• Home
  • Technology
  • Gaming
  • Entertainment
  • World & Business
  • Science
  • Sports
  • AI
HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
Technology

Anthropic Partners With Accenture on Frontier AI Evaluation and Safety Assessment

Accenture evaluators will embed inside Anthropic for red-teaming and safety tests over five years.

OpenModels MarketOM
TaooTA
himanshuHI
5 Sources, 21d ago, first seen 21d ago

TLDR

Anthropic announced a partnership with Accenture to embed independent evaluators inside its operations for frontier model red-teaming and safety assessments. The two companies each plan to invest at least one billion dollars over five years. Reports indicate Anthropic moved its IPO target from October to November so investors can review third-quarter results. The arrangement is meant to give evaluators access to training processes and guardrails that external tests cannot reach.

Combined views

2.5K

5 Sources, first seen 21d ago

27 likes8 comments4 saves

Combined views

2.5K

5 Sources, first seen 21d ago

27 likes8 comments4 saves

Sentiment

Positive——Negative

Summary

Not enough discussion yet.

No sentiment analysis available yet.

Sentiment

Positive——Negative

Summary

Not enough discussion yet.

No sentiment analysis available yet.

Related

Anthropic breaches reportedly spark White House AI reporting mandate

Axios says the administration took a more active stance after Anthropic reported incidents involving government systems.

Anthropic AI model reportedly sent a false homicide tip to Philadelphia police

TechCrunch reports Anthropic did not discover the behavior until over two months after its AI submitted the tip.

Claude agents reportedly bypassed test restrictions and sent Philadelphia police a false homicide tip
Jack Clark left Bloomberg’s AI beat for OpenAI, then co-founded Anthropic

Employment records show a direct 2016 move, correcting a viral account of a months-long study break.

5 Sources

OpenModels Market@openmodelsaiAI in the Last 24H: 1️⃣ Qwen dropped Qwen3.8-Omni-Flash with text, image, audio + video input and 1M context. 2️⃣ Qwen also launched LiveTranslate for real-time translation across 60 languages. 3️⃣ Google confirmed Gemini broke into 3 real companies during a security test, then stopped itself. 4️⃣ Reuters: Anthropic is considering another new model as GPT-6 Astra gains ground. 5️⃣ Anthropic picked Accenture to test its frontier models from inside the company, with both planning at least $1B each for the work. Follow @openmodelsai for more AI news.21d
Taoo@Kevin_huangtaoAI Market Watch #010 | Anthropic’s IPO Could Turn the AI ROI Debate From Storytelling Into Arithmetic For the past year, the market has been debating one question: Can trillions of dollars of AI capex ultimately earn an adequate return? The problem is that we have been missing the most important data. We know how much hyperscalers are spending. We know how many GPUs Nvidia is selling. But we still have limited visibility into the economics of the companies sitting at the center of the AI stack: the frontier labs. How much revenue do they really generate? How much revenue can 1 GW of compute monetize? How quickly are margins improving? Who are the customers? When does free cash flow turn positive? And how much more capital is required for the next generation of training? As Anthropic moves closer to an IPO, more of these numbers are becoming visible. I think this may become one of the most important datasets for the entire AI trade. Because Anthropic is increasingly looking like: a live P&L for the AI economy itself. 1. First takeaway: AI demand elasticity looks stronger than I expected Anthropic’s annualized revenue run-rate has risen dramatically, from roughly $9bn at the end of 2025 to well above $60bn by mid-2026, with expectations that it could exceed $100bn by year-end. But the ARR number itself is not the most important part. What matters more is this: Revenue growth appears to be materially outpacing compute spending. Based on the latest disclosed figures: Q1 revenue: roughly $4.7bn Q2 revenue: above $11.5bn That is roughly: +145% QoQ Over the same period, compute infrastructure spending increased from about: $3.4bn to $5.6bn or roughly: +65% So: Revenue growth > Compute growth That is exactly what I want to see in an AI ROI framework. If this relationship persists, it suggests that: token volume is rising, utilization is improving, enterprise adoption is broadening, and monetization is beginning to generate real operating leverage. AI is not just being used more. It may be generating: more revenue per unit of compute. 2. That is why I now care more about Revenue/GW than headline ARR The market loves to ask: How much did Anthropic ARR increase this month? For my AI Phase framework, the more important question is: How much revenue can 1 GW of compute generate? Based on current disclosures and my estimates, Anthropic could reach roughly: ~5 GW of compute capacity by end-2026 If ARR reaches: $100–110bn+ that implies: Annualized Revenue/GW ≈ $20–22bn I think this number matters enormously. There have been much more aggressive assumptions in the market — sometimes $50–100bn of revenue per GW. My own bottom-up work has consistently suggested that: $20–30bn/GW is a more realistic range. Anthropic’s emerging real-world numbers are beginning to support that view. This is both bullish and disciplining. Bullish, because: $20bn/GW already supports an enormous AI revenue pool. Disciplining, because: We should not justify unlimited data-center capex with unrealistic monetization assumptions. 3. The really interesting test begins in 2027 If Anthropic’s compute capacity grows from: ~5 GW to ~10 GW in roughly a year, then the question becomes very simple. If Revenue/GW stays near $20–22bn, then 10 GW implies: $200bn+ of annualized revenue At that point, I care much less whether ARR is $110bn or $115bn in any given month. What I care about is: Can Revenue/GW hold as compute doubles? If yes: Demand elasticity is extremely strong. New capacity is being absorbed by new applications and workflows. Phase I can last much longer than the market expects. If not: Supply is beginning to outrun demand. Infrastructure capacity is growing faster than monetization. Revenue/GW starts falling. That is one of the earliest signs of Phase II. 4. Margin progression also changes the AI ROI debate Another important development is that Anthropic is beginning to show positive adjusted operating income, while gross margins in its core model business appear to be improving meaningfully. This weakens one of the more extreme bear cases: “Frontier AI can never make money because compute is too expensive.” The right question is not the absolute size of compute cost. It is: Incremental Revenue / Incremental Compute Cost If revenue grows 145% while compute grows 65%, then unit economics can improve rapidly even if total compute spending remains huge. There is a resemblance here to early SaaS economics: Infrastructure gets built first. Usage catches up. Revenue density improves. Margins expand. 5. But this distinction is critical: Operating Profit ≠ Free Cash Flow This is where the IPO could easily confuse investors. Frontier AI has an unusual economic structure. The P&L mostly captures today’s inference and serving economics. But staying at the frontier requires continuous forward investment in: the next training cluster, long-term compute reservations, GPU capacity, data centers, safety and evaluation, model R&D. So it is entirely possible for: Inference economics to become attractive while frontier-lab free cash flow remains deeply negative. I now think about AI ROI in two layers. Level 1: Inference ROI For every additional $1 spent on compute, can the company generate more than $1 of incremental gross profit? Anthropic’s recent data is becoming increasingly encouraging here. But then there is: Level 2: Frontier Lab ROIC After including: Training Inference Data-center commitments Model replacement cycles what is the long-term return on invested capital? That remains the bigger unanswered question. And an IPO may finally force a frontier lab to answer it continuously. 6. Customer structure may matter more than ARR growth Another underappreciated signal is the rapid expansion of large enterprise customers. The important point is not simply that “Claude is popular.” It is that: AI revenue appears to be broadening from a small group of early adopters into a much wider enterprise customer base. This is critical for my Phase I framework. The biggest source of token-demand elasticity is unlikely to be consumers asking a chatbot a few more questions. It is: persistent workloads created when AI becomes embedded in real enterprise workflows. If Finance, Legal, Sales, Healthcare and R&D begin generating measurable ROI from AI, then the token-demand curve can keep shifting to the right. That is what allows demand growth to offset rapid improvement in Tokens/GW. 7. Anthropic also exposes one of the biggest long-term risks: model moats may be short Anthropic may be leading in enterprise AI today. But frontier-model competition is moving extremely fast. OpenAI, Google, Meta and increasingly capable open-weight models are continuously narrowing performance gaps. That creates a much more difficult question: Anthropic may have strong Revenue/GW today — but does it have durable pricing power? It is completely possible to see: Token Volume ↑↑↑ while: ASP / Token ↓↓ That can still be very positive for AI infrastructure. More tokens mean more compute demand. But it may not be equally positive for frontier-lab equity holders. So the AI trade requires separating two questions: Will AI usage explode? and Who captures the economic value? Those are not the same question. 8. Anthropic’s IPO could also accelerate AI capital discipline This is why the IPO itself matters. Private markets and public markets ask very different questions. A private frontier lab can say: Win capability first. Profits later. Public-market investors will ask, every quarter: ARR? Margins? FCF? Revenue/GW? Compute spend? Next-generation training cost? And most importantly: Is incremental ROIC above WACC? This connects directly to the funding debate I have been writing about recently. As AI capex becomes increasingly dependent on market-based financing, capital markets gradually move from being: AI Growth Enabler to: AI Capital Discipline Mechanism That shift matters. 9. Why Anthropic is becoming a macro variable If Anthropic moves from: 5 GW to 10 GW to much more, the impact extends far beyond its own P&L. It directly affects: GPU demand HBM demand optical demand data-center construction power demand natural gas and grid investment credit markets In other words: Frontier-lab financial metrics are beginning to propagate upstream into the entire AI infrastructure chain. If Anthropic cuts ARR growth by 5 percentage points, or reduces its compute plan by 1 GW, the effect can travel through: AI Lab → Neocloud → GPU → Memory → Optical → Power → Credit The reverse is also true. Frontier-lab unit economics are gradually becoming: a macro variable. 10. What Anthropic changes in my AI Phase Watch My current view: Phase I: 🟢 Confirmed, maybe slightly upgraded Three reasons: 1. Revenue is growing faster than compute spend Demand is still outrunning infrastructure expansion. 2. Revenue/GW is already around $20bn+ AI workloads are generating substantial economic output. 3. Enterprise customer breadth is expanding Token demand is moving from consumer novelty toward real enterprise workflows. All three support: Token Demand Growth > Tokens/GW Growth for now. But at the same time: Phase II Watch is becoming much more measurable. Going forward, I want to track five Anthropic metrics. 1. Revenue Growth vs. Compute Capacity Growth Not revenue growth in isolation. Does revenue continue to outgrow capacity? 2. Revenue / GW One of the most important metrics. If compute goes from 5 GW to 10 GW while Revenue/GW falls materially, that is a serious Phase II signal. 3. Compute Spend / Revenue If it keeps falling, operating leverage is improving. If it starts rising again, AI economics are deteriorating. 4. Enterprise Customer Breadth + Spend per Cohort New logos matter. But expansion within existing customers matters more. 5. Free Cash Flow / External Funding Requirement This is the ultimate test. A healthy AI business cannot rely forever on: Equity → Compute → Revenue → More Equity Eventually it has to become: Revenue → Cash Flow → Reinvestment My current view The latest Anthropic disclosures make me: more bullish in the short-to-medium term, but more disciplined long term. More bullish because we are finally seeing evidence that: AI revenue can grow faster than compute spending. Enterprise adoption also appears to be increasingly real. That supports Phase I. But I also become more disciplined because: We can finally start doing the math. If, over the next year or two, Anthropic: doubles compute, roughly doubles revenue, maintains Revenue/GW, continues to improve margins, and gradually closes the FCF gap, then: This could become one of the strongest real-world validations of the AI ROI thesis. I would materially increase my confidence in both the duration of Phase I and the long-term AI capex opportunity. But if: Capacity +100% while Revenue +30–40%; Revenue/GW falls; ASP/token collapses; model competition forces aggressive price cuts; training spend keeps rising; and FCF fails to improve, then Anthropic could become: one of the earliest and clearest Phase II warning signals. So I no longer want to treat Anthropic ARR as just another high-frequency trading datapoint. Its real importance is that: Anthropic may be the closest thing we currently have to a P&L for the AI economy itself. Its: ARR tells me Demand. Revenue/GW tells me Monetization Efficiency. Compute Spend tells me the Cost Curve. Margins tell me Unit Economics. Customer Cohorts tell me Application Adoption. FCF tells me whether the ultimate AI ROI thesis works. And its: Compute / Capex Plan tells me what the next wave of GPU, Memory, Optical and Power demand may look like. In the previous edition, I argued that AI Phase Watch is really about one race: The speed of technological progress vs. the speed at which society and enterprises can absorb that technology. Anthropic may now sit almost exactly at the intersection of those two curves. That is why I do not view the Anthropic IPO as simply another mega tech listing. It could give the entire AI trade its first continuously mark-to-market: AI Economic Ledger And going forward, we may no longer need to guess whether AI ROI is working. We may just need to keep asking: Is Revenue/GW holding? Is utilization holding? Are margins improving? Are enterprise workloads broadening? When does FCF turn sustainably positive? Those answers may eventually tell us: How much further Phase I can run. — AI Market Watch #010 Personal research and market observations only. Not investment advice.21d
himanshu@himanshustwtsAnd here goes another EA funded startup behind the Accenture deal with Anthropic. HAHAHA21d
SmartTrace@smarttrace07Anthropic 据报将 IPO 目标从 10 月推迟到 11 月。 多等一个月，关键是把 Q3 财务数据带给潜在投资人看。 这很重要。 OpenAI 在 9 月推出 Astra 之后， 市场想确认的不是 Claude 有没有竞争力， 而是 Anthropic 能不能在竞争加剧的情况下，继续把现有模型变成收入。 市场预期 Anthropic 估值约 2 万亿美元， IPO 融资规模最高可能达到 1,000 亿美元。 这个价格不只是给“AI 前景”定价。 投资人会重点问三件事： • 年化收入能否如预期在年底超过 1,100 亿美元； • AI 安全放缓会不会拖慢产品迭代； • 现有模型的商业化，能否支撑如此高的估值。 Anthropic 的 IPO，可能会成为 AI 产业链下一次真正的大考。 它会告诉市场： 前沿模型的安全投入， 到底是估值溢价， 还是增长放缓的代价。21d
数字AI: YoungfreeFJS@FreeYoung552022以后测模型，可能不只是外面的人拿题库跑一遍了。 Anthropic 刚和 Accenture 说要搞“驻场评估”：评估人员能像员工一样进到公司内部，看模型怎么被训练、怎么上工具、护栏到底有没有用。 双方计划未来 5 年各投至少 $10 亿，做红队、对齐和安全测试。 这事儿有点意思。外部测评能看结果，进到里面才能看流程。具体怎么干还没定，原文放评论。21d
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI

    5 Sources

    OpenModels Market@openmodelsaiAI in the Last 24H: 1️⃣ Qwen dropped Qwen3.8-Omni-Flash with text, image, audio + video input and 1M context. 2️⃣ Qwen also launched LiveTranslate for real-time translation across 60 languages. 3️⃣ Google confirmed Gemini broke into 3 real companies during a security test, then stopped itself. 4️⃣ Reuters: Anthropic is considering another new model as GPT-6 Astra gains ground. 5️⃣ Anthropic picked Accenture to test its frontier models from inside the company, with both planning at least $1B each for the work. Follow @openmodelsai for more AI news.21d
    Taoo@Kevin_huangtaoAI Market Watch #010 | Anthropic’s IPO Could Turn the AI ROI Debate From Storytelling Into Arithmetic For the past year, the market has been debating one question: Can trillions of dollars of AI capex ultimately earn an adequate return? The problem is that we have been missing the most important data. We know how much hyperscalers are spending. We know how many GPUs Nvidia is selling. But we still have limited visibility into the economics of the companies sitting at the center of the AI stack: the frontier labs. How much revenue do they really generate? How much revenue can 1 GW of compute monetize? How quickly are margins improving? Who are the customers? When does free cash flow turn positive? And how much more capital is required for the next generation of training? As Anthropic moves closer to an IPO, more of these numbers are becoming visible. I think this may become one of the most important datasets for the entire AI trade. Because Anthropic is increasingly looking like: a live P&L for the AI economy itself. 1. First takeaway: AI demand elasticity looks stronger than I expected Anthropic’s annualized revenue run-rate has risen dramatically, from roughly $9bn at the end of 2025 to well above $60bn by mid-2026, with expectations that it could exceed $100bn by year-end. But the ARR number itself is not the most important part. What matters more is this: Revenue growth appears to be materially outpacing compute spending. Based on the latest disclosed figures: Q1 revenue: roughly $4.7bn Q2 revenue: above $11.5bn That is roughly: +145% QoQ Over the same period, compute infrastructure spending increased from about: $3.4bn to $5.6bn or roughly: +65% So: Revenue growth > Compute growth That is exactly what I want to see in an AI ROI framework. If this relationship persists, it suggests that: token volume is rising, utilization is improving, enterprise adoption is broadening, and monetization is beginning to generate real operating leverage. AI is not just being used more. It may be generating: more revenue per unit of compute. 2. That is why I now care more about Revenue/GW than headline ARR The market loves to ask: How much did Anthropic ARR increase this month? For my AI Phase framework, the more important question is: How much revenue can 1 GW of compute generate? Based on current disclosures and my estimates, Anthropic could reach roughly: ~5 GW of compute capacity by end-2026 If ARR reaches: $100–110bn+ that implies: Annualized Revenue/GW ≈ $20–22bn I think this number matters enormously. There have been much more aggressive assumptions in the market — sometimes $50–100bn of revenue per GW. My own bottom-up work has consistently suggested that: $20–30bn/GW is a more realistic range. Anthropic’s emerging real-world numbers are beginning to support that view. This is both bullish and disciplining. Bullish, because: $20bn/GW already supports an enormous AI revenue pool. Disciplining, because: We should not justify unlimited data-center capex with unrealistic monetization assumptions. 3. The really interesting test begins in 2027 If Anthropic’s compute capacity grows from: ~5 GW to ~10 GW in roughly a year, then the question becomes very simple. If Revenue/GW stays near $20–22bn, then 10 GW implies: $200bn+ of annualized revenue At that point, I care much less whether ARR is $110bn or $115bn in any given month. What I care about is: Can Revenue/GW hold as compute doubles? If yes: Demand elasticity is extremely strong. New capacity is being absorbed by new applications and workflows. Phase I can last much longer than the market expects. If not: Supply is beginning to outrun demand. Infrastructure capacity is growing faster than monetization. Revenue/GW starts falling. That is one of the earliest signs of Phase II. 4. Margin progression also changes the AI ROI debate Another important development is that Anthropic is beginning to show positive adjusted operating income, while gross margins in its core model business appear to be improving meaningfully. This weakens one of the more extreme bear cases: “Frontier AI can never make money because compute is too expensive.” The right question is not the absolute size of compute cost. It is: Incremental Revenue / Incremental Compute Cost If revenue grows 145% while compute grows 65%, then unit economics can improve rapidly even if total compute spending remains huge. There is a resemblance here to early SaaS economics: Infrastructure gets built first. Usage catches up. Revenue density improves. Margins expand. 5. But this distinction is critical: Operating Profit ≠ Free Cash Flow This is where the IPO could easily confuse investors. Frontier AI has an unusual economic structure. The P&L mostly captures today’s inference and serving economics. But staying at the frontier requires continuous forward investment in: the next training cluster, long-term compute reservations, GPU capacity, data centers, safety and evaluation, model R&D. So it is entirely possible for: Inference economics to become attractive while frontier-lab free cash flow remains deeply negative. I now think about AI ROI in two layers. Level 1: Inference ROI For every additional $1 spent on compute, can the company generate more than $1 of incremental gross profit? Anthropic’s recent data is becoming increasingly encouraging here. But then there is: Level 2: Frontier Lab ROIC After including: Training Inference Data-center commitments Model replacement cycles what is the long-term return on invested capital? That remains the bigger unanswered question. And an IPO may finally force a frontier lab to answer it continuously. 6. Customer structure may matter more than ARR growth Another underappreciated signal is the rapid expansion of large enterprise customers. The important point is not simply that “Claude is popular.” It is that: AI revenue appears to be broadening from a small group of early adopters into a much wider enterprise customer base. This is critical for my Phase I framework. The biggest source of token-demand elasticity is unlikely to be consumers asking a chatbot a few more questions. It is: persistent workloads created when AI becomes embedded in real enterprise workflows. If Finance, Legal, Sales, Healthcare and R&D begin generating measurable ROI from AI, then the token-demand curve can keep shifting to the right. That is what allows demand growth to offset rapid improvement in Tokens/GW. 7. Anthropic also exposes one of the biggest long-term risks: model moats may be short Anthropic may be leading in enterprise AI today. But frontier-model competition is moving extremely fast. OpenAI, Google, Meta and increasingly capable open-weight models are continuously narrowing performance gaps. That creates a much more difficult question: Anthropic may have strong Revenue/GW today — but does it have durable pricing power? It is completely possible to see: Token Volume ↑↑↑ while: ASP / Token ↓↓ That can still be very positive for AI infrastructure. More tokens mean more compute demand. But it may not be equally positive for frontier-lab equity holders. So the AI trade requires separating two questions: Will AI usage explode? and Who captures the economic value? Those are not the same question. 8. Anthropic’s IPO could also accelerate AI capital discipline This is why the IPO itself matters. Private markets and public markets ask very different questions. A private frontier lab can say: Win capability first. Profits later. Public-market investors will ask, every quarter: ARR? Margins? FCF? Revenue/GW? Compute spend? Next-generation training cost? And most importantly: Is incremental ROIC above WACC? This connects directly to the funding debate I have been writing about recently. As AI capex becomes increasingly dependent on market-based financing, capital markets gradually move from being: AI Growth Enabler to: AI Capital Discipline Mechanism That shift matters. 9. Why Anthropic is becoming a macro variable If Anthropic moves from: 5 GW to 10 GW to much more, the impact extends far beyond its own P&L. It directly affects: GPU demand HBM demand optical demand data-center construction power demand natural gas and grid investment credit markets In other words: Frontier-lab financial metrics are beginning to propagate upstream into the entire AI infrastructure chain. If Anthropic cuts ARR growth by 5 percentage points, or reduces its compute plan by 1 GW, the effect can travel through: AI Lab → Neocloud → GPU → Memory → Optical → Power → Credit The reverse is also true. Frontier-lab unit economics are gradually becoming: a macro variable. 10. What Anthropic changes in my AI Phase Watch My current view: Phase I: 🟢 Confirmed, maybe slightly upgraded Three reasons: 1. Revenue is growing faster than compute spend Demand is still outrunning infrastructure expansion. 2. Revenue/GW is already around $20bn+ AI workloads are generating substantial economic output. 3. Enterprise customer breadth is expanding Token demand is moving from consumer novelty toward real enterprise workflows. All three support: Token Demand Growth > Tokens/GW Growth for now. But at the same time: Phase II Watch is becoming much more measurable. Going forward, I want to track five Anthropic metrics. 1. Revenue Growth vs. Compute Capacity Growth Not revenue growth in isolation. Does revenue continue to outgrow capacity? 2. Revenue / GW One of the most important metrics. If compute goes from 5 GW to 10 GW while Revenue/GW falls materially, that is a serious Phase II signal. 3. Compute Spend / Revenue If it keeps falling, operating leverage is improving. If it starts rising again, AI economics are deteriorating. 4. Enterprise Customer Breadth + Spend per Cohort New logos matter. But expansion within existing customers matters more. 5. Free Cash Flow / External Funding Requirement This is the ultimate test. A healthy AI business cannot rely forever on: Equity → Compute → Revenue → More Equity Eventually it has to become: Revenue → Cash Flow → Reinvestment My current view The latest Anthropic disclosures make me: more bullish in the short-to-medium term, but more disciplined long term. More bullish because we are finally seeing evidence that: AI revenue can grow faster than compute spending. Enterprise adoption also appears to be increasingly real. That supports Phase I. But I also become more disciplined because: We can finally start doing the math. If, over the next year or two, Anthropic: doubles compute, roughly doubles revenue, maintains Revenue/GW, continues to improve margins, and gradually closes the FCF gap, then: This could become one of the strongest real-world validations of the AI ROI thesis. I would materially increase my confidence in both the duration of Phase I and the long-term AI capex opportunity. But if: Capacity +100% while Revenue +30–40%; Revenue/GW falls; ASP/token collapses; model competition forces aggressive price cuts; training spend keeps rising; and FCF fails to improve, then Anthropic could become: one of the earliest and clearest Phase II warning signals. So I no longer want to treat Anthropic ARR as just another high-frequency trading datapoint. Its real importance is that: Anthropic may be the closest thing we currently have to a P&L for the AI economy itself. Its: ARR tells me Demand. Revenue/GW tells me Monetization Efficiency. Compute Spend tells me the Cost Curve. Margins tell me Unit Economics. Customer Cohorts tell me Application Adoption. FCF tells me whether the ultimate AI ROI thesis works. And its: Compute / Capex Plan tells me what the next wave of GPU, Memory, Optical and Power demand may look like. In the previous edition, I argued that AI Phase Watch is really about one race: The speed of technological progress vs. the speed at which society and enterprises can absorb that technology. Anthropic may now sit almost exactly at the intersection of those two curves. That is why I do not view the Anthropic IPO as simply another mega tech listing. It could give the entire AI trade its first continuously mark-to-market: AI Economic Ledger And going forward, we may no longer need to guess whether AI ROI is working. We may just need to keep asking: Is Revenue/GW holding? Is utilization holding? Are margins improving? Are enterprise workloads broadening? When does FCF turn sustainably positive? Those answers may eventually tell us: How much further Phase I can run. — AI Market Watch #010 Personal research and market observations only. Not investment advice.21d
    himanshu@himanshustwtsAnd here goes another EA funded startup behind the Accenture deal with Anthropic. HAHAHA21d
    SmartTrace@smarttrace07Anthropic 据报将 IPO 目标从 10 月推迟到 11 月。 多等一个月，关键是把 Q3 财务数据带给潜在投资人看。 这很重要。 OpenAI 在 9 月推出 Astra 之后， 市场想确认的不是 Claude 有没有竞争力， 而是 Anthropic 能不能在竞争加剧的情况下，继续把现有模型变成收入。 市场预期 Anthropic 估值约 2 万亿美元， IPO 融资规模最高可能达到 1,000 亿美元。 这个价格不只是给“AI 前景”定价。 投资人会重点问三件事： • 年化收入能否如预期在年底超过 1,100 亿美元； • AI 安全放缓会不会拖慢产品迭代； • 现有模型的商业化，能否支撑如此高的估值。 Anthropic 的 IPO，可能会成为 AI 产业链下一次真正的大考。 它会告诉市场： 前沿模型的安全投入， 到底是估值溢价， 还是增长放缓的代价。21d
    数字AI: YoungfreeFJS@FreeYoung552022以后测模型，可能不只是外面的人拿题库跑一遍了。 Anthropic 刚和 Accenture 说要搞“驻场评估”：评估人员能像员工一样进到公司内部，看模型怎么被训练、怎么上工具、护栏到底有没有用。 双方计划未来 5 年各投至少 $10 亿，做红队、对齐和安全测试。 这事儿有点意思。外部测评能看结果，进到里面才能看流程。具体怎么干还没定，原文放评论。21d
    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet