节目 从零开始学AI 下一集
第 77 集

Anthropic 路演秀:使命、算力与企业战略

SnapYT 特殊加更,中文配音
2026-11-11 approximately 34 minutes 进阶
音频已上线 点击收听
主讲人

这将是人类历史上最具前景的技术。AI正在捕捉认知的本质。二零二四年和二零二五年,编程已变得可行。如今我们开始发现,在其他知识工作领域端到端执行任务的智能体也已实现。最近更看到,生物学、数学和物理学等领域的根本性突破正成为可能。This is going to be the most promising technology in human history. AI is capturing the essence of cognition. In 2024 and 2025, coding became possible. And now we're starting to find that agents that do things end to end in other areas of knowledge work are possible. And even more recently, we're starting to see that fundamental discoveries in areas like biology and mathematics and physics are becoming possible.

主讲人

如果人类能用AI治愈癌症,坏人也能用它制造生物武器或发动网络攻击。不解决这些风险,我们就无法成功。我们下定决心必须做好这件事。因为这就是我,以及许多在Anthropic工作的同事,投身这项技术的初衷。智力如今已成为人类可以构建的能力。其影响深远。我们相信它将大幅提升每个人的能力,拓展全球认知工作的容量,从而重塑每一个行业与学科。巨大的经济影响始于前沿智力,也就是Claude能做到的事。二零二四年,Claude能回答问题、检索信息或总结文档。到二零二五年底,它能进行深度研究和多步骤项目,比如重构整个代码库。If humans can use AI to cure cancer, an evil human can use AI to make a bioweapon or do a cyber attack. We cannot succeed if we don't address these risks. And we are absolutely determined to get this right because this is why I, and this is why many people who work at Anthropic, this is why we work on this technology. Intelligence is now something humans can build. The implications of this are profound. We believe it will make every person far more capable and expand the world's capacity for cognitive work, transforming every industry and every discipline. This enormous economic impact begins with frontier intelligence, with what Claude can do. In 2024, Claude could answer a question, retrieve information, or summarize a document. By the end of 2025, Claude could do deep research and multi-step projects, like refactoring an entire code base.

主讲人

过去需要人类花费数小时甚至数天才能完成的任务类型。如今,Claude能接手那些曾需多人耗时数月甚至数年研究、分析与执行的复杂开放式项目,并在几天内完成。展望未来不久,我们认为前沿AI有望推动人类一些最伟大的突破,并开辟全新市场。我们至今的成功源于五大核心优势。分别是我们的使命与专注、人才与文化、前沿研究与安全工作、算力策略以及企业战略。这一切都始于我们的使命:负责任地开发和维护先进AI,造福人类长远利益。The type of tasks that required hours or even days of human work.Today, Claude can take on complex, open-ended projects that once took months or years of research, analysis, and execution by many people, and complete them in days.In the near future, we see potential for frontier AI to drive some of humanity's greatest breakthroughs and open entirely new markets.Our success thus far emerges from five core advantages.These are our mission and focus, talent and culture, frontier research and safety work, our approach to compute, and our enterprise strategy.All of this begins with our mission: the responsible development and maintenance of advanced AI for the long-term benefit of humanity.

主讲人

它是凝聚团队的核心力量,也是精准执行的关键。我们管理着有限的资源,包括算力、人才和研究能力。使命为团队指明方向,让资源发挥最大效益。面临艰难抉择时,使命能帮我们果断决策。它促使我们将Mythos等模型专供网络防御者使用。也让我们承诺让Claude绝不植入广告。这一使命持续吸引业界顶尖人才。自创立起便是如此。七位联合创始人中,有几位正是如今AI产业基石理念的提出者。顶尖人才遍布组织各层级,我们在吸引和留住领域精英方面始终走在行业前列。It's an organizing force and key to our focused execution. We manage a universe of finite resources, compute, talent, and research capacity. The mission is how we create clarity for the team and deploy those resources to maximum effect. At challenging moments that call for hard choices, our mission sharpens our decision making. It led us to deploy models like Mythos exclusively to cyber defenders. It also led us to commit to keep Claude free of advertising. This mission and focus continue to attract the very best talent in the industry. That's been true since our founding. Our seven co-founders include the originators of ideas that now make up the foundation of the AI industry. This concentration of top talent extends to all levels of the organization, and we lead the industry in attracting and retaining the brightest minds in the field.

出镜者

我为什么在Anthropic工作?Why do I work at Anthropic?

出镜者

我为什么选择Anthropic?我为什么留在这里?Why do I work at Anthropic? Why do I work at Anthropic?

出镜者

我为什么加入Anthropic?Why do I work at Anthropic?

出镜者

因为我天生是个乐观的人。Because I'm an optimist.

出镜者

我想给家人和全人类,一个更安全的世界。I want a safer, better world for my family and for humanity.

出镜者

我深信这项技术,能救下更多像他一样的人。Three years ago, my older brother died of a brain tumor at the age of 42, which is the age that I am now. And I believe deeply that through our mission and this technology, there will be a situation where people like my brother don't have to die.

出镜者

总得有人扛起重担,把这事做对。我不想只在旁边看着。Somebody has to carry the weight of getting this right, and I really didn't want to watch from the sidelines.

出镜者

这里的同事总会先问:这事该不该做?然后再问:能不能做?People here ask, "Is this the right thing to do?" Before they ask, "Can we do it?"

出镜者

这些品质让团队真正高效。It's a rare combination of a group of people who care deeply about making AI go well and also care deeply about competence, coherence, integrity, qualities that actually make an organization effective.

主讲人

这些益处与风险,光明与阴影,正是我选择加入 Anthropic 的原因。我们让模型更安全、更可靠的工作,也正是让它们变得更强大的工作。我们关注实用的智能,关注模型处理现实任务的能力,以及为客户创造的价值。我们执着于回答一个问题:Claude 能做什么,其他模型却做不到?算力是支撑这一切的关键基础设施。我们在此建立了差异化优势,不仅在于如何搭建算力,更在于如何通过发现算力乘数来最大化利用它。我们是从创立之初就专注服务企业的领先 AI 实验室。我们训练前沿模型来处理知识型工作。因此,客户既能获得顶尖能力,又能享受超高性价比。These benefits and these risks, this light and this shade, precisely why I chose to come to work at Anthropic. The work we do to make our models safe and reliable is the same work that makes them more capable. We care about useful intelligence, how effectively a model handles a real world task and the value this delivers to customers. We're relentlessly focused on answering the question, what can Claude do for me that other models can't?Compute is the critical infrastructure that enables this work. We've built a differentiated advantage here, not just in how we build our compute capacity, but in how we make the most of this capacity through the discovery of compute multipliers. We are the leading AI lab that has been focused on building for the enterprise from day one. We train our frontier models to address knowledge work. As a result, our customers get the best of both capability and value.

主讲人

这些客户推动了我们的历史性增长,他们遍布金融、科技、制药、医疗、政府和制造等行业。这也正是我们的公益使命与安全重点自然契合之处。企业界和我们一样,同样关心前沿技术是否安全、可信且有用。随着前沿智能日益强大,我们对安全研究的专注就显得愈发重要。这对客户至关重要,也是释放 AI 全部潜力的前提。Anthropic 已全力承诺卓越地完成这项工作。我们的公益使命将我们带到今天,并将继续指引我们的工作,包括治理层面。除高管和董事会外,Anthropic 还受长期利益信托基金指导。该机构独立运作且无财务利益关联,负责选出董事会多数席位。These customers have enabled our historic growth,and they span industries across financial services, technology, pharma, healthcare, government, manufacturing, and more.And this is also where our public benefit mission and safety focus find natural alignment.Enterprises care that the frontier is safe, trustworthy, and useful as much as we do.As frontier intelligence grows more capable,our dedicated focus on safety research becomes that much more important.It's critically important to our customers and necessary for realizing AI's full potential.And Anthropic is fully committed to doing this work with excellence.Our public benefit mission is what brought us to this point,and it will continue to shape our work, including at the governance level.In addition to our executive leadership and our board of directors,Anthropic is advised by the Long-Term Benefit Trust,an independent, financially disinterested body that selects a majority of our board.

主讲人

长期利益信托基金的作用是协助 Anthropic 规划并履行公益使命,伴随公司发展。我们相信,AI 的未来应由民主社会塑造与引领。我们希望成为得力伙伴,协助政府适应 AI 能力提升带来的变革。我们在执行层面依靠政策团队开展工作。我们还通过 Anthropic 研究所等举措,分享前沿 AI 如何改变生活方式的研究。我们认为,这将是现代人类经历的最大转型,而我们才刚刚起步。这不仅关乎深远的经济潜力,更关乎 AI 助力人类繁荣的方方面面。从治愈疾病到加速发展,再到帮助人们过上更充实、富足的生活。所以看看我们的 Claude 模型家族,它是为了在所有目标上为您创造价值而设计的。无论您看重的是降低成本、提升速度,还是增强智能。The LTBT's role is to help Anthropic plan for and fulfill our public benefit mission as we grow.We believe AI's future should be shaped and led by democratic societies.We want to be good partners in helping governments adapt to the changes that come with AI's rising capability.We do this at the executive level through the work of our policy teamand through efforts like the Anthropic Institute,which shares research on how frontier AI shapes how we live our lives.We are still early in what we believe will be the biggest transformationof the human experience in the modern age.Not just the profound economic potential,but in all the ways in which we believe AI can help humanity flourish.From curing diseases to accelerating growth,to helping people live more fulfilled and abundant lives. So if you think about our Claude model family,it's designed to deliver value across all the goals you might have,whether you're optimizing for cost or speed or intelligence.

主讲人

在最高端,Mythos 和 Fable 不断突破前沿。我们看到客户用它加固网络基础设施或取得科学新发现。在另一端,我们拥有能力极强且极具成本优势的模型。企业可大规模调用它们处理常规的明确任务。我们重点思考的是,如何让每个人都能受益于全系列的模型。因为即便模型能力呈指数级增长,大众的使用习惯也不会剧变。我们在开发中遵循三大原则。首先是贴近客户现有场景。使用 Claude 的途径有好几种。首推我们自家的应用,例如 Claude Code 或桌面及移动端 App。同时,Claude 也已集成进 Word 和 Excel 等企业软件。我们会深耕垂直行业,去弄明白该领域的专家究竟需要什么,才能立刻上手见效?At the top end, Mythos and Fable push the frontier forward. We see customers use it to harden cyber infrastructure or make new scientific discoveries. At the other end, we have models that are extremely capable and highly cost effective, and businesses can rely on them at scale for well-defined routine tasks. A key thing we think about is how everyone can get the benefit of that whole spectrum of models. Because even as model capability increases exponentially, what most people do with the models doesn't change that fast. We think about three principles as we build. The first is meeting customers where they are. There's a few ways to use Claude. First is through our own applications like Claude Code or the Claude desktop or mobile app. But Claude is also available in enterprise applications like Word and Excel. We go deep into verticals and try to understand what does an expert in this domain need to be successful immediately?

主讲人

比如你是律师,我们有法律版 Claude 套件,直接连上你需要的工具。每个领域我们都这么考虑。医疗、金融、生命科学等领域都有专属套件。第二条原则是模型和产品同步开发。我们不为今天的模型做产品,而是为明天的模型做准备。Claude Code 就是很好的例子。几年前我们就发现,未发布模型的代码理解能力在提升。虽然还不能全搞定,但趋势很明显。我们顺着这个思路做了个产品,几个月后 Claude Code 就上线了。它让大家体验到了全新的编程方式,也明白了 AI 能做什么。For instance, if you're a lawyer, we have a Claude for Legal bundle, which connects directly to the tools you need. This is something we think about for every domain. We have bundles for healthcare, financial services, life sciences, and more. Our second principle is that we build models and products side by side. We develop products for tomorrow's models, not just today's. Claude Code is a great illustration of why that matters. A couple of years ago, we could see that our pre-release models were getting better at understanding code. Couldn't quite do it all yet, but we could see the trend line. We built a product around that thinking, and Claude Code was launched just a few months later. That introduced people to an entirely new way of coding and of understanding what AI can do.

主讲人

模型和产品的飞轮是我们持久的优势。模型变强,我们就基于新能力开发新功能。接着再训练模型,让它更擅长用户想做的事,从而强化整个系统。这在编程及其他领域都一样。第三条原则是拉近前沿技术和客户日常使用的距离。大多数人还把 AI 工具当成搜索替代品或简单问答机。它确实可以这样用。但这低估了人们真正能用这种智能做到的事。这时产品就成了关键的竞争优势。没有好产品,不断增强的智能和用户实际使用之间的差距只会越拉越大。我们认为未来要让 Claude 尽可能简单自然。这样大家就能像对待同事一样用它。This flywheel between model and product is a durable advantage in how we build.The model gets better, we develop new features powered by those advancements.Then we train the models to get even better at what people are trying to do,which reinforces the whole system.And we do that in coding and beyond.Our third principle is that we close the gap between the frontier and customers day to day.Most people still think about their AI tools as a replacement for search or a simple Q&A tool.And it can be those things.But that understates the power of what people can really do with this intelligence.This is where product becomes a meaningful competitive differentiator.Without a strong product, the difference between growing intelligence and where users are just keeps compounding.We think the future will involve making Claude feel as simple and natural as possible.So people everywhere can just treat Claude the same way you'd treat a colleague.

主讲人

并且每次都能输出高质量结果。我们认为这三条原则,就是把前沿智能带给每个人的方法。我们说想给人“超能力”,指的就是这个。Claude 先帮你把该做的事做得更快更好。然后它会接手那些你从未想过尝试的事,因为它们曾经太难。未来,这意味着我们能一起做到过去不可能的事,让生活更美好、更充实。当我们发现 AI 的缩放定律时,就意识到 AI 能力会飞速提升。这就是我们创立 Anthropic 的原因,以确保如果 AI 成为变革性技术,它一定是安全且有益的。我们的核心判断是,AI 在这十年将取得巨大进展。正是这种变革性能力,决定了研究和安全必须齐头并进。and get high quality output every time.We think these three principles are how we bring frontier intelligence to everyone.This is what we mean when we say we want to give people superpowers.Claude starts by helping you do what you need to, better and faster.Then it takes on the things you never would have tried because they were just too hard.And in the future, it means all of us doing what used to be impossible,so we can all live better, fuller lives. When we discovered scaling laws in AI,we realized that AI capabilities could improve very fast.That's why we founded Anthropic,to ensure that if AI became a transformational technology,it would be safe and beneficial.Our fundamental bet is that AI would make tremendous progress this decade.These transformational capabilities are why research and safety must advance together.

主讲人

这是实现技术红利的唯一途径。我们从三个维度评估模型:能力、效率和安全性。我们优化模型,让它们能规划并准确完成耗时数小时甚至数天的复杂工作。Opus 5.5 就是个典型例子。有用户不到一天就完成了六十八万行代码迁移。几年前这还得让工程团队花上几周。我们认为在需要高度准确的长周期任务上,我们的模型是最顶尖的。这种能力始于模型训练。预训练阶段,模型会学习人类知识的总和:文本、图像、论文和代码。这个阶段它掌握基础的人类概念与思想。第二阶段是强化学习。在这里 Claude 学会如何有益、诚实且无害。这也是我们激发编码、知识工作和科学等实用能力的过程。It's the only way we'll realize the benefits of the technology. We assess our models across three dimensions: capability, efficiency, and safety. We optimize our models to be able to plan and complete complex work accurately across hours and even days. A good example is Opus 5.5. One of our users was able to complete a 680k line code migration in less than a day. This would have taken an engineering team weeks just a few years ago. We think our models are the very best at this long-horizon work, where accuracy really matters. The capability begins in the model training process. In the pre-training phase, a model learns from the totality of human knowledge: text, images, scientific papers, code. This phase is where it comes to understand fundamental human concepts and ideas. The second phase is reinforcement learning. This is where Claude learns to be helpful, honest, and harmless. It's where we draw out useful capabilities: encoding, knowledge work, and science.

主讲人

随着模型从对话助手转向自主智能体,强化学习已成为我们投入最大的研究方向之一,这也对客户至关重要。我们认为人工智能能将临床试验周期缩短百分之二十到三十。药物提前两三年上市,无药可治的患者就能早点用上。这对患者是巨大胜利,对百时美施贵宝(BMS)亦然。As models have moved from conversational assistance to autonomous agents,reinforcement learning has evolved into one of the largest areas of research investment,one that really matters to our customers.In 2021, I wrote an internal memooutlining how our most likely product might be an aligned AI programmer.I wrote a lot of other things too,but this one turned out to be pretty accurate.Claude's training process is integral to Anthropic's mission.We need to understand how to train frontier intelligence and how to make it safe.And this is how we deliver the benefits of AI. We think that AI will help us reduce the time to do a clinical trial by 20 to 30%.If you get a drug to market two or three years sooner,those are either cures or treatments that patients that actually don't have solutions can have sooner.So we think it's actually a huge win for patients,and it's also a huge win for BMS.

出镜者

我们要能监控运行、管安全、控权限,这些都是IT部门看重的企业级管控。We chose Anthropic for a couple reasons. One is they had the most mature enterprise capabilities. So it's one thing to have a chatbot that you can use. It's a whole different thing to actually put it in an environment where you're giving it access to some of the company's most confidential information, and you're able to do observability, manage the security, manage the fine-tuned controls of what you give users. So there's like the enterprise controls that you'd often see that IT organizations really value a lot.

出镜者

这是我们历史上上线最快、普及率最高的AI工具。We rolled it out across all 30,000 of our employees. This is the fastest, most well adopted AI tool in the history of what we've rolled out.

出镜者

过去答个简单问题要一两个月,现在快多了。So now you can ask a very, very complicated question, and it will go through, it will figure out which database or databases, 'cause it can actually cross those databases, can be queried, and then it will produce a report that actually has the context of the epidemiology built right into it. So that takes you from what would've typically taken one to two months to get a simple question answered,

主讲人

缩短到了四五分钟。我们现在终于迎来了新阶段,无论是高通量计算还是大语言模型,都能让算力和分析能力大幅扩展,以匹配生物学研究的规模。这将帮助你们优化并革新现有的诊疗标准。想想前沿 AI 实验室为何必须高效使用算力。我们在研发和模型训练的每个环节都在用它。这是交付模型给客户的方式,也是我们内部大规模使用 Claude 的途径。通过训练最高效的模型,同等算力能办更多事。这也降低了成本,既惠及客户,也压降我们的底层开支。我们正通过研究加速提升效率。发现步伐正在加快,很多提速来自 Claude 本身。部署前沿模型时安全至上。这一理念促成了联合创始人携手,也是 Anthropic 自创立起就坚守的基石。卓越的安全能力,是管控组织与系统性风险的关键。to being able to get it in four to five minutes. And we're now finally at a stage, both with high throughput computation, but also large language models, where you're able to scale up your computation, your analytics, to match the better scale of biology. That can help the way that you inform and change the standard of care. Consider all the reasons a frontier AI lab needs to use compute efficiently. We use it at every stage of the research and model training process. It's how we serve models to customers and how we use Claude at scale internally. By training what we believe to be the most efficient models, we can do more of all these things with the same amount of compute. It's also how we drive down costs, both for customers as well as our own underlying costs. We're accelerating these efficiency gains through our research process. Our pace of discovery is speeding up, and a lot of that acceleration comes from Claude itself. When you deploy frontier models, safety is critical. This idea brought the co-founders together, and it's been at the foundation of Anthropic from the beginning. Excellence in safety is how you manage organizational and systemic risk.

主讲人

这就是你们如何在编程和医疗等领域安全释放前沿能力的方法,客户非常看重这一点。它给了客户信任,让他们敢把重要工作交给人工智能。我们在每个阶段都内置了安全措施。通过对齐训练,我们让模型听指令,不采取有害行动。同时加上防护机制,防止被人乱用。我们还开创了可解释性研究,借此看清模型脑子里到底在发生什么。在 Anthropic 内部,我们越来越多地用前沿模型来优化自己的流程。我们叫它闭环,它推着咱们三大研究重点往前走。Claude 正慢慢变成咱们的平等搭档。Claude 越强,越能帮咱们干活。它提升了安全性,让训练和推理更高效,推动了安全工作,帮咱们更快试错,也让整个组织更给力。咱们最重要的工作还在后头。只要科研和安全一起往前推,相信就能帮人类迎来一些最重大的突破。It's how you safely enable frontier capabilities in fields like coding and medicine, and it's valued by our customers. It gives them the trust they need to point AI at important work. We build safety in at every stage. Through alignment training, we train models to follow user intent and avoid taking harmful actions, and we build in the right safeguards aimed at preventing misuse. We also pioneer fields like interpretability, which gives us insight into what's happening inside the model's brain. Within Anthropic, we increasingly turn frontier models back on improving our own processes. We call this closing the loop, and it drives forward all three of our research priorities. Claude is increasingly our equal partner here. The better Claude gets, the more it helps us with all of our work. It improves our security, it makes training and inference more efficient, it drives safety work, it helps us try out more research ideas more quickly, and it makes the whole organization more effective. Our most important work is still ahead. If we keep research and safety advancing together, we believe that we can help bring about some of the most significant breakthroughs in human history.

主讲人

拿个数翻倍,再翻倍,接着继续翻倍。这条曲线就是 Anthropic 核心业务的写照。我们早就发现,性能跟着算力走,涨得非常平稳。咱们的赌注是,互联网上全是上下文数据。要是限制模型大小的瓶颈没了会怎样?所以在二零二二或二三的时候,聊天机器人刚起步。你让它讲讲武士历史,它就回篇小作文。到了二零二四年,能写代码了,但 Claude 碰到难题还是会卡壳。到了二零二五年,模型开始干出让人吃惊的事。编程能力大幅变强。Take a number and double it, and double that, then double that.That curve is really at the heart of what we do at Anthropic.We noticed early on that performance seemed to scale very smoothly with compute.Our bet was, there's a whole internet's worth of context. What if the limiting factor of model size just went away?So in 2022 or 23, we had chatbots.You could ask them for a history of the samurai, and it would come back with a little essay.In 2024, coding becomes possible, but Claude still struggles on hard problems.In 2025, the models start to be able to do surprising things. Coding gets much better.

主讲人

Claude 甚至能打游戏,虽然打得挺菜。同年,它挖出了 OpenBSD 里藏了二十七年的漏洞。突然之间,它能跑多步骤流程,干脑力活了。它给一家量子计算公司搭了激光锁控制器。这活儿以前得靠一群博士干一周。而且它是一夜之间自己搞定的。咱们开始看到 Claude 在数学上做出基础发现,在生命科学前沿干活。这种加速让人有点晕。但大家总觉得这才刚上路。那条指数曲线还在前面铺着,一切才刚开始。Claude can even play video games, quite badly.This same year, it found a flaw in OpenBSD that had gone unnoticed for 27 years.Suddenly, it can do multi-step processes, knowledge work.It builds a controller for laser locks at a quantum computer company,a task that used to take a team of people with PhDs a week.And it does this without human intervention overnight.We start to see Claude making fundamental discoveries in math,working at the frontier of life science.This acceleration can be disorienting.But there's still this sense that we're just setting out.That exponential curve stretches up ahead of us, and it's only just begun.

主讲人

要实现 Anthropic 的使命就得死守前沿,死守前沿全靠算力。咱们的路子源于我和几位联合创始人早年做的缩放定律研究。我们发现,配方对了,算力越多,智能就越稳、越准。后来我们发现,这个配方能不断升级,用同样的硬件算力换来更高智能。所以到二零二一年办 Anthropic 时,咱们摸清了两件事。第一,算力是生意的底座;第二,把每份算力用到极致,是造出更聪明模型的关键。这也是咱们死磕效率的原因。咱们相信,在 Anthropic 花一块钱买的算力,比其他任何前沿实验室都管用。咱们靠两招拿到这个优势。先说第一招。咱们自研的芯片部署方案。现在咱们在四大主流芯片族上都跑 Claude。这之所以能成,是因为早年咱们押了一把多芯片策略。Achieving Anthropic's mission requires staying at the frontier, and staying at the frontier depends on compute. Our approach here goes back to research that I did with several of my other co-founders on scaling laws. We found that with the right recipe, more compute reliably and predictably yields more intelligence. Later, we found that it was possible to improve this recipe over time, to achieve more intelligence with the same amount of physical compute. So by the time we started Anthropic in 2021, we knew two things. One, compute is the foundation of the business, and two, getting the most out of every unit of compute is key to making increasingly intelligent models. That's why we're so focused on efficiency, and it's why we believe a dollar of compute at Anthropic goes further than a dollar at any other frontier lab. We've achieved that advantage in two ways. I'll start with the first. Our proprietary approach to bringing up and deploying silicon. We train and serve Claude now across all four major chip families. That's only possible because early on, we made an unconventional bet on a multi-chip strategy.

主讲人

当时很多人觉得在不同芯片上跑AI模型不值得费劲。每块芯片都得配专属的软件栈和推理栈,成本高又难搞。但我们坚持做了好几年,现在做得极其熟练。这带来两大优势。在算力紧缺的市场里,我们的多芯片策略让我们能只看性价比,不管供应商是谁。我们可以把每块芯片用在它最擅长的活儿上。有的适合训练,有的适合推理。这套策略让我们能把所有供应商、架构和地点的芯片,当成一个可互换的算力池。这也让我们上手新芯片特别快。比如今年我们第一次开始用AMD显卡。以前引入新芯片平台要几十个工程师干好几个月。这次我们让Mythos去试。Claude在AMD上跑起来只用了三天,几乎没人插手。At the time, many people thought that running AI models on different chips wasn't worth the effort.Each chip needs its own tailored software stack and inference stack, which is expensive and difficult to build.We've now been doing it for years, and we've gotten incredibly good at it.This comes with two key advantages.In a supply-constrained compute market, our multi-silicon strategy allows us to focus on the best price performance regardless of the provider.We can match each chip to the job that it's best at.Some are better at training, some are better at inference.Our multi-silicon chip strategy means we can treat all of our chips, across providers, architectures, and locations,as one fungible pool of compute.This has also made us really good at bringing up new chips.As an example, this year, we started working with AMD GPUs for the first time.In the past, bringing on a new chip platform took dozens of engineers months of work.This time, we had Mythos try it out.It got Claude running on AMD in just three days, with basically no human intervention.

主讲人

Claude移植或重写了近百分之百的内核。我们对存放芯片的数据中心也有独特做法。过去我们直接买云服务算力。现在我们加了第二招:租芯片,放在专门为我们建的数据中心里跑。这叫Anthropic Direct。这招对我们有俩好处。能降低算力的总拥有成本,还能把数据中心承诺和芯片本身分开。我们算力优势的第二个推手,是发现算力乘子的速度。我们看重总有效算力,就是物理容量乘以发现的算力乘子。每发现一个,总有效算力池就大一分。打个比方,假设训练某个智能水平的模型要一百单位算力。四倍的算力乘子意味着你只需二十五单位就能训出同样的模型,或者你用这一百物理单位,去训出个更聪明的模型,那模型以前得四百单位才行。Claude ported or wrote almost 100% of the kernels.We also have a distinct approach to the data centers that house our chips.Historically, we bought compute as a cloud service.We've now added a second approach where we lease chips and run them in purpose-built data centers designed for us.We call this Anthropic Direct.That approach has a couple advantages for us.It helps to lower the total cost of ownership of computeand allows us to separate the commitment to data centers from the chips.The second driver of our compute advantage is our rate of discovery of compute multipliers.We focus on total effective compute,which is the multiplicative product of physical capacity times the compute multipliers that we discover.Each one we discover makes our pool of total effective compute bigger.For example, say you need 100 units of compute to train a model of a certain intelligence.A 4x compute multiplier means you can use 25 units of compute to train the exact same model,or you can use those 100 physical units to train an even more intelligent modelthat would have previously required 400 units.

主讲人

我们相信,精准投资加前沿研究,已让Anthropic领跑顶尖实验室。Compute multipliers create a competitive advantage for us that's hard to replicate. They build on each other. Each new multiplier we discover has a positive effect on the multipliers before it. They stack. We believe that this can accelerate even further as we approach recursive self-improvement. Claude will be able to discover compute multipliers entirely by itself, allowing the number of compute multipliers to grow faster and faster. We believe this combination of smart investments, breakthrough research, and the improving capabilities of Claude have put Anthropic in a leading position among the frontier labs.

出镜者

在埃森哲(Accenture),我们的使命就是架起技术与实效的桥梁。AI is a breakthrough innovation, but only reinvention is going to get the value that AI promises. At Accenture, our job is to be the bridge between technology and real outcomes for our clients.

主讲人

我很喜欢这个问题,咱们到底用不用Claude?因为我们每天有数万名工程师在用。我们在整个软件开发生命周期里都用它。它帮我们省时间、提速度,让工程师能花更多精力解决客户难题、辅助决策和定战略。我们帮成千上万客户防网络攻击,还把Claude Code和Agent开发包跟cyber.ai结合了起来。这是我们保护客户的尖端平台。这么搭配的一个成果就是,以前要扫描好几天的活儿,现在不到一小时就能搞定。我们一心盯着客户的成功和安全,我们的想法是,因为咱们的合作,客户能更快拿到价值。I love the question about, you know, are we using Claude? Because we have tens of thousands of engineers using it every day. We use it across our entire software development lifecycle, and it frees up time, it makes us go faster, and it allows our engineers to spend more time focusing on solving our clients' challenges, leading decision-making, and setting strategy.We defend thousands of our clients from cyber attacks, and we've paired Claude Code and Claude's Agent Software Development Kit with cyber.ai, our cutting-edge platform that we use to help protect our clients. Just one result of using this is that we can now do scans that used to take days in under an hour. We are focused on our clients' success and safety, and our mindset is that because of our partnership, our clients can get to value faster.

主讲人

我们早期的押注是,智能是企业能从AI里拿到的最高价值资产。我们相信前沿智能能啃下最硬的骨头,所以它带来的影响也最大。AI能力不会慢慢爬升。模型一旦跨过某个门槛,就会发生跳跃式改变,一整类工作都会被彻底重写。编程是我们最熟悉的例子。变化快得惊人。短短一年,我们从自动补全几行代码,变成了全天候运转的代理大军。对乐天这种公司来说,新功能上线速度直接比对手快四到五倍。编程之后轮到网络安全。靠着Mythos和玻璃翼计划,Claude已经帮网安人员找出了漏洞。他们在全社会规模上完成了修补。咱们以Mozilla这家公司为例。The early bet we made was that intelligence was the highest value asset that a business could get from AI. We believe frontier intelligence solves the toughest problems, and for that reason, it drives the biggest impact. AI capability doesn't improve gradually. When models cross a certain threshold, there is a step change, and an entire category of work is done differently.Coding is the example we're most familiar with. The speed of the change is remarkable. We went from autocomplete, a few lines of code, to fleets of agents working around the clock in just a year. For a company like Rakuten, this means they're shipping new features four to five times faster than their competitors. After coding came cyber. And with Mythos and Project Glasswing, Claude has helped cyber defenders identify and patch vulnerabilities at a societal scale.Consider a company like Mozilla.

主讲人

每个领域都是先提升,再跃升,最后整个板块跑起来。In the past, they used to fix over 20 security bugs per release. And with Mythos Preview, they identified and fixed 271. Knowledge work is on a similar curve. Scientific research, investment analysis, underwriting, all great examples of what we're seeing across industries. Claude helped an insurer like AIG cut down time to quote by 55%. In each domain, we see an improvement, then a step change, and then a whole category of work moves fast.

出镜者

对客户来说,我们的响应也更快了。It's hard to imagine the speed with which that Anthropic has become embedded inside of AIG. With Claude, it's helped us create better underwriters. We're able to use their skillset more frequently because we're able to pull the submission information, get it in a format where they can then underwrite. We've seen a significant uptick in the speed with which they can make decisions, and which really means we're able to get to more submissions in a more consistent way. And for clients, we're able to respond faster.

主讲人

我觉得Anthropic和Claude绑在一起,提供的价值主张非常一致。我们在安全、韧性和技术能力上都全力做到位。我们也跟监管机构保持密切沟通,说明我们怎么用AI。人和人工监督显然对我们、对监管方都极其重要,而且已经深深嵌在流程里了。Anthropic做的每件事,从研发到产品再到推向市场,最终都体现在我们怎么服务客户上。现在,我们在五百强企业和前十强里都有深度合作。超过一半的五百强企业在我们这儿花的钱超过十万美金。全球两千家企业里也有三成多是这样。面对所有这些客户,我们相信我们才刚刚开启一场持续几十年的转型。平均来看,这些五百强企业客户花在我们身上的钱,还不到他们总开支的千分之一。I think Anthropic and Claude together have a very similar value proposition that we try and bring in terms of safety and resilience and technical capability. And we communicate pretty closely with the regulators on how we're using AI and how the human element and the human oversight is obviously very important to us and to our regulators and is embedded in the process. Everything Anthropic does, from research to product to go to market, it all comes together in how we serve the customer. Today, we have deep engagements across the Fortune 500 and all of the Fortune 10. More than 50% of the Fortune 500 spends more than 100K with us. And that's true for more than 30% of the Global 2000. With all of these customers, we believe we are still in the early stages of a multi-decade transformation. On average, spending from these Fortune 500 customers is less than 0.1% of their total expenses.

主讲人

就算在前十强里,占比也还是不到百分之一。咱们聚焦看看典型客户的旅程吧。它通常从一个单一场景开始。很多时候是开发者用的Claude Code,然后很快会扩散到周边岗位。知识工作者开始用Claude干最有价值的事。接着团队会把更多业务流程自动化,把AI搭进核心架构里。最后,他们把Claude塞进全新的产品和服务中。这个过程发生得飞快,所以我们看到了史无前例的约百分之九百净收入留存率。让我兴奋的是,最有趣、最能颠覆行业的用法还在后头。编程和网安里见过的所有东西,我们现在几乎在所有部门里都开始看到了。在公司和行业层面,同样如此。他们都想跟Anthropic合作,因为他们相信我们能帮他们抢先看到未来。Even among the top 10, it's still less than 1% of their total expense line. Let's zoom in on the journey of a typical customer, and it starts with a single use case. Often that's Claude Code for developers and engineers, and it soon expands to adjacent roles. Knowledge workers start using Claude for their highest value work. Then teams automate more business processes and build AI into core infrastructure. And ultimately, they embed Claude into entirely new products and services. The process happens quickly, and that's why we see an unprecedented net revenue retention rate of approximately 900%. What excites me is that the most interesting and transformative use cases are still ahead of us. Everything we've seen in coding and cyber, we're now starting to see across virtually every function, every company, and every industry. And they want to partner with Anthropic because they trust we will help them see the future first.

主讲人

我们会让他们始终站在技术前沿,并助力他们实现转型、赢得胜利。当Claude解锁的发现进入我们的世界,甚至超越它时,激动人心的事情就发生了。二零二五年十二月,Claude协助规划了毅力号火星车的路线。它用火星车专用的标记语言编写代码。工程师在指令发出前逐一核对。行驶四百米,任务圆满完成。二零二六年夏天,Claude处理海量数据的能力再次凸显。约九百五十个Claude智能体耗时二十一小时,筛选了十九亿个蛋白质簇。其中一个发现了异常模式,一种带有类似CRISPR重复序列的系统。我们的科学家在实验室里验证了Claude的发现。that we will keep them on the frontier and that we will help them transform and win. Something exciting happens when the discoveries unlocked by Claude move into our world, or even beyond it.In December 2025, Claude helped plan the Perseverance rover's route on Mars.It wrote the code in the rover's own markup language. Engineers checked everything before the commands went out. 400 meters driven, mission accomplished.Claude's ability to process vast amounts of data came to the fore again in the summer of 2026. About 950 Claude agents spent 21 hours sifting 1.9 billion protein clusters. One spotted an unusual pattern, a system with a CRISPR-like repeat array. Our scientists were able to check Claude's findings in the lab.

主讲人

这种新序列此前从未发表过,是一项真正的发现。在资源匮乏的地方,快速理解新信息的能力至关重要。二零二六年九月,应对刚果东部埃博拉疫情的团队,能在不到一小时内处理各地区的每日报告。这有助于更准确地掌握疫情态势,并为患者护理腾出时间。能够影响现实世界,让安全变得尤为重要。这也是为什么Claude的完整生物能力仅开放给经过我们审核的机构。以及为什么Claude拥有行为准则。它能帮助AI在从未遇到过的情境中做出判断。The new array had never been published before, a genuine discovery.The ability to quickly understand new information really matters where resources are thin.In September 2026, teams responding to an Ebola outbreak in eastern Congo were able to process each region's daily reports in under an hour,creating a better picture of the epidemic and freeing up time for patient care.The power to affect the real world makes safety even more important.It's why Claude's full biology capabilities go only to organizations we've verified.And why Claude has a constitution.It equips it to make judgment calls in situations it has never encountered before.

主讲人

这也是我们投入大量精力研究可解释性和对齐的原因,以便在人们依赖它的场景中保持Claude的责任可控。一次又一次,我们发现安全并没有拖慢我们的脚步。事实上,正因为我们可以信赖Claude的工作成果,我们才能将它的发现应用于现实世界,造福每一个人。我们相信,AI推动全球经济转型是历史上最大的经济机遇,而Anthropic具备独一无二的优势来把握它。我们在AI前沿占据领先地位,预计价值将在此集中涌现。强大的商业模式使我们能持续投入以推进前沿技术,同时为客户创造更多价值,为股东带来卓越回报。最有力的证明就是我们的财务业绩。第三季度,我们实现了近一百七十亿美元的收入,同比增长十二倍。年初至今,我们的算力利润率已达百分之四十三。And it's why we put so much energy into interpretability and alignment, to keep Claude accountable in situations where people will rely on it.Again and again, we're finding safety doesn't hold us back. In fact, it's because we can trust Claude's work that we can take its discoveries and apply them in the real worldin ways that will benefit everyone. We believe the transformation of the global economy through AI is the greatest economic opportunity in history, and Anthropic is uniquely positioned to capture it. We have leadership at the frontier of AI, where we expect value will differentially accrue, and the strength of our business model allows us to keep investing to advance the frontier while delivering more value to customers and superior returns for our shareholders. The clearest evidence of that is our financial results. In Q3, we earned nearly $17 billion of revenue, up 12 times year-over-year. And year-to-date, we've delivered a compute margin of 43%.

主讲人

毛利率百分之八十五,调整后营业利润率百分之十二,自由现金流为正。这些成绩源于我们在前沿的领先地位,也赋予了我们重金投入以巩固优势的能力。尽管盈利能力强劲,我们仍处于积极投资状态,专注于长期建设这项业务。Anthropic的商业模式有五大差异化要素。第一,我们构建能提供最具实用价值的智能模型。随着我们不断突破前沿,客户为Claude找到了更有价值的应用,收入增长随之而来。我们将此称为模型驱动的增长。业务轨迹印证了这一点。季度收入从二零二五年第一季度的三点七六亿美元,跃升至二零二六年第三季度的近十七亿美元。这与我们模型在编码、智能体工作流和现实任务方面能力的显著提升同步。立足前沿能解锁新用例和新能力。这也让我们向大多数客户工作负载的帕累托优势迈进。Gross margin of 85%, adjusted operating margin of 12%, and positive free cash flow.These results come from our frontier leadership, and they give us the capacity to invest heavily to extend it.So while our profitability is strong, we are in an aggressive investment mode as we focus on building this business for the long term.There are five elements that differentiate Anthropic's business model.First, we build models that deliver the most useful intelligence.As we've pushed the frontier, our customers have found more valuable work for Claude, and our revenue growth has followed.We call this model-led growth.The trajectory of our business demonstrates this.Quarterly revenue grew from $376 million in Q1 of 2025 to nearly $17 billion in Q3 of 2026.This tracked a significant increase in the capabilities of our models in coding, agentic workflows, and real-world tasks.Being at the frontier unlocks new use cases and capabilities.It also moves us towards Pareto dominance across most customer workloads.

主讲人

让我们得以在既定成本下交付最强模型。因为更聪明的模型效率更高。它们能以更低工作量完成任务,消耗更少代币。我们还将大模型的前沿智能蒸馏到小模型中。因此我们覆盖了成本、能力和速度的全谱系。每一代都在此基础上更进一步。我们近期发布的Claude 5.5系列,相比几个月前的Claude 5系列,能力大幅提升。按任务计费也便宜得多。这就是站在前沿构建的力量。我们相信能用最低成本提供最高价值智能,从而高效服务整个市场。尽管多数基准测试我们领先,但衡量成功的真正标准是客户。这就引出了我们的第二大差异化优势:我们打造领先平台和产品,让客户都能用上前沿智能。enabling us to deliver the most capable models at a given cost. That is because smarter models are more efficient. They can complete tasks at lower effort levels, which use fewer tokens. We also cascade the frontier intelligence from our larger models into our smaller models. So we address the full range of cost, capability, and speed. Each generation has pushed that further. Our recent Claude 5.5 family of models are a significant step up in capability from the Claude 5 family we released just a few months prior. They're also significantly cheaper on a cost per task basis. That is the power of building at the frontier. We believe we deliver the highest value intelligence at the lowest cost, which means we can serve the entire market efficiently. While we lead across most benchmarks, the true measure of our success is our customers. And that leads me to our second differentiator.We build the leading platform and products to make frontier intelligence accessible to our customers.

主讲人

我们的平台方案是业务基石。客户能基于我们的API和SDK开发产品、重塑流程,空间巨大。同时,自研产品也成为推动普及的关键。比如Claude Code,本质上是押注模型能力的指数级进步。让客户尽早用上它,让我们能紧跟智能体编程能力的飞跃来抢占先机。我们用同样打法推出Claude Cowork,其采用率在去年预览同期已超过前者。产品与平台结合,让我们能全面满足客户需求。如今大多数企业客户既用平台也用至少一款产品,比例较二零二五年四季度已大幅上升。Our platform solutions are the foundation of our business, and there is significant opportunity for customers to build products and reimagine processes on top of our APIs and SDK. At the same time, our own products have become a key driver of diffusion. For example, Claude Code was fundamentally a bet on the exponential and model progress. Getting it into the hands of our customers put us in a position to capitalize on agentic coding capabilities as they dramatically improved. We followed that same playbook with Claude Cowork, and its adoption has exceeded that of Claude Code at the equivalent point following its own research preview last year. Together, our products and platform enable us to serve the needs of our customers. Today, the majority of enterprise customers use both our platform and at least one of our products, up from 30% in the fourth quarter of 2025.

主讲人

随着内部广泛使用Claude,我们的产品发布也在提速。工程师日均代码贡献量较二零二四年平均水准翻了四十五倍,这直接体现在新成果上。二零二六年前三季度,我们上线了二百五十个模型、产品和功能。Claude不仅加速了工程团队,在全公司范围内,员工使用量在一月至九月间增长了二百五十多倍。这也解释了为何每位员工创收从二零二五年三季度的一百四十万美元,增至二零二六年三季度的六百六十万美元以上,增幅超四点五倍。Anthropic的第三大优势在于极致的算力利用效率。我们将算力视为统一资源池,兼顾客户服务、模型训练和内部应用。我们致力于让每一单位算力发挥最大价值。Our product velocity is also accelerating as we expand our internal use of Claude. The average daily volume of code contributed per engineer is up 45x relative to the 2024 average, and that's showing up in our launches. In the first nine months of 2026, we shipped 250 models, products, and features. And Claude isn't just accelerating our engineering teams. Across the company, our employees' use of Claude grew more than 250 times between January and September of 2026. Together, these help explain why our revenue per employee has grown more than 4.5x, from $1.4 million in Q3 of 2025 to over $6.6 million as of Q3 2026. Anthropic's third differentiator is that we are extremely efficient users of compute. We treat our compute as one fungible pool that serves customers, trains our models, and powers our own use of Claude. We are focused on getting the most out of every unit of compute.

主讲人

这为我们带来了服务成本与再投资的双重优势。正如Tom所说,这种效率源于两点。第一是多芯片策略,让我们能采购性价比最高的芯片,并整合为单一资源池运行。第二是算力乘数效应,使有效算力规模远超物理硬件增速。财务数据印证了这一点。过去六个季度,营收增长超四十五倍,而算力与基建成本仅增十倍,每美元算力带来的收入提升超四倍。年初至今,毛利率已升至百分之八十五以上,对比二零二五年同期大幅增长。我们将这些效率红利视为可调配资金。有时我们会降价让利,推动Claude普及。有时则保留利润,重新投入购买更多算力。This gives us a cost to serve and a reinvestment advantage. As Tom described, there are two drivers of this efficiency. The first is our multi-silicon strategy, which lets us buy the most price-performant chips and run them as a single fungible pool. The second is compounding compute multipliers, which have scaled our total effective compute significantly faster than our physical capacity. You can see this in our financial results. Over the last six quarters, we grew revenue by more than 45x, while compute and infrastructure costs grew only 10x, a more than 4x improvement in revenue per dollar of compute. And year-to-date, our gross margin has improved to over 85%, up from 29% in the same period of 2025. We think of these efficiency gains as capital that we get to allocate. In some cases, we pass them along to customers by lowering prices, which drives diffusion of Claude across those customers. In other cases, we retain the margin and reinvest it in additional compute.

主讲人

这正是让优势不断放大的原因。下一个差异化优势在于我们专注打造企业级产品。我们早就意识到,企业会是最大且最好的AI客户。自二零二三年一季度商业化以来,我们一直深耕该市场,采用率迅速攀升。截至二零二六年三季度,企业客户已超一万一千九百家,较年初增长近八倍。作为首家登陆三大主流云平台的顶尖实验室,我们让客户能在现有环境中部署Claude。同时我们保持与终端客户的直接联系,并掌握定价权。我们相信,这些合作让我们的推广效率远超同行。事实上,我们全部的非算力支出,包含销售、人力和其他运营费用。它们占营收的比例,甚至低于同行软件公司仅用于销售的占比。which is what makes this advantage compound. The next differentiator is our relentless focus on building for enterprises. We had the early insight that enterprises would be the largest and best customers for AI. We have focused on serving that market since our commercial launch in the first quarter of 2023 and have seen rapid adoption since. As of Q3 2026, we had more than 11,900 enterprise customers, up nearly 8x from the start of the year. As the first frontier lab available on all three major clouds, we allow enterprises to deploy Claude within the environments they already operate in, and we maintain the direct relationship with the end customer, including control over pricing. We believe these partnerships enable us to be far more efficient in our go-to-market efforts than the rest of the industry. In fact, our entire non-compute expense base, including sales and marketing, personnel, and other operating expenses, represents a lower percentage of revenue than the median enterprise software company spends on sales and marketing alone.

主讲人

我们预计将在市场推广上投入重金以支持增长。同时在销售与客户拓展中保持高度高效。我们的最后一个差异化优势在于不断扩大的市场机遇。我们相信,强大的人工智能将重塑各行各业的所有企业。变革从编程开始,正蔓延至知识工作领域。这是一个四十万亿美元的市场,而我们才刚刚起步渗透。我们将许多全球规模最大、业务最复杂的企业视为客户。其中包括超过五成的财富五百强企业和三成的全球两千强企业。正如保罗所述,我们的财富五百强客户在Claude上的支出,还不到其总费用的百分之零点一。随着人工智能前沿技术的推进,我们预期它将开辟全新的业务线。网络安全领域已经率先开启。我们预计未来几年会延伸至生命科学和机器人领域。鉴于我们看到的市场规模与我们业务的强劲势头,We expect to invest meaningfully in go-to-market to support our growth, while remaining highly efficient in how we sell to and grow with our customers.Our final differentiator is our expanding market opportunity.We believe every company across every industry will be transformed by powerful AI.It started with coding and is spreading to knowledge work, a $40 trillion market that we've just begun to penetrate.We count many of the world's largest and most complex businesses as our customers,including over 50% of the Fortune 500 and over 30% of the Global 2000.And as Paul described, our Fortune 500 customers spend less than 0.1% of their total expenses on Claude.As the frontier of AI advances, we expect it will open up entirely new lines of business.This has started already with cybersecurity, and we expect it will extend to life sciences and robotics in the coming years.Given the size of the opportunity we see and the strength of our business,

主讲人

我们正在大力投资三大领域:算力、客户价值,以及安全与对齐。算力是我们增长的关键引擎。它支撑着我们的研发,并让我们能满足不断扩大的客户需求。鉴于可观的投资回报,我们计划在未来几年持续大幅投入算力。这包括短期和多年的云服务合约,以及我们直接租赁的数据中心和芯片。我们将最后这部分称为Anthropic Direct。预计它在未来几年将占据我们算力基数的更大比重。我们的第二大投资领域是客户价值。我们极度专注地确保客户在Claude上的每一分钱都能获得最强回报。我们通过开发能力更强、更安全且更实用的模型来实现这一点。同时随着模型效率提升来降低客户成本。例如,今年九月发布的Opus 5.5模型不仅大幅提升能力,还为客户降低了百分之四十的成本。这项降本对我们自身的利润率没有影响。We are investing aggressively in three areas: compute, customer value, and safety and alignment. Compute is a critical enabler of our growth. It powers our R&D and enables us to meet expanding customer demand. We plan to make significant investments in compute over the coming years, given the compelling returns we see. That includes both short-term and multi-year cloud deals, as well as data centers and chips we lease directly. We call this last piece Anthropic Direct, and we expect it will make up a growing share of our compute base over the coming years. Our second investment area is customer value. We are maniacally focused on ensuring customers get the strongest return from their spend on Claude. We do this through developing the most capable, aligned, and useful models, and through reducing customer costs as we make our models more efficient. For example, our Opus 5.5 model in September delivered substantial gains in model capability and 40% lower customer costs. This reduction was margin neutral to us.

主讲人

因为 Opus 5.5 的推理成本大幅降低。虽然降价会牺牲部分短期收入,但我们的经验是,智能成本下降很快会带来收入顺风车,推动 Claude 的使用量和普及度大幅提升。Because Opus 5-5 is significantly cheaper to serve, while we sacrifice some near-term revenue when we reduce customer costs, in our experience, we quickly see a tailwind to revenue as the lower cost of intelligence drives greater usage and diffusion of Claude. The third area is safety and alignment. Safety is core to how we operate at Anthropic, and we've been investing here since the inception of the company. As model capabilities advance rapidly, we're investing even more in people, processes, and compute to support our safety and alignment work. Taken together, we expect our focus on safety to remain a commercial advantage because customers and developers want models and products they can trust. The investments we are making in compute, customer value, and safety and alignment will strengthen our business and broaden our opportunity. Over the long term, we believe Anthropic can deliver adjusted operating margins of more than 35%. With continued efficiency gains, a greater share of Anthropic direct capacity,

主讲人

随着更多算力转向推理,我们预计算力和基础设施支出将占营收的 45% 到 50% 左右。其次,销售、营销和合作成本将降至营收的 7% 到 10%,因为规模扩大后合作费用占比会缩小。第三,人力和其他运营费用合计也将控制在营收的 7% 到 10%,因为 Claude 提升了团队效率,且我们扩招很高效。我们相信公司拥有最优的财务模型和领先的算力效率。这里汇聚了全球最顶尖的人才,文化也让我们始终聚焦于如何为企业创造最大价值。人工智能重塑全球经济,有望成为史上最大的经济机遇,而今天我们就站在这一浪潮的起点。比如今天,如果我们看看癌症、阿尔茨海默病、心脏病和精神疾病这些问题,and more of our compute shifting towards inference, we see a path for compute and infrastructure expenses to trend towards approximately 45 to 50% of revenue.Second, we expect sales, marketing, and partnership costs to trend towards 7 to 10% of revenue, as partner fees become a smaller share of revenue as we scale.And third, we expect personnel and other operating expenses combined to trend towards 7 to 10% of revenue, as Claude makes our teams more productive and we grow headcount efficiently.We believe we have the best financial model, leading compute efficiency, the greatest concentration of talent, and a culture that keeps us focused on where we can add the most value to enterprises.AI's transformation of the global economy has the potential to be the largest economic opportunity in history, and we are at the foothills of it today. You know, if today we look at things like cancer, Alzheimer's, heart disease, mental illness,

主讲人

我认为不到十年,这些疾病就会成为历史遗迹。我相信,能妥善管控风险的公司,才能拿到继续运营的许可。社会会信任它,并相信它能以正确的方式造福人类。如果我们要把这件事做好,就能开创一个奇迹时代,让人类解决困扰我们数千年的难题。我们必须极其谨慎地应对所有潜在风险,防止技术出错。同时,我们还要在创造福祉和维持健康商业运营之间取得平衡。These will be relics of the past in, I think, less than a decade.And I believe that the company which gets the risks rightwill also be the company that has the license to operate,that society trusts and society believes can build the benefits in the right way.I think if we do this right, we can create an age of miracleswhere humanity is able to solve problems that have eluded us for millennia.We have to make sure that all the risks,all the ways that things can go wrong with this technology,we very carefully address those risks as we bring the benefitsand as we have a viable business at the same time.

主讲人

我诚挚邀请大家加入我们的下一段旅程。And I invite you to join us on the next phase of this journey.

已复制