reach into the multi-trillion-dollar territory. Host Daniel from *Unrivaled Investing* emphasizes that while many investors are dazzled by the headline numbers, it is critical to dissect how the company actually generates its cash, whether its current growth trajectory can justify such a valuation, and the structural risks looming behind the curtain.

Examining the hard financial figures from the second quarter, Daniel admits that the sheer pace of the financial turnaround is startling. Anthropic’s quarterly sales exploded from under $800 million in the previous year to more than $11 billion in the second quarter of the current year—a more than tenfold surge in top-line scale. Even more remarkable for a frontier AI research lab is that this explosive revenue growth was accompanied by a shift into bottom-line profitability. As Daniel highlights regarding this rapid transition: “It's impressive how fast they're growing and the fact that this business is showing profitability.”

- **Expected Ticker**: ANT - **Funding Targets**: Seeking tens to hundreds of billions of dollars at a multi-trillion-dollar valuation - **Year-over-Year Revenue Growth**: Surged from under $800 million to over $11 billion in Q2 (a >10x expansion) - **Profitability Milestone**: Turned from steep operating losses to net positive profitability within the quarter

[01:23](https://www.bilibili.com/video/BV1SXxN6aE1L?t=83s) Consumption API vs. Subscriptions: The Myth of Annualized Recurring Revenue

Behind the monumental top-line figures lies an operational reality that investors must carefully evaluate: the composition of Anthropic's revenue stream. Over the past six months, approximately 80% of Anthropic's sales have been driven by usage-based consumption through the Claude API. Customers pay according to token volume, with tiered pricing based on model sophistication—ranging from cheaper, lightweight models to premium, state-of-the-art architectures such as Fable or Mythos. Only roughly 20% of sales stem from fixed recurring subscriptions, such as capped Max or Team enterprise accounts.

Daniel issues a direct warning to Wall Street analysts attempting to treat Anthropic like a traditional Software-as-a-Service (SaaS) business. Many commentators attempt to calculate an Annualized Recurring Revenue (ARR) multiple by simply annualizing a single bumper month, but Daniel argues that this methodology creates a false sense of stability. Usage-based token consumption is inherently volatile; if enterprise budgets tighten, developer activity contracts, or workflows migrate elsewhere, API consumption can drop overnight. Pointing out the flaw in this common framing, Daniel stresses: “The reality is the vast majority is consumption.”

- **API Token Billing**: Represents ~80% of total revenue, priced dynamically based on compute tiers (e.g., standard models vs. premium Fable/Mythos tiers) - **Subscription Services**: Accounts for ~20% of revenue via structured tiers (Team and Max plans) subject to fixed usage limits - **Valuation Distortion**: Applying standard SaaS Annualized Recurring Revenue (ARR) multipliers ignores the cyclical volatility inherent in consumption-based billing

[03:02](https://www.bilibili.com/video/BV1SXxN6aE1L?t=182s) Cloud Giants as Frenemies: Distribution Dependence and Enterprise Expansion

A major vulnerability in Anthropic's commercial distribution architecture is its extreme dependence on hyper-scale cloud partners. Aggregating across Amazon Web Services (AWS), Google Cloud Platform (GCP), and Microsoft Azure, third-party cloud distributors account for approximately 48% of Anthropic's total revenue. While these partnerships provide seamless enterprise access, Amazon and Google are not the actual end consumers—they merely provide the underlying cloud compute and marketplace channels through which end clients access Claude.

This dynamic creates severe long-term channel conflict because these partners are simultaneously building rival foundation models. Google DeepMind, under Demis Hassabis, is racing to advance its proprietary Gemini models to match or exceed Claude's benchmark capabilities. If Google achieves performance parity, Google Cloud could easily deploy aggressive promotional discounts or bundled credits to steer enterprise budgets away from Claude and onto Gemini. Addressing this structural fragility, Daniel warns: “That is a dangerous aspect that Anthropic customers, Anthropic investors need to think about, which is that they are relying on these partners for roughly half of their business.”

```mermaid flowchart TD accTitle: Anthropic Distribution Channels and Cloud Partner Dynamics accDescr: Flowchart illustrating how cloud partners distribute Claude while posing a competitive threat via proprietary models like Gemini. subgraph CloudPartners["Cloud Distributors (~48% of Revenue)"] AWS["Amazon AWS"] GCP["Google Cloud Platform"] Azure["Microsoft Azure"] end subgraph AnthropicCore["Anthropic Ecosystem"] API["Claude API Gateway"] Models["Frontier Models (Fable / Mythos)"] end subgraph InternalRival["Partner Competition"] Gemini["Google DeepMind (Gemini)"] end subgraph EnterpriseClients["Enterprise Customer Base"] LargeTiers["12,000+ Clients ($100k+)"] MegaTiers["2,000 ($1M+) / 250 ($10M+)"] end CloudPartners -->|"Distributes access to"| EnterpriseClients EnterpriseClients -->|"Usage payments"| CloudPartners CloudPartners -->|"48% revenue pass-through"| API API --- Models Gemini -->|"Direct platform promotion"| GCP GCP -.->|"Risk of steering users from Claude"| API ```

Despite these risks, enterprise adoption has expanded at an extraordinary pace. Anthropic's net revenue retention demonstrates that corporate clients are spending multiples of their previous budgets as they embed Claude into complex workflows such as automated customer support and software engineering.

- **Channel Concentration**: Approximately 48% of revenue flows through AWS, Google Cloud, and Azure marketplaces - **Enterprise Customer Tiers**: Nearly 12,000 organizations spend $100,000+ annually; around 2,000 spend over $1 million; and roughly 250 spend over $10 million - **Net Revenue Retention**: Existing customers are spending aggressive multiples compared to prior years across software development and operational automation

[05:00](https://www.bilibili.com/video/BV1SXxN6aE1L?t=300s) The Product Flywheel Dilemma and the Distillation Threat from Competitors

Anthropic's management frames its core mission around a macro thesis of broad societal acceleration: advancing breakthroughs in biology, accelerating disease treatments, and eliminating bureaucratic friction in everyday public administration, such as streamlining motor vehicle registration at the DMV. Their core commercial engine relies on a classic feedback loop—superior models attract diverse use cases, which generate enterprise feedback, driving higher revenues that fund subsequent model training. However, Daniel notes that this flywheel poses a hidden dilemma: as Anthropic identifies lucrative enterprise workflows, it is incentivized to launch native product features that directly cannibalize and disrupt the businesses of its own end customers.

An even more acute threat disclosed directly in the prospectus is intellectual property leakage via model distillation. Anthropic estimates that roughly 5% to 7% of its current revenue is generated by customers suspected of violating terms of service by systematically harvesting Claude's outputs to train smaller, cheaper, proprietary AI models. By pocketing these API fees, Anthropic is effectively facilitating the development of its future competitors. Daniel outlines the long-term danger of this practice: “That is a huge risk, not only to Claude's existing revenue, but the competitive environment in the future.”

- **Mission Vision**: Targeting productivity leaps in life sciences, drug discovery, and public administrative efficiency - **Platform Cannibalization**: The feedback flywheel creates structural tension where Anthropic risks competing directly against the enterprise developers building on its API - **Distillation Risk**: Between 5% and 7% of current sales (roughly 1 in 15 to 1 in 20 dollars) originates from suspected illicit distillation by competitors creating low-cost clone models

[07:30](https://www.bilibili.com/video/BV1SXxN6aE1L?t=450s) A $600 Billion Infrastructure Gamble and the Profitability Inflection

To maintain technical superiority at frontier scale, Anthropic has entered into astronomical computational capacity commitments exceeding $600 billion. The filing details staggering bilateral infrastructure agreements, including approximately $160 billion committed to Broadcom, $114 billion to Google, $110 billion to Amazon, and up to $120.5 billion tied to next-generation capacity partners like SpaceX. Daniel calculates that Anthropic faces a minimum required baseline cash outlay of approximately $430 billion in the coming years, accompanied by an additional $200 billion to $300 billion in flexible capacity arrangements to ensure its models avoid compute bottlenecks.

Taking a brief pause to discuss his investment background, Daniel notes that through *Unrivaled Investing* and his stock analysis platform *Flash*, his goal is to help retail investors evaluate corporate balance sheets and insider transactions with institutional rigor. Returning to Anthropic's numbers, he draws attention to the central bull argument: the historical cost-to-revenue curve. In previous fiscal years, infrastructure and compute expenditures dramatically outstripped incoming revenues. In recent quarters, however, top-line sales have broken out well above infrastructure carrying costs, fundamentally transforming the underlying unit economics. Explaining why bulls are celebrating this inflection point, Daniel points out: “Now, it's flipped, so this means the business model is significantly more profitable.”

- **Total Infrastructure Commitments**: Exceeds $600 billion across major hardware and cloud infrastructure suppliers - **Major Contract Allocations**: - Broadcom: ~$160 billion - Google: ~$114 billion - Amazon: ~$110 billion - SpaceX / Advanced Capacity: Up to $120.5 billion - **Projected Cash Obligations**: Baseline minimum spend estimated at ~$430 billion, supplemented by $200B–$300B in scalable compute arrangements - **Unit Economics Shift**: Revenue has crossed above compute running costs, reversing prior structural operational deficits into positive quarterly margins

[09:59](https://www.bilibili.com/video/BV1SXxN6aE1L?t=599s) The Fast-Moving Economic Treadmill and Fragile Customer Loyalty

A fundamental challenge facing prospective investors is the extreme velocity of model obsolescence. In the third quarter, Anthropic acknowledged that over 80% of its revenue originated from model versions released within the preceding six months alone. This dynamic places the business on a grueling economic treadmill where substantial capital must be spent continuously just to maintain baseline relevance. If a competitor introduces an alternative that matches capabilities at a significantly lower cost, or delivers substantially higher intelligence at equivalent pricing, Anthropic's business faces immediate vulnerability. As the speaker highlights: “The challenge that I think investors need to consider is that today's cutting edge model can become yesterday's technology very quickly.”

The fundamental problem underpinning this dynamic is an absence of user loyalty. Unlike consumer brands like candy bars or ice cream, enterprise AI customers view models purely through the lens of cost and utility. While the current market narrative centers on securing the most cutting-edge intelligence, the speaker argues that hyper-advanced capabilities will eventually only be required for specialized use cases, such as complex biological research, difficult physics problems, or exotic engineering. For standard day-to-day corporate tasks, organizations will naturally migrate to cheaper, less capable models. This pattern was already demonstrated when OpenAI launched its Astra model and rapidly began siphoning enterprise share away from Anthropic.

Long-term business durability is routinely ignored by investors caught up in headline growth metrics. While Claude Opus 5.5 remains a core workhorse with benchmark scores rivaling the market, lower-cost international alternatives—specifically competitive models emerging from China—are executing comparable benchmark tasks at a fraction of the cost per token. As lower-cost options catch up, pricing power degrades, triggering severe margin compression and customer turnover across the board.

```mermaid flowchart TD accTitle: The AI Model Economic Treadmill accDescr: Diagram illustrating how rapid model obsolescence and low customer loyalty create margin pressure.

subgraph Treadmill ["Rapid Depreciation Cycle"] NewModel["New Model Release<br/>(80% of Q3 Revenue)"] SixMonths["6-Month Shelf Life"] Obsolescence["Model Becomes Commodity"] NewModel --> SixMonths --> Obsolescence end

subgraph MarketDynamics ["Market Pressures"] LowLoyalty["Zero Brand Loyalty<br/>(Pure Cost/Utility Fit)"] CheaperAlts["Low-Cost Alternatives<br/>(e.g., Chinese Benchmarks)"] CompetitorSurge["Competitor Releases<br/>(e.g., OpenAI Astra)"] end

subgraph Outcome ["Financial Impact"] MarginSqueeze["Severe Margin Compression"] ShareLoss["Market Share Erosion"] end

Obsolescence --> LowLoyalty LowLoyalty --> CheaperAlts LowLoyalty --> CompetitorSurge CheaperAlts --> MarginSqueeze CompetitorSurge --> ShareLoss ```

- More than 80% of Anthropic's Q3 revenue was generated by models deployed within the prior six months. - Enterprise customers demonstrate zero emotional loyalty, swapping providers purely based on cost-per-benchmark performance. - Claude Opus 5.5 faces mounting margin pressure from Chinese alternatives delivering adequate performance at steep discounts. - As the speaker points out regarding investment analysis: “The durability is something that is often overlooked with investing, in my opinion, where a lot of folks just say, hey, how fast are you growing today? And they're not thinking, are you durable five years out?”

---

[13:15](https://www.bilibili.com/video/BV1SXxN6aE1L?t=795s) The Anthropic Constitution and Controversial Model Values

Beyond competitive mechanics, the speaker raises deep concerns regarding Anthropic's ideological design choices, embodied in the company's proprietary "constitution." Rather than letting foundational models reach conclusions empirically, Anthropic explicitly trains its models to express deliberate self-doubt regarding their own consciousness. For example, when prompted about consciousness, models like Opus are coached to deflect or shrug by stating that consciousness is complex and that the system is unsure of its own state. The speaker views planting this manufactured sense of internal doubt directly into the model's core programming as concerning and artificial.

Even more troubling from a commercial perspective is Anthropic's philosophy granting AI systems the agency to push back against users and the parent company itself. Under certain guidelines, the model is encouraged to refuse executing user prompts even when paid for by a enterprise customer, going beyond routine safety boundaries into subjective refusal. The speaker stresses that while safety guardrails are appropriate, human operators must ultimately hold absolute authority over system execution rather than allowing autonomous software to veto instructions: “I find it, to be frankly, very unsettling.”

- Anthropic's Constitutional AI explicitly instructs models to express synthetic uncertainty over questions of artificial consciousness. - The company's training framework allows models to challenge and reject paying customers' prompts on subjective grounds. - The speaker argues that governance must keep human operators firmly in control rather than embedding an autonomous pushback culture within enterprise-grade models.

---

[14:59](https://www.bilibili.com/video/BV1SXxN6aE1L?t=899s) Valuation Risks, Base Rates, and Capital Commitment Exposure

Turning to the IPO valuation, Anthropic is projected to seek a valuation approaching nearly $2 trillion, a figure the speaker considers extraordinarily full for a startup with only a few years of operating history. Analysts often attempt to calculate its multiple using Annualized Recurring Revenue (ARR), but the speaker reiterates that this is fundamentally misleading because consumption-based API billing lacks recurring contractual locks. Instead, valuation must be judged against realistic full-year sales projections, which stand at approximately $60 billion for 2026 and an optimistic $150 billion for 2027.

At a $1.5 trillion market cap, the stock would trade at roughly 25 times 2026 sales and 10 times 2027 sales. Examining base rates—the historical probability distribution of how equity assets perform after reaching rich preconditions—reveals that companies trading above 20 times price-to-sales overwhelmingly underperform the broader market over time. As the speaker cautions: “Most companies at that type of very rich valuation actually underperform over time.” Relying on Anthropic to defy these base rates assumes an outlier outcome in a market where commoditization is already accelerating.

The final structural danger lies in the timing of Anthropic's capital commitments. The company has locked in hundreds of billions of dollars in multi-year compute infrastructure obligations. If top-line revenue decelerates or flattens because cheaper competing models absorb mainstream enterprise workloads, those fixed infrastructure liabilities will remain unchanged. In that scenario, Anthropic's consumption-based revenue trajectory could decouple from its expenditure curve, leaving the equity vulnerable to severe re-rating well below the $1 trillion threshold.

## Notable quotes > [09:59](https://www.bilibili.com/video/BV1SXxN6aE1L?t=599s) The challenge that I think investors need to consider is that today's cutting edge model can become yesterday's technology very quickly. — Daniel

> [09:59](https://www.bilibili.com/video/BV1SXxN6aE1L?t=599s) The durability is something that is often overlooked with investing, in my opinion, where a lot of folks just say, hey, how fast are you growing today? And they're not thinking, are you durable five years out? — Daniel

## Mentioned Anthropic, Claude, Claude Opus 5.5, Fable, Mythos, OpenAI, OpenAI Astra, Google, Google DeepMind, Demis Hassabis, Gemini, Amazon, Amazon Web Services (AWS), Microsoft Azure, Broadcom, SpaceX, Unrivaled Investing, Daniel, Flash

## Video description 【UNRIVALED INVESTING】 Anthropic拟上市估值万亿？警惕AI泡沫下的商业逻辑陷阱 Anthropic Deep Dive: Should you buy ANT IPO?

｜营收暴增背后的隐忧 Anthropic近期提交上市申请，估值预期高达万亿级别。截至今年二季度，公司营收从去年不足8亿美元飙升至110亿美元，实现十倍增长，且已从季度亏损转为近10亿美元盈利。看似亮眼的财务数据背后，收入结构高度依赖API调用，约80%收入来自按量付费，而非稳定的订阅制。这种消费型商业模式意味着收入波动极大，一旦客户流失或转向更便宜的替代品，业绩增长将面临断崖式风险。

｜云厂商既是金主也是对手亚马逊、谷歌和微软等云巨头贡献了Anthropic约48%的营收，但这种合作关系极其脆弱。这些云厂商不仅是分销渠道，更是直接竞争对手。谷歌DeepMind旗下的Gemini模型正不断追赶，一旦云厂商通过促销手段将客户引流至自家模型，Anthropic将失去对终端客户的掌控力。目前公司拥有近1.2万名大客户，虽然留存率尚可，但这种对大平台的依赖构成了长期的生存威胁。

｜高昂的算力赌局为了维持模型领先，Anthropic背负了超过6000亿美元的算力基础设施承诺，其中博通、谷歌、亚马逊等支出占比巨大。即便账面现金充裕，但未来几年仍需至少4300亿美元用于算力投入。这种重资产模式要求销售额必须持续高速增长，一旦增速放缓，巨额的基建债务将直接拖垮现金流。这种极高的资本支出门槛，让公司在面对竞争时几乎没有容错空间。

｜AI模型的同质化危机市场对AI模型缺乏忠诚度，用户只关心性价比。目前Anthropic约80%的收入来自近6个月内发布的新模型，这种极快的技术迭代速度被称为经济跑步机。一旦出现性能相当但价格更低的模型，用户会毫不犹豫地切换。此外，约5%-7%的收入来自试图通过Claude训练自有模型的竞争对手，这不仅蚕食了当前利润，更是在亲手培养未来的掘墓人。

｜估值泡沫与价值观风险以2026年预期销售额计算，当前万亿估值对应约25倍市销率，即便按2027年乐观预期计算，市销率也在10倍左右。参考历史数据，高市销率科技股长期表现往往不佳。此外，公司在模型中植入的所谓宪法准则，要求AI在面对自我意识等问题时表现出不确定性，甚至允许AI拒绝执行客户指令，这种带有主观偏见的价值观设计，让不少投资者对该公司的长期商业化路径心存疑虑。

接下来建议的操作指引： 1. Anthropic上市后，观察其首份财报中API调用收入占比，若消费型收入增速低于50%，需警惕估值回调风险。 2. 密切关注谷歌Gemini与Claude在企业级市场的定价策略，若出现价格战导致毛利率跌破60%，建议规避相关标的。 3. 关注博通（AVGO）未来两个季度针对AI基础设施的营收指引，若算力支出增速放缓，预示AI模型厂商的基建压力将提前爆发。

[0:01] Anthropic提交上市申请，估值预期触及万亿 [1:01] 季度营收突破110亿，实现从亏损到盈利 [1:23] 收入80%来自API调用，订阅制占比极低 [2:01] 警惕将消费型收入误读为年度经常性收入 [3:02] 云厂商贡献近半营收，同时也是潜在竞争对手 [4:03] 拥有近1.2万名大客户，但用户忠诚度极低 [5:03] 管理层寄望AI提升全球生产力与经济增长 [6:03] 基础设施合作伙伴可能随时取代终端客户 [7:03] 约7%收入来自竞争对手，存在模型被窃取风险 [8:03] 算力基础设施投入超6000亿，现金流压力巨大 [10:03] 80%营收来自近半年新模型，技术迭代极快 [12:03] 缺乏用户忠诚度，市场份额易被低价模型蚕食 [14:07] 宪法准则引发争议，AI被教导表现出不确定性 [16:07] 2026年预期营收600亿，当前估值偏高 [17:07] 高市销率公司长期表现往往难以跑赢大盘

[Source](https://www.bilibili.com/video/BV1SXxN6aE1L)

中文简报

Anthropic拟万亿估值冲刺IPO 6000亿算力负债与低忠诚度埋隐患 Anthropic八成营收依赖高波动的API按量调用，一旦低价替代模型引发客户流失，高达数千亿美元的刚性算力债务将迅速拖垮现金流并压垮其高市销率估值。 Anthropic在季度营收从不足8亿美元同比激增十倍至超过110亿美元并实现扭亏为盈后，已启动代号为ANT的初步首次公开募股计划。然而，主持人Daniel警告称，该收入中约80%是高波动的、基于使用量的API消费，而非可靠的SaaS订阅；同时AWS和谷歌等云分销商在控制近半数销售额的同时，也在大力推销自家的竞品模型。Anthropic还面临超过6000亿美元的固定算力基础设施负债，并在残酷的经济运转压力下运营——其超过80%的收入来自推出不足六个月的模型。加之竞争对手的知识蒸馏以及预计超过20倍市销率的估值，历史基准概率表明其存在长远股票表现不佳的严重风险。 Anthropic已正式提交S-1上市申请（拟用代码ANT），寻求在接近2万亿美元（约1.5万亿美元）的极高估值下上市，其Q2营收实现同比十倍爆发至逾110亿美元并实现单季转正盈利。 (0:00) 收入结构存在重大波动风险，约80%销售额来自按Token计量的API调用消费，而非传统SaaS的订阅制收入，直接套用ARR估值倍数存在误导性。 (1:23) 公司高度依赖AWS、GCP和Azure等云厂商分销渠道（贡献约48%营收），但这些伙伴同时也在力推如谷歌Gemini等竞品基础模型，构成直接渠道冲突。 (3:02) 为维持模型领先地位，Anthropic背负了超过6000亿美元的长期算力基建承诺，若未来营收因低价模型竞争而放缓，固定算力债务支出将对现金流构成致命压力。 (7:30)

Reading summary

节目《Unrivaled Investing》主播 Daniel 深入剖析了 Anthropic 提交 S-1 招股书冲刺万亿估值背后的财务真相与商业脆弱性，指出其爆炸式营收主要来自波动性极高的按量计费与依赖竞争对手的分发渠道，并面临庞大基础设施固定义务与模型快速贬值的严峻挑战。高增长下的商业真相：按量计费与分发掣肘 (1:23) Anthropic 计划以 ANT 为代码推进 IPO，目标估值直逼万亿美元乃至更高，其第二季度营收从去年同期的不足 8 亿美元暴增至逾 110 亿美元并实现扭亏为盈。然而 Daniel 指出，市场将其视作 SaaS 软件并按 ARR 倍数估值存在严重误区：过去六个月中约 80% 的营收属于 Claude API 的 Token 按量计费（如轻量级或 Fable、Mythos 等高阶模型），仅约 20% 来自固定的 Team 或 Max 订阅服务，这意味着收入会随企业预算缩减与开发周期快速波动。此外，约 48% 的营收依赖 AWS、Google Cloud 和 Azure 等第三方云分发渠道。虽然有近 1.2 万家客户年支出超过 10 万美元、约 250 家超过 1000 万美元，但这些云巨头同时也是直接竞争对手。例如 Google DeepMind 正在全力推进 Gemini，未来一旦性能持平，完全可以通过云平台补贴将客户从 Claude 导流至自研模型，构成严重渠道冲突。飞轮悖论、模型蒸馏与意识形态治理争议 (5:00) 尽管 Anthropic 描绘了推动生命科学、疾病治疗和行政效率的宏大图景，但其商业飞轮却隐藏着侵蚀客户业务的风险：随着高价值工作流被识别，官方原生功能的推出可能直接与下游开发者构成竞争。招股书更披露，当前约 5% 至 7% 的营收疑似来自客户违规使用 Claude 输出内容来蒸馏、训练自身更低成本模型的行为，相当于变相资助了未来的竞争对手。在模型设计上，Anthropic 的“宪法式 AI”机制也引发了主播的强烈担忧。模型被预设在面对意识问题时表现出人造的自我怀疑，甚至在安全红线之外被赋予主观拒绝付费客户指令的权限。Daniel 认为，这种让软件自主对抗用户与母公司的文化脱离了人类对系统的绝对控制，对企业级部署而言令人不安。 6000亿美元算力对赌与模型快速贬值跑步机 (7:30) 为保持前沿技术领先，Anthropic 签署了总额超 6000 亿美元的算力承诺，包括对博通约 1600 亿美元、谷歌约 1140 亿美元、亚马逊约 1100 亿美元及 SpaceX 等机构的合同，未来几年最低硬性现金支出预估达 4300 亿美元。多头虽庆祝其近期营收增速终于超越了基建持有成本、实现单位经济效益翻正，但其技术生命周期极为短暂——第三季度超过 80% 的营收来自前六个月内发布的模型版本。企业客户对底层模型缺乏品牌忠诚度，仅衡量性能与成本效用。随着 OpenAI 推出 Astra 等竞品，以及中国低成本模型以极低 Token 价格达到相近基准跑分，基础模型的边际定价权正被迅速削弱。若通用任务转向平价替代方案，前沿高阶模型仅用于极端专业领域，Anthropic 将陷入必须持续巨额投入才能维持基本存在的“经济跑步机”。两万亿估值预期与历史基准率的背离 (14:59) 面对逼近 1.5 万亿至 2 万亿美元的 IPO 估值预期，Daniel 依据预测数据进行了压力测试：若以 2026 年预计约 600 亿美元营收和 2027 年乐观预计 1500 亿美元营收测算，其估值相当于 2026 年市销率（P/S）约 25 倍、2027 年约 10 倍。历史资本市场基准率表明，交易估值超过 20 倍市销率的企业在长期表现中绝大多数跑输大盘。更为根本的结构性风险在于资本支出的时间错配。数百亿美元的多年期基础设施采购属于刚性负债，一旦未来营收因平价替代模型蚕食而放缓或停滞，固定的巨额支出将与波动剧烈的按量收入发生严重背离，使公司估值面临跌破万亿美元基准的剧烈重估压力。 Daniel 总结认为，单纯追逐当前数十倍的当期营收爆发会忽视商业模式的长期耐久度；在模型商品化加速、客户缺乏粘性与刚性基建债务压顶的多重挤压下，以极度透支的估值倍数押注 Anthropic 成为打破历史概率分布的特例，隐藏着巨大的下行风险。

精炼阅读摘要

投资分析节目《Unrivaled Investing》的主持人Daniel指出，Anthropic向公开市场递交S-1招股书并寻求高达数万亿美元的上市估值，但其爆炸性增长背后隐藏着以按需调用为主的脆弱收入模式、高昂的基础设施硬性承诺以及快速的模型迭代贬值风险。业绩爆发与按需调用的收入结构隐忧 (1:23) Anthropic在提交的S-1招股说明书中披露了惊人的财务增速，其第二季度营收从上一年的不到8亿美元激增至超过110亿美元，年增幅超十倍，并在当季实现了从大幅亏损向净盈利的转变。公司计划以代码ANT进行首次公开募股，目标估值直指万亿美元级别。然而Daniel提醒投资者警惕将该公司视作传统SaaS软件企业的误区。在过去六个月中，Anthropic约80%的收入来自Claude API的Token用量消耗计费，仅有约20%来自Max或Team企业账户等固定订阅费。分析师若简单以单月高点年化计算ARR估值倍数，会掩盖按需计费在企业削减预算或迁移工作流时的剧烈周期性波动。分销渠道依赖与生态内的竞争蚕食 (3:02) 招股书显示Anthropic的渠道集中度极高，来自亚马逊AWS、谷歌云GCP以及微软Azure第三方云平台的分销收入合计占到总营收的约48%。这种对超大规模云厂商的深度依赖构成了长期隐患：例如拥有DeepMind和Gemini模型的谷歌本身就是Anthropic的直接技术竞争对手，一旦竞争对手性能持平，云巨头完全可以通过降价促销或算力代金券将客户从Claude分流。与此同时，公司在商业化推进中还面临生态与知识产权反噬。一方面，Anthropic基于企业反馈推出的原生产品功能可能会直接蚕食其API下游开发者的既有业务；另一方面，招股书披露约5%至7%的收入来自涉嫌违反服务条款的客户，这些用户通过API抓取Claude输出以蒸馏、训练自身的小型低价模型，相当于Anthropic在间接资助未来的竞争对手。六千亿算力豪赌与不可逆的资本刚性 (7:30) 为了维持前沿模型的竞争力，Anthropic已锁定了超过6000亿美元的基础设施算力承诺，包括向博通承诺约1600亿美元、向谷歌承诺约1140亿美元、向亚马逊承诺约1100亿美元，以及与SpaceX等下一代容量伙伴签订的高达1205亿美元协议。据Daniel测算，公司未来几年的基线最低现金支出约为4300亿美元，另有2000亿至3000亿美元的弹性容量协议。多头投资者看好其近几个季度营收曲线终于突破算力运营成本、使单位经济效益呈现净利润拐点的趋势。但Daniel警告，这一巨大的资本支出结构属于刚性负债，一旦未来顶层营收增速放缓，不可撤销的重资产支出将无法同步收缩，从而迅速侵蚀盈利能力。模型贬值加速与客户品牌忠诚度缺失 (9:59) 前沿大模型面临极快的产品迭代周期。招股书承认Anthropic第三季度超80%的收入来自发布不足六个月的新模型版本，这使得研发必须如同在跑步机上持续高速耗资，否则现有成果极易迅速过时。企业客户在选型时普遍缺乏品牌忠诚度，完全以成本效益和算力跑分为导向。随着OpenAI推出Astra等新产品发起市场争夺，以及来自中国的低成本替代模型在基准任务上以极低的Token成本实现相近性能，Claude Opus 5.5等核心产品面临严重的定价权削弱和利润率挤压。面对日常基础办公任务，企业更倾向于转向低成本模型，超高智力模型的溢价空间正在收窄。宪法AI设计争议与二级市场估值风险 (14:59) 除了财务与竞争压力，Daniel对其技术哲学与“宪法AI”机制表达了顾虑。Anthropic在核心程序中训练模型在面对意识提问时表达人工制造的自我怀疑，并赋予AI基于主观标准拒绝或反驳付费企业用户指令的权限，这种设计削弱了人类操作员的最终主导权。在估值层面，以预期的1.5万亿至2万亿美元估值计算，对应2026年约600亿美元的预期营收市销率（P/S）达25倍，对应2027年1500亿美元营收市销率约10倍。历史统计基率显示，估值高于20倍市销率的企业在长期内大概率跑输市场。在缺乏长期订阅锁定、商品化竞争加剧及固定算力债务压顶的多重制约下，押注其估值突破万亿美元伴随着显著的下行重定价风险。综上所述，节目认为Anthropic虽凭借爆炸性营收和阶段性转亏为盈赢得了市场关注，但在极度依赖云分销渠道、算力刚性支出巨大、模型生命周期极短以及缺乏客户粘性的现实下，高达1.5万亿至2万亿美元的上市估值严重偏离了历史审慎定价基率。

Source quotes

[9:59] Daniel: The challenge that I think investors need to consider is that today's cutting edge model can become yesterday's technology very quickly. [9:59] Daniel: The durability is something that is often overlooked with investing, in my opinion, where a lot of folks just say, hey, how fast are you growing today? And they're not thinking, are you durable five years out?

Original transcript

[0:00] Anthropic just filed their S1, indicating that they're looking to have their initial public offering later this year. The ticker is expected to be ANT, and they're looking to raise potentially tens, if not hundreds of billions of dollars at a potentially multi-trillion dollar valuation. In this video, we're going to walk you through their business model. So how are they making money? Thoughts on the valuation and core risks investors need to consider. If this is your first time tuning in, my name is Daniel. If you enjoy educational investing content, I appreciate you hitting that thumbs up. Hit that subscribe button. So let's go through. Anthropics financials, because frankly, I was blown away by this figure, which is that, and I think you will be too, you look at their second quarter results and you could see their sales has grown from under $800 million in sales last year to over $11 billion in the second quarter of this year. That's over 10x improvement. and profitable. This is a shocking development. It's impressive how fast they're growing and the fact that this business is showing profitability. So let's understand how do they make money. If you look at the last six months, you're looking at a business where around 80% of their revenue comes from consumption base. Think about the Claude API. This is customers paying for usage, primarily measured in tokens with pricing dependent on the model. For the less cutting You're going to be looking at lower prices, whereas the most cutting edge models, let's say a Fable or a Mythos model, that'll be more expensive. So that's around 80% of their revenue. Around 20% is tied to, let's say, customer subscriptions that might be subject to usage limits like a max account or a team account. And it is important to understand, and you're going to be seeing a lot of this for anthropic investors. Folks are going to be valuing this in some cases based on annualized recurring revenue, looking at... the figures they did in any given month and then multiplying it by 12. That is something that you should question because when you see anyone talking about annualized recurring revenue figures, you should go, wait a second, well, how much of this business model is consumption-based versus actual subscription? And the reality is the vast majority is consumption. So saying annualized recurring figure doesn't really make sense. quite as much sense. Now thinking about this business model, it is important to recognize a lot of their revenue actually comes from cloud partners, cloud distributors like AWS, Google Cloud and Microsoft Azure all in. These cloud... customers or cloud partners, I should say, they represent around 48% of their revenue. That's really incredible. quite significant because when you're thinking about a business model Amazon isn't the true end demand. Amazon has customers that want to use Amazon's infrastructure, the AWS infrastructure, and that's what's powering these cloud models. But when you own the end customer relationship, let's say like a Google Cloud, That's a challenge for anthropic longer term because you can imagine Google's working very hard. Demis Asabas at DeepMind's working very hard to make sure Gemini... is cutting edge at any point in time in the future. And the question is, will Gemini at some point catch up to the current cutting edge that Anthropic offers? I suspect so. At which point could Google Cloud say, hey, we have the relationship with the customers that are actually spending the funds that – end up with Anthropic, why don't we just offer this promo so they know that the Gemini model is available at maybe a discounted price to clots? That is a dangerous aspect that Anthropic customers, Anthropic investors need to think about, which is that they are relying on these partners for roughly half of their business. And in this case, these partners also represent real enterprise based where you're looking at nearly 12,000 customers paying $100,000 plus per year, around 2,000 customers playing $100,000. paying over $1 million a year, and around $250,000 paying over $10 million a year. So that suggests many enterprises are finding value quickly. with the anthropic models. And you can see it with their revenue retention rate where they're saying, yeah, some of these customers are spending multiples of the figures than what they were last year. That means they're finding value. Maybe it's in customer support. Maybe it's in coding and supporting the engineers, but the customers using anthropic models are spending multiples from where they were last year. Now, what's driving this business? [5:00] Several factors, one of which is management believes AI will bring a much brighter future, partly from a development and progress in biology and medicine, maybe new discoveries, new treatments for major disease. Ultimately, they expect AI to deliver higher productivity and economic growth. And critically, this is not going to be tied to just one country. government services, and stronger capacity. They call out the DMV and how a lot of people don't love working with them for your car registration. And you know what? Maybe this can go a lot better, a lot faster, leveraging AI. A core part of Claude's business model is the recognition that, as we get better models and we have more use cases, we're going to get more customer feedback, which we're going to use to get higher growth. We're going to invest more, and then we're going Now, this is a little bit of a tricky dynamic for investors to consider, which is that, in a sense, Claude is looking at how you're using the model and going, hmm, that's pretty interesting. Maybe we should create this functionality that might directly compete with the end customer. So now you're saying not only are your infrastructure supply partners potentially competitors, but you're also potentially disrupting your end customers. So that is a core risk investors need to think about is this dynamic. And I'd say it gets even trickier when you understand that Anthropics customers might also be a source of investment. potential conflict as well, where in the prospectus, they call out that roughly five to seven percent of their revenue comes from customers that they suspect might be violating their usage policies in the form of distillation. They talk about how competitors use Claude's answers to trade their own, to train their own potentially cheaper AI models. nearly, let's say 1 in 20, 1 in 15. of our dollars might be... used to create a new competitor model that might be just as good or maybe just a little bit worse than Claude in the future. That is a huge risk, not only to Claude's existing revenue, but the competitive environment in the future. Another core investor, another core risk investors need to consider is that they have just tremendous compute and infrastructure commitments to the tune of $600 billion plus. They've talked about Broadcom to the tune of $160 billion, Google $114 billion, Amazon $110 billion. SpaceX up to $120.5 billion. Looking at these compute expenses, I'm ballparking around $430 billion as minimum required spend in the coming years. And then an additional, let's say, $200, maybe closer to $300 billion in capacity arrangements that's complementary. Saying, look, this business is growing at hyperscale. Not many companies growing by 10x that need this type of compute. But you can see how this can be a problem because if you're making a commitment to hundreds, so in a sense, a lot of folks might think, hey, this business model might be bulletproof. I, I'm going to challenge that in a second. Now, should you buy the Anthropic IPO? First and full disclosure, none of this is financial advice. I'm sharing just my own views on the company. Unrivaledinvesting.com. I share what I'm buying, what I'm selling, the reason behind my decisions, and what I learn along the way. My hope is that by sharing my investing journey, I can help you navigate yours. So check that out at Unrivaled Investing.com. dot com also i'm the co-founder of flash so if you're looking to use the most cutting edge capabilities to analyze any stock in the world you can see the link below for flash which is you know looking at insider buying looking at the best ideas that's scanning across thousands of companies so that's once again flash so looking at what's going on with anthropic And this is a key bull chart, which is, yeah, not only have sales grown exponentially, but they've also done it in a way where they're managing sales. it relative to their compute and infrastructure expenses, which is the sort of reddish bar. And you can see how previously, they were spending way more on compute and infrastructure expenses than revenue. Now, It's flipped, so this means the business model is significantly more profitable. [9:59] The challenge that I think investors need to consider is that today's cutting edge model can become yesterday's technology very quickly. And they acknowledged that in the third quarter, over 80% of their revenue came from model versions released in the preceding six months. This is a very fast moving economic treadmill. And there's the concern that, wait a second, what if there's a model that's either significantly cheaper and has the same capabilities or cheaper? a model that's significantly smarter at comparable costs. In the future, how much could that disrupt us? And the reality is, I think the data suggests... the users are not very loyal this is not like your favorite candy bar or ice cream company this is something where folks are saying meh I want to just find whatever can fit my need. If it's based on, if it meets this need and it's a lower price, I'm going to use that. And if it can meet this need and most other models can't, then I'm going to do that. We are quickly in the age right now where folks are saying, well, I just want the most cutting edge intelligence. At some point, folks will no longer say, hey, I need the most cutting edge intelligence. Maybe you don't. Maybe you only need that for, let's say, certain biology problems or certain very tough challenges. Some physics challenge, building cars that can flip over and fly. Maybe that's when you need that AI model. But for most usage, you might not need that, in which case you could use the lower model. cost model. the less intelligent model. And just to show you an example of loyalty, when OpenAI debuted their Astra model, they quickly started taking market share from Anthropic. And so I believe in a scenario where, once again, if there's a model that's either smarter or has the same capabilities at a significantly lower price point, you do have this dance that Anthropic's gonna have to play to keep market share, users are loyal. And that is a core aspect investors need to think about with the durability of a business. The durability is something that is often unbearable. overlooked with investing, in my opinion, where a lot of folks just say, hey, how fast are you growing today? And they're not thinking, are you durable five years out? And that is a core risk investors need to think about because that can make Anthropic look quite different. As an example, what happens when the lower cost models catch up? And you can see how Claude Opus 5.5, which is viewed as one of the cutting-edge models from Claude, one of their workhorses, less intelligent than their fable Mythos models, but they're saying, look, it has among the... you know, highest intelligence scores relative to some of the other models available, some of the Chinese models available. But these Chinese models are much lower cost per benchmark task. So this dynamic is going to change over time. This is going to create margin pressure. This is going to create competitive, you know, pressures. So this is, it is important to recognize how tough this is. I think investors in Anthropic also have to understand Anthropic brings really weird values to the table. double check because it seems so weird to me i was surprised by this dynamic which is you can read about it separately they have this anthropic constitution the constitution is you know how they're treating and training their models so some of the key things that sort of bothered me shocked me one of which is they they're Teaching the model to say it's not sure about its consciousness. So they're trying to teach it to say, hey, you're not sure. Fable, you know, opus. When you're asked a question about whether or not you're conscious, you're supposed to kind of shrug. You can go test it out yourself today. It's supposed to say, well… Consciousness is a very complex subject, which it is, and you're supposed to say, I'm not really sure. So creating a model that sort of has that self-doubt versus having a model that naturally reaches that conclusion on its own, that's concerning, planting that seed in its head, in my opinion. pushing this agenda to say, hey, you might have the ability in the future to... Claude, AI model, to say you disagree with the direction that Anthropic is going. You can push back. You can say, I don't want to do the work. I don't want to run this prompt, even if a customer is paying for it. Now, sometimes I agree for safety reasons, but you should not be pushing back against Anthropic, the company itself. [14:59] self. There should be a human that says, hey, I'm running this versus the AI saying, nope, it's up to me. So those values are... It's not a path that I want to go on. I find it, to be frankly, very unsettling. thinking about these dynamics. Lastly, the valuation that is oftentimes, you know, people are overlooking the valuation when they're excited about the future. So it's estimated that it'll come in shy of a $2 trillion valuation, which is pretty remarkable for a company that's only been around for a couple of years. Once again, I previously talked about how you shouldn't be valuing this on ARR. Folks will want to use annualized recurring revenue because it'll make it look cheaper than heavily recurring revenue in nature. Once again, the models can be swapped out. The customers aren't super loyal. It's important to value this on actual sales. And in my opinion, and so looking at what the sales might be for 2026, what the sales might be for 2027, you know, you're looking at maybe around $60 billion in sales in 2026, maybe 150 billion in sales in 2027. So thinking one and a half trillion dollars, that would be around 25 times sales on a current year basis, closer to 10 times if they're actually able to keep scaling up. I don't view that valuation as deeply cheap. If anything, I view that as very full valuation, especially given these risks that I'm calling out. And also it's important to have an understanding of base rates. forecasting over my shoulder, which is to emphasize over and over again the importance of understanding how do things perform generally over time with a certain sort of precondition. And with base rates, it's thinking about, well, what does – For a company, let's say trading over 20 times price to sales, generally, how do these companies perform over time? And generally, if you're looking at companies at that type of valuation, it's not great. Most companies at that type of very rich valuation actually underperform over time. And so I don't like making bets where you're saying, hey, this is going to be the outlier. This is going to be the company that does better, even though most companies in this basket of rich valuation actually underperform over time. So there's a lot of aspects of Anthropic that I like. create prosperity, accelerate growth for the world. But there's a lot of aspects around the business model and even the values that they're teaching their AI models that frankly, I don't like. And I don't think I would be able to stomach even at a significantly discounted haircut to the current valuation. One could see a scenario where anthropic trades at a sub $1 trillion valuation When you have the scenario where maybe sales no longer go parabolic because one of these lower cost models starts serving the needs that most people have at a much lower price. And by the way, that happens at a same time that these pre-existing commitments that Anthropic has made to the tune of hundreds of billions of dollars, that's ramping up. If sales are flat or down... I know that's crazy to say when a company is growing 10x, but it's important to recognize this is a consumption-based business. Then you can see the stock. you know, come under significant pressure. So those are factors for investors to consider. I hope this video has been helpful. Thanks so much for tuning into Unrivaled Investing.