Cerebras CEO 谈快速 AI 推理与计算未来

Netflix 以前用信封寄送 DVD，当互联网变快后，它变成了电影制片厂，对吧？这开启了一个全新的业务，一些根本不同的东西。这就是速度带来的变化，我认为快速 AI 也是如此。现在，我们正在取代大家都能看到的东西：编码、设计、SaaS 工具，但一旦我们开始围绕这一点进行根本性的重组，你就会看到这种新的商业模式和生产力上的根本性飞跃，我对此充满期待。太酷了。今天在 No Priors 节目中，我们有 Andrew Feldman，Cerebras 的联合创始人兼 CEO。Cerebras 成立于 2010 年代中期，专注于 AI 的新工作负载，特别是机器学习领域，然后转型为我们今天所处的基础模型世界中的极速推理。Cerebras 最近上市，目前在股票市场上市值约 630 亿美元。Andrew，感谢你加入 No Priors。

Netflix used to deliver DVDs in envelopes, and when the internet got fast, they became a movie studio, right? It opened up an entirely new business, something fundamentally different. That's what happens with speed, and I think that's what fast AI does. Right now, we're replacing things that everybody can see, coding, design, the SAS tools, but once we start sort of fundamentally reorganizing around this, you're going to see this sort of new business models and fundamental jumps in productivity, and I'm eager for that. That's so cool. Today in No Priors, we have Andrew Feldman, the co-founder and CEO of Cerebras. Cerebras was founded in the mid-2010s to focus on new workloads for AI, particularly the machine learning world, and then has made the transition into very fast inference for the foundation model world that we live in today. Cerebras recently went public and is currently worth about $63 billion on the stock market. So, Andrew, thank you for joining us in No Priors.

哦，非常荣幸。很高兴再次见到你们。

Oh, what a pleasure. It's good to see you guys again.

是的，首先恭喜你。你的公司 Cerebras 刚刚上市。截至今天，市值 600 亿美元，这相当惊人。

Yeah, so first of all, congratulations. So, your company Cerebras just went public. As of today, it's a $60 billion market cap, which is pretty amazing.

确实惊人。

Pretty amazing.

是的，我记得一两年你前上过我们节目，在早期的一集中，当时和你交谈很愉快，显然今天我们也很兴奋能请你来。你能跟我们说说从那以后业务是如何发展的吗？另外，给我们的观众提醒一下你们是做什么的，专注于什么，以及如何前进的。

Yeah, and I think you were with us a year or two ago on the show in one of the earlier episodes, and it was a pleasure to talk to you then, and obviously we're very excited to have you on today. Could you tell us a bit how the business evolved since that time and what you folks just a reminder for our audience what you do, what you're focused on, how you're moving forward.

我们制造 AI 计算机，对吧？专为加速 AI 工作负载而设计和优化的计算机。目前，我们在推理方面是最快的，不是快一点点，而是快很多，比 GPU 快 15、18、20 倍。所以，从 2025 年左右开始，AI 模型变得足够智能，开始有用。人们开始使用它们，你知道，我们通过训练制造 AI，并通过推理使用它。因此，随着人们开始使用它，它开始融入他们的日常工作。速度变得至关重要，我们被需求压得喘不过气来。

We build AI computers, right? Computers designed and optimized to accelerate AI workloads. And right now, we're the fastest at inference, not by a little, but by a lot, 15, 18, 20x faster than GPUs. And so, what happened was starting in about 2025, AI models got smart enough to be useful. People began using them, and you know, we make AI with training, and we use it with inference. So, as people began to use it, it began to be integrated into their day-to-day work. Speed became fundamentally important, and we were just crushed with demand.

是全面更快，还是特定用例？

Is it faster across the board, or is it specific use cases?

全面更快。大模型、小模型、美国模型、中国模型、万亿参数模型或 10 亿参数模型，全面覆盖。

Faster across the board. Big model, small models, US models, Chinese models, trillion parameter models or 1 billion parameter models, across the board.

嗯。

Mhm.

然后年底，我们与 OpenAI 签署了一项协议。这算是硅谷有史以来最大的交易之一，金额超过 200 亿美元。接着在三月，我们与 AWS 签署了协议，未来将部署在他们的数据中心。所以，这就像旋风般的一年半，追逐供应并努力满足需求。

And then what happened was at the end of the year, we signed a deal with OpenAI. It's sort of one of the biggest deals ever in Silicon Valley. It's sort of north of 20 billion dollars. And then in March, we signed an agreement with AWS, where we'll be deployed in their data centers going forward. And so, it was just a whirlwind year and a half of chasing supply and trying to meet the demand.

在过去一年半里，你们做了什么？是制造产能的提升？还是新的芯片设计？还是别的什么？你能帮大家解释一下发生了什么吗？

And what you have done in the last year and a half, was it the ramp in manufacturing? Was it a new chip design? Was it something else? Could you help educate folks on what happened?

事情是这样的：我们制造了一台非常非常快的机器，但很长一段时间没人关心。他们……实际上，恕我直言，很多人反对，说这只是一个奇怪的架构。他们称之为错误。比如 Cerebras 被说成是错的。

What happened was we built a really really fast machine, and for a long time nobody cared. And they right it and that's because AI... Actually, forgive me for saying so, but a lot of people objected and said this is just a weird architecture. They called it wrong. Like Cerebras called it wrong.

是的。

Yeah.

他们确实这么做了。我认为要变得根本性地更好，你不能构建一个相似的架构，对吧？你不可能通过对 GPU 架构进行微小修改就获得 15 或 20 倍的提升。这很可能普遍成立。如果你渴望根本性的改进，你的设计必须不同。从一开始，我们就选择了晶圆级，这意味着我们制造一块 46,000 平方毫米的芯片，一块餐盘大小的芯片，而其他人都在制造邮票大小的芯片。他们告诉我们我们疯了，这永远行不通。他们列出了为什么不可能的理由。但在 2019 年，我们证明了这是可能的。我们开始交付，并不断改进。但当我们快的时候，它还是个新奇事物。当它是新奇事物时，没人关心你是否快，因为它没有被使用。所以，从大约 2023 年到 2025 年初，人们指着 AI，但没有人每天在工作中使用它。

They did. I think to be radically better, you can't build something that is a similar architecture, right? You're not going to get 15 or 20 times better than the GPU with a minor modification to their architecture. And that's probably true across the board. If you're going to aspire to a radical improvement, your design has to be different. And from the beginning, we chose wafer scale, which means we build a 46,000 square millimeter chip, a chip the size of a dinner plate, whereas everybody else is building chips the size of postage stamps. They told us we were out of our minds, it would never work. They listed reasons why it was impossible. But in 2019, we proved it was possible. We began delivering it, and we improved on it, and we improved on it. But we were fast when it was a novelty. And when it's a novelty, nobody cares if you're fast because it's not being used. And so, from about 2023 to the beginning of '25, people pointed at AI, but nobody used it every day in their work.

嗯。

Mhm.

一旦你每天在工作中使用某样东西，它就不能慢。我的意思是，你们会等一个网站加载多久？

And once you use something every day in your work, it can't be slow. I mean, how long will you guys wait for a website to resolve?

我完全没有耐心。

I'll have no attention.

对，完全正确。就是这样。我的意思是，慢速搜索的市场有多大？是零。拨号上网的市场有多大？是零。慢速推理的市场也将如此。但我们必须等到它足够智能、有用。这发生在 2025 年。这就是为什么需求爆发，像 Cognition、Cursor、Lovable 以及其他公司开始飞速增长。你们投资的许多公司也在疯狂增长。OpenAI 等等。而我们正好拥有合适的产品。

Right. That's exactly right. That's exactly the way it is. I mean, how big is the market for slow search? It's zero. How big is the market for dial-up internet? It's zero. That's how big the market for slow inference will be. But we had to wait until it was smart enough to be useful. And that happened in 2025. And that's why you got this sort of explosion of demand and companies like Cognition and Cursor and Lovable and just all these others that began ramping extraordinary. Many of the ones you guys have invested in are ramping like crazy. OpenAI and others. And we were right there with the right product.

嗯。

Mhm.

我想我第一次见你是在 2016 年左右。那时，人们并没有说 AI 听起来很奇怪，对吧？你在谈论机器学习。当时的模型是卷积神经网络和 RNN，你知道，还有 GAN 等东西刚刚出现。

I think I first met you back in 2016 or something like that. And at the time, people weren't like saying AI sounded weird. Right? You were talking about machine learning. And the models of the time were convolutional neural networks and RNNs and you know, just the emergence of GANs and things like that.

我们当时试图区分椅子和猫，对吧？那很棒。所以，他的博士论文就像猫和椅子。就像，“哇，看看我们走了多远。”我的意思是，这难以置信。

We were trying to tell the difference between a chair and a cat. Right? That was great. So, his PhD is like a cat and or a chair. It's like, "Woah, look how far we've come." I mean, it's unbelievable.

是的。是的。你认为是什么给了你逆市场而行的远见？因为正如你所说，我想我们很多人都相信这个市场会非常重要。而你比其他人更甚，对吧？因为你真的在这个领域创办了一家公司。但市场花了一些时间才真正扩展到现在的程度，正如你所说，现在这是一个巨大的用例，人们真正关心推理速度等。当时是什么给了你这样做的信念？

Yeah. Yeah. What do you think gave you the foresight to build against the market? Because to your point, I think a lot of us believed in that this market would be really important. And you more than others, right? Since you actually started a company in it. But then it took some time for the market to really expand to the point where to your point now it's this massive use case, people really care about speed of inference and other things. What gave you the conviction back then to do this?

远见、合适的联合创始人、一点点傲慢和一点点运气的结合。你知道，我们看到了 AI 在地平线上作为一种新的工作负载。作为计算机架构师，新的工作负载就是机会。好吧，进入 x86 世界非常非常困难，对吧？那里没有什么新东西发生，而且几代人都没有变化。

Combination of vision, the right co-founders, and a little bit of arrogance, and a little bit of luck. You know, we saw AI on the horizon as a new workload. And as computer architects, new workloads are opportunity. All right, it's very very hard to enter in the x86 world, right? Where there's not nothing new is happening there, and nothing has happened for generations.

但你知道，当图形计算兴起时，出现了独立 GPU，有了英伟达；当移动计算爆发时，有了 ARM。有趣的是，英特尔、AMD 以及所有你认为本应占据优势的公司，都没能分到一杯羹。所以我们知道，这种新工作负载会消耗大量算力，需要全新的、专用的架构，而且必须非常不同。这个架构不能是现有产品的衍生。这些都是我们的豪赌，而且 100% 逆向而行。

But you know, when graphics emerged, you got the discrete GPU, and you got Nvidia, and when mobile compute hit, you got ARM. And it was interesting that not Intel, not AMD, not all sorts of people who you would have thought have been really well positioned to win in that business, they all got no share. And so we knew that this new workload would eat a lot of compute. It would require a new architecture, dedicated architecture, and that ought to be very different. The architecture could not be a derivative of what's existing. Those were our big bets, and they were 100% contrarian.

嗯。

Mhm.

结果证明它们完全正确。

And they turned out to be dead right.

有没有那么一刻，你怀疑过这是否能行得通，毕竟这花了你不少时间？

Were there moments where you just doubted whether this would work, given that it took time for you?

有的。我们曾有一段时间——我们在解决一个从未被解决的问题。

Yeah. We had a period — we're solving a problem that had never been solved before.

嗯。

Mhm.

我的意思是，在计算机行业 70 年的历史中，一直有人试图构建一个可规模化的产品。事实上，Gene Amdahl，我们领域之父之一，算力界的拉什莫尔山人物，也惨败了。

I mean there'd been efforts across the entire 70-year history of the computer industry to build a way-for-scale product. In fact, Gene Amdahl, sort of one of the fathers of our field, one of the guys on Mount Rushmore of compute, failed miserably to do it.

嗯。

Mhm.

从 2017 年中到 2019 年中，我们一直造不出来。我们每月花费约 800 万美元。

We had a period between about 2017, middle of 2017 and middle of 2019, where we couldn't build it. We were spending about 8 million a month.

嗯。

Mhm.

每六周开一次董事会，我们说：‘我造不出来。不，还是不行。’没错，‘唉’是正常的。我是说，那是巨额资金，也是投资者巨大的信念。每次失败分析后，我们都进步一点。然后在 2019 年夏天，我们成功了。它开始工作了。第一次，我们坐在洛斯阿尔托斯市中心一个临时搭建的办公室里，那栋楼不是为硬件人员设计的。我们盯着电脑，那感觉就像看油漆干一样无聊。

You have a board meeting every 6 weeks saying, 'I can't build it. No, still not working.' And right, 'Oof' is right. I mean, that's a huge amount of money. And a huge amount of conviction your investors have. And each time we did a failure analysis, we got a little bit better at it. And then in the summer of '19, we yielded it. And it began to work. And the first time we were sitting in a little makeshift shift office in downtown Los Altos in a building that was not designed for hardware guys. And we're staring at a computer, which is about as exciting as watching paint dry.

嗯。

Mhm.

它居然在运行。我们半小时说不出话来。对吧？就像：‘没人能做到，但它成功了。我们做到了。’

And it's working. And we just couldn't speak for half an hour. Right? It's like, 'Nobody's been able to do this and it's working. And we did this.'

这太棒了，因为这是技术层面。然后还有市场层面，对吧？在市场层面，如你所说，这些工作负载变得真正重要需要时间。那么，有没有你怀疑市场是否存在的时刻？

And that's amazing because that's the technical side of it. And then there's a market side, right? And also on the market side, to your point, it took time to get to the point where these workloads were really important. So, were there moments where you doubted whether the market existed?

你知道，我们解决了计算机行业最难的问题，但没人在乎。没人在乎。第一代可能只卖了十几台，第二代大概卖了 300 台，现在第三代要卖几万台。我们有 2 到 3 年时间领先于市场，但根本没人关心我们快得惊人。

You know, we solved it — we solved this sort of the hardest problem in the computer industry and nobody cared. Nobody. It was like, the first gen we might have sold a dozen. The second gen we probably sold 300. And now we're going to sell tens of thousands in the third gen. We had a 2 or 3 year period where we were ahead of the market. And absolutely nobody cared that we were blisteringly fast.

你找到了一些先驱客户，他们的起点很不典型，对吧？有些主权实体提前购买。你是如何考虑在需求超前这段时期保持韧性的？

And you found some pioneering customers that were like atypical in terms of starting point, right? There were some sovereigns who really bought ahead. Like, how did you think about being resilient to this period of being ahead of demand?

我认为有一条新计算机架构铺就的道路。通常你从超级计算机领域开始，因为那些人热爱速度，不在乎你的软件是否成熟。

I think there's a path that has been laid down by new computer architectures. And often you begin in the supercomputer world because those guys love speed and they don't care if your software is immature.

嗯。

Mhm.

所以我们在那里大获全胜。我们拿下了国家实验室，劳伦斯利弗莫尔、桑迪亚，还有欧洲的欧洲并行计算中心 LRZ。然后我们又赢得了油气领域和制药领域的客户，他们都长期使用大量算力。但历史上存在一个巨大的鸿沟，因为没有一个客户能提供足够的量来进入主流。我们赢得了一个主权实体 G42。他们成为了战略伙伴和亲密朋友，给了我们一个 10 亿美元的订单。凭借这个，我们得以转型公司，改变供应链，部署足够大的集群来大规模实战测试。你知道，硬件的一个挑战是你的 QA 实验室不可能像某些客户部署的规模那么大。

And so, we sort of ran the table there. We wanted National Labs and at Lawrence Livermore and at Sandia and in Europe at European Parallel Computing Centre at LRZ. So, we ran the table there and then we won some guys in the oil and gas space and we won some guys in pharma, all of whom have long histories of using extraordinary amounts of compute. But then historically there's this giant chasm because none of them provide the volume to get to mainstream. And we won out a sovereign G42. And they became a strategic partner and close friends and they placed a billion-dollar order on us. And with that we were able to sort of transform the company. We're able to change our supply chain. We're able to deploy equipment in big enough clusters that we could battle test at scale. You know, one of the challenges in hardware is your QA lab can't be as big as some of the customers you want to deploy to.

嗯。

Mhm.

对吧？你不能在 QA 实验室里放价值一亿美元的自有设备。他们与我们合作，我们开始为他们训练模型，开始为他们做推理。他们是卓越的合作伙伴。这是 G42 的 CEO Peng 和他的主席 Sheikh Tahnoun。我们找不到更好的合作伙伴了。所以当 OpenAI 出现时，当 AWS 出现时，我们有能力，我们准备好了。对吧？我们经过了实战测试，跨越了鸿沟。我们有了一座桥，所以能够满足需求。

Right? I mean, you can't put a hundred million dollars in your QA lab worth of your own gear. And they worked with us and we began training models for them. We began doing inference for them. They've been an extraordinary partner. This is Peng who's CEO of G42 and his chairman Sheikh Tahnoun. We couldn't ask for better partners. And so we were able to when OpenAI came along, when AWS came along, we had the capacity. We were ready. Right? We'd battle tested. We'd sort of gotten over the chasm. We'd had a bridge and so we could meet the demand.

是的，我认为这种路径依赖在这个领域有时被低估了，因为从数千万、数亿美元的订单到 200 亿美元的积压，中间肯定有东西。那是多年的工作。

Yeah, I think that kind of path dependence is sometimes undervalued in this field because the ability for you to go from a tens, hundred million-dollar order to twenty billion of backlog, like there's got to be something in the middle. It's years of work.

这是多年的工作，而且你知道，我经常想，我相信你的许多听众在软件领域，你们可以扩展得非常快。

It's years of work and you know, I think often and I'm sure many of your listeners are in the software world and you guys can scale so fast.

嗯。

Mhm.

对吧？但当你制造东西时，你想翻倍，就得打电话给你的制造合作伙伴、你的代工厂。他们得找电力、租厂房、增加产线、制作测试夹具。对吧？每一步都需要时间和努力来增长。我们今年打算将制造能力提升 10 倍。

Right? But when you're building things, you have to — you want to double, you got to call your manufacturing partner, your CM. They have to find power. They have to rent a building. They have to add more lines. They have to make test fixtures. Right? Each step takes real time and effort to grow. We're going to try and increase manufacturing 10x this year.

嗯。

Mhm.

对吧？这差不多是硬件历史上最快的速度了。

Right? That's about as fast as anybody in the history of hardware.

所以对你们来说，软件栈的成熟度更关乎规模，对吧？

And so the maturity of the software stack for you guys that's more scale, right?

你知道，我们创办公司时，Sarah，我的一位联合创始人，Gary 我认识。我们向你展示过。一位联合创始人说 Andrew，构建一个编译器大约需要 10 年。我说：‘不，那太疯狂了。那是大公司的说法。我们 5 年就能搞定。’结果花了大约 10 年。

You know, when we started the company, Sarah, one of my co-founders, Gary I know. We presented to you. One of my co-founders said Andrew, it's going to take about 10 years to build a compiler. I said, 'No, that's crazy. That's big company talk. We can do it in five.' Takes about 10 years.

非常真实。

Very true.

构建编译器需要很长时间。这是一个极其困难的软件。现在我们有了一套不错的软件栈。

It takes a long time to build a compiler. It's an extraordinarily difficult piece of software. And now we've got a good software stack.

顺便问一下，因为你十多年来一直相信这场革命会发生，所有这些 AI 生成的代码对 Cerebras 内部有多大的相关性？

Can I ask you as an aside actually just because you have for more than a decade believed that this revolution's going to happen, how much is all of this AI-generated coding relevant for Cerebras internally?

非常大。我会说，8 个月前我们每个工程师在 token 上花费不到 1000 美元，现在大概是 25000 到 30000 美元，而且还在飞速增长。

Hugely. I would say that 8 months ago we weren't spending a thousand dollars an engineer on tokens, and we're probably at 25 or 30,000 right now, and it's ripping.

我认为它并非对所有人都适用。我觉得这是事实。有些人天生就适合这种模式，对吧？他们 7x24 小时运行着 8 到 10 个智能体，已经将编码风格转变为管理智能体。

I think it's not useful for everybody. I think that's truth. I think there are some people who have sort of the perfect mindset for it. Right? And they are running eight or 10 agents 7 by 24. They've moved their coding style to being one in which they govern agents.

嗯。

Mhm.

他们会考虑如何做质量保障，所以运行着一个 QA 智能体。他们会思考如何弥补编码模型的一些弱点，对吧？模型往往过于冗长，还经常删掉注释。所以他们真的深入思考了，这就像一种谜题，完美契合他们的思维方式。他们从 10 倍效率的人变成了 100 倍效率的人。我认为我们其他人，包括我自己，都在艰难前行。我们试图弄清楚如何让它适用于我们不同的工作——CEO、CFO、会计、市场营销。但对少数人来说，它确实是个好工具。至于其他人，我们努力向他们展示别人在做什么、最佳实践是什么。

Whether they think about how to QA, so they've got a QA agent running. They think about how to sort of remedy some of the weaknesses in the coding models, right? They're often verbose. They often cut out comments. So, they've really thought about it and it's a type of puzzle that is the perfect fit for their mind. And they've gone from being sort of 10x guys to being 100x guys. I think the rest of us, myself included, we're sort of limping along. We're trying to figure out how we can make it work for our different jobs. For being the CEO, for being the CFO, for being accountants, for being in marketing. But for a small number it is such a tool. And then the rest we try and try and show them what others are doing, what best practices are.

你们现在大约有 800 人？

You're about 800 people now?

800 到 850 人，是的。

800, 850, yeah.

人均市值很高。

It's a lot of market cap per person.

我喜欢这个说法，是的，不错。

I like that, yeah. That's good.

总体来看是个好指标。当你思考未来方向——扩大业务、战略布局——你预测什么？你们能走向何方？

A good metric overall. When you think about where to go from here, making business bigger, strategic directions, what do you predict? Where can you go from here?

我认为我们……

I think we...

除了交付。

Besides delivery.

嗯，当你有超过 200 亿的积压订单时，交付每天都非常重要。我认为我们必须继续无所畏惧。我觉得公司发展到 1000、2000、3000 人时的一个通病是，他们不再像以前那样冒险了，对吧？你从无畏的工程文化变成了“下个版本我们能做什么”。我认为这极具破坏性。我们为从事无畏的工作而自豪。我们希望雇佣那些做无畏工作的人。我们要守护那种文化：宁愿在追求非凡中失败，也不愿在平庸中成功。平庸是可怕的。所以这些是我担心的事情。还有招聘。对吧？你有那么多空缺，很容易将就。很容易随便找个人填坑。嗯，差不多就行，先让人坐进来。但那就是死亡。所以我们深思熟虑，我每天花大量时间与候选人交谈。这些都是我每天担忧和思考的事情。

Well, when you've got a backlog that's north of 20 billion, delivery is pretty important every day. I think we have to continue to sort of be fearless. I think one of the malaise of companies as they get to 1,000 to 2,000, 3,000 people is they stop taking the type of risks that they were taking before, right? You move from being a fearless engineering culture to sort of being what can we get in the time frame in the next rev. And I think that's extraordinarily damaging. And we take such pride in doing fearless work. We want to hire people who do fearless work. We want to sort of guard that culture that says we would much rather fail in pursuit of the extraordinary than succeed in the ordinary. That is a horrible thing to do. And so those are some of the things that worry me. I think recruiting. Right? You have so many openings and it is so easy to settle. And it's so easy to just try and put a butt in a seat. Yeah, pretty good. Let's get that butt in a seat. I mean that is death. And so we think really hard and I spend a meaningful part of every day talking to candidates. Those are things that I worry about, I think about every day.

我们有很多创始人和领导者收听播客，他们可能已经拥有成功的企业，正在等待市场或判断自己是否仍然正确。他们思考如何从 800 人招聘到几千人。我们谈到了管理自己的心态——比如“我适合这个十年吗？”当长时间没有外部反馈时，你如何留住和激励员工？

We have a lot of founders and leaders who listen to the podcast, who are thinking about maybe they have a successful business and they're managing through the period of waiting for the market or trying to figure out if they're still right. They think about how to hire from 800 to several thousand. We talked about the managing of your own psychology when you're like, am I right for this decade? How do you keep and motivate employees when there wasn't external feedback for this long period of time?

首先，我同情他们。CEO 是一个非常孤独的职位。你在建设一家企业。你们都知道，作为领导者是孤独的，而且不容易。人们不喜欢承认这一点。尤其是对于我们这些喜欢解决问题——特别是那些别人说无法解决的问题——的人来说。你从那种不服输的心态中获得动力，对吧？当别人说解决不了时，你心里想：“你解决不了。”

Well first I have empathy for them. I mean being CEO is an extraordinarily lonely thing. And you're building a business. You guys know this, that being a leader it is lonely. And it's not easy. And people don't like to say that. Especially for those of us who like to solve problems, specifically the problems everyone else says can't be solved. You sort of gain fire from that chip on your shoulder, right? When they say it can't be solved, you say in your head, you can't solve it.

对，没错，那就是……

Right. Right, that's...

我不知道这是不是只是我的……

I don't know if that's just my...

不，你说得对。完全正确。

No, that's right. That's exactly right.

你知道，你曾在顶级风投公司工作。你想按自己的方式做事，对吧？所以你按自己的方式走了出来。你对自己说：“我能做到。”这并不容易。这是一方面。另一方面，你必须热爱这个过程，对吧？如果你不喜欢建设，这些事情就太难了。为了钱做这些事是很糟糕的。有比创造非凡事物并与英伟达这样强大的对手竞争更容易的赚钱方式。这不是最简单的路。你必须热爱做“大卫”。我是职业“大卫”。这是我的第五次创业，我挑战“歌利亚”。这就是我的谋生之道。我对自己说，我们卖出的每一美元、每一百万、每一十亿，如果不是因为我们的头脑，他们的肌肉力量早就抢走了。你必须热爱这一点。如果你不爱，那将是一条非常漫长的路。

You know, you were at a top venture firm. You wanted to do it your way. Right? And so you stepped out in doing it your way. And you said to yourself, I can do this. And it's not easy. And that's one thing. The other thing is you have to love the journey, right? These things we do are too hard if you don't like the building. Right? That you do this for the money is a horrible thing. There are way easier ways to make money than trying to create something extraordinary and compete with somebody as strong as Nvidia. That is not the easiest path. You got to love being a David. Right? I'm a professional David. This is my fifth startup. I compete against Goliath. That is what I do for a living. And I think to myself that every dollar, every million dollars, every billion dollars we sell, if it wasn't for our brains, their muscle would have taken it in a heartbeat. And you got to love that. And if you don't love that, it's a very long road.

你怎么看？关于何时放弃，有两种观点：一种是不管怎样都坚持下去，希望最终能成功；另一种是不断重新评估你的道路是否正确，有时放弃才是最明智的选择。你怎么看？或者你认为什么时候是放弃的正确时机？

When do you think, because there are two views of the world in terms of when to give up on something? One argument is just keep going no matter what, and hopefully things work out eventually. The other view is you should be constantly reassessing whether the journey you're on is the right one. And there are some moments where actually giving up is the smartest possible thing you can do. What's your view on that? Or how do you think about when's the right time to give up on something?

我认为，当你列出一系列关于获胜所需条件的假设，而它们全都得到负面结果时，就是放弃的正确时机。

I think it is clearly the right time to give up when you've laid out a set of hypotheses about what it's going to take to win. And they all come back negative.

是的，但我看到人们会依次这么做，对吧？他们说：“哦，我只需要再测试一件事。”然后测试了，不行。又说：“我需要再测一个。”如此循环……

Yeah, but I see people kind of do this sequentially, right? They say, "Oh, I just need to test one more thing." And they test and it doesn't work. And they say, "I need to test one more." And so...

滑坡效应是个恶魔。在所有事情中——道德情境、生活中——滑坡效应都是你必须警惕的，对吧？我认为有时需要其他前 CEO 或经验丰富的企业家站在你这边，他们可以提醒你：“记得一年前你说过，如果到了这一步还没有这个，你就会放弃。”他们把你拉离滑坡。就像温水煮青蛙的故事：你说过如果水这么热你就会跳出来，但水慢慢变热了。

The slippery slope is a beast. The slippery slope in all things. In ethical situations, in your life. I mean, the slippery slope is really something you have to guard against. Right? And I think sometimes having other former CEOs or other really seasoned entrepreneurs who are on your side and who can share with you, "Remember a year ago you said if you got to this point and you didn't have this." And to remind you. So they pull you back off that slippery slope, right? They said, you know, the old frog in the warm water thing is like you said if it got this hot you were going to get out. And it slowly kept getting warmer.

别人能有效地让你对自己的指令负责吗？

Can other people keep you effectively accountable to instructions?

对你自己的思考负责。

To your own thinking.

是的。

Yeah.

如果你理解为什么行不通。对吧？如果你能明确说出哪些事情必须改变才能成功，并且可以设定一个时间框架。

If you understand why it's not working. Right? If there are some things that you can articulate that have to change in order for it to work and you can put some sort of time frame on it.

嗯，但这是一个极其困难的问题。我认为很可能很多努力都应该被截断。嗯，是的。那些人会把精力重新部署到他们拥有的新想法上。是的，这有点像，你知道，生活中存在机会成本，对某些人来说，这是他们一生中生产力或能做事情的最佳时刻，所以时间成本极高。嗯，在你们的情况下，显然成功了。是什么让你们决定上市？同样，关于何时上市、为何上市、好处和缺点，存在不同意见。你当时是怎么想的，是什么让你决定现在上市？

Um but that is an extraordinarily hard question. And I think it's probably the case that lots of efforts ought to be truncated. Mhm. Yeah. And those people sort of redeploy their efforts to new and different ideas that they have. Yeah, it's kind of like I have you know there's opportunity cost in life and for some people it's the best moment of their lives that in terms of productivity or things I could do and so you know, the cost of time is extremely high. Um you know, in your guys' case obviously it worked out. What made you all decide to go public? Similarly, there's differing opinions on when to go public, why to go public, what's the benefits, what's the drawbacks. What was that in your mind and what made you decide to go out now?

首先，上市是用一些专业投资者（专注于科技投资的风险资本家）来换取另一类投资者，从而稍微降低你的资本成本。对吧？这其实就是正在发生的事情。嗯，我们突然从像你这样的专业人士变成了我爸爸。对吧？这就是一种权衡。作为回报，你必须同意接受一套极其严格的规则。我认为你的问题因为一个事实而变得复杂：历史上第一次有四到五家公司可以在不上市的情况下筹集巨额资金。这在 OpenAI、Anthropic 和可能 Databricks 之前从未发生过。

First, going public is exchanging some professional investors, venture capitalists who specialize in technology investing, for a different class of investors and in so doing reducing your cost of capital a little bit. Right? This is really what's happening. Mhm. Suddenly we go from pros like you to my dad. Right? That's sort of the trade-off. And in return for that, you have to agree to be governed by a set of extraordinarily stringent rules. I think your question is complicated by the fact that there have been for the first time in history four or five companies that can raise huge amounts of money without going public. That this was never a thing before OpenAI and Anthropic and maybe Databricks.

硅谷的期权包时间线来自哪里？大概是 4 年的时间线。

Option package timeline from Silicon Valley comes from. It's like a 4-year timeline.

是的，过去这通常是你上市所需的时间。对吧，过去是 4 年，那是你获得数亿美元估值的方式，对吧？但我认为

Yeah, it used to be how long it would take you to get public. Right, it used to be 4 years and that was the way you got a valuation in the hundreds of millions, right? But I think

有一个要约收购周期。

have a tender cycle.

没错。

That's right.

并且达到一定规模。

And at a certain scale.

我们花了 10 年。我认为这改变了很多。对吧，我们所做的是开放二级市场，让人们出售股份，对吧？如果你要把职业生涯的大部分押在我们身上，我们认为让你在过程中获得适度的流动性是完全合理的。我认为如果这需要十年，你必须非常不同地思考。但我认为对于极少数公司，特别是那三家，它们能够在私募市场以公开市场估值筹集公开市场资金。我认为对于世界上的其他公司，如果你想要超高估值，如果你想要随之而来的合法性。历史上，大公司喜欢与美国其他上市公司做生意，你通过审计账目、让他们看到你是谁而获得可信度和合法性，这与私营时不同。我认为所有这些理由都是合理的。我也认为我们可以为公开市场提供一些独特的东西，对吧？我们将是第一个也是在一段时间内唯一一个纯粹的 AI 公司。我们是唯一一家你可以 100% 收入来自这个确切市场的公司。没有游戏，没有图形，没有 PC，就是这个。这是一个机会，一个我们认为有趣的差异化因素。我认为所有其他事情都有办法解决。你可以为投资者带来回报。我认为 Elon 和 Ali 都非常有创意，允许员工出售股份，并允许拥有 10 年基金的投资者在这个过程中找到一些流动性。但最重要的是，对我们来说，这是一个从企业青春期毕业到企业成年的机会。

It took us 10. And I think that changes a lot. Right, what we did is we opened up the secondary market and let people sell, right? If you're going to bet big chunks of your career with us, we thought it would be perfectly reasonable for you to find modest liquidity as you went along. I think you have to think very differently if it's going to take you a decade. But I think for a very small number of companies, those three in particular, they've been able to raise public market money at public market valuations in the private market. I think for the rest of the world, if you want super high valuations, if you want the legitimacy that comes with it. Historically, large companies like doing business with other public companies in the US and you get a credibility and a legitimacy from having your books audited, from them being able to see who you are that is different than when you're private. I think all of those are reasonable reasons. I also think we could offer the public market something unique, right? We would be the first and only for a period of time AI pure play. We are the only company that you can 100% of the revenue this exact market. There's no gaming, there's no graphics, there's no PC, this is it. And that was an opportunity, a differentiator that we thought was interesting. I think there are ways around all the other things. You can deliver returns to your investors. I think both Elon and Ali have been really creative about allowing employees to sell and allowing investors who have 10-year funds to find some liquidity in the process. But I think more than anything, for us, it was an opportunity to graduate from corporate adolescence to corporate adulthood.

你能谈谈吗？我很好奇，OpenAI 的交易是怎么发生的？你知道，你认为在哪个时刻你意识到你们很适合他们？

Can you talk a little bit about I'm so curious like how did the OpenAI deal happen? You know, what were um what do you think was the point at which you knew that you were a good fit for them?

我想我在 2025 年仲夏与 Sam 谈过。他第一次说，我们一直在努力跟上需求。我们现在看到了快速推理的重要性。这产生了一系列试验和一些测试。我们比竞争对手快得多。感觉非常好。我们喜欢与超级聪明的客户交谈，对吧？我的意思是，我做不了消费者业务。我有一个规则：如果我妈妈买或使用它，我就不想制造或销售它。因为我真的想要超级聪明的客户，他们用我们的东西做非常有趣的事情。所以我们接触了他们的一些人，他们说：“哇。我们现在明白了。”在感恩节，感恩节前夜，我们签署了条款清单。你知道，4 周后，12 月 24 日，我们签署了一份大型主协议。所以非常快。

I think I spoke to Sam in the middle of summer in 2025. And he said for the first time, he said we've been trying so hard just to keep up with demand. We now see the importance of fast inference. That produced a set of trials and some testing that was done. And we were so much faster than the competition. It felt really good. And what we love is talking to super smart customers, right? I mean I can't do consumer. I have a rule that if my mother buys it or uses it, I don't want to make it or sell it. Because I really want super smart customers who are doing really interesting things with our stuff. And so we got in with some of their guys and they were like, "Whoa. This is We understand now." And at Thanksgiving, the night before Thanksgiving, we signed a term sheet. And you know, 4 weeks later on the 24th of December, we signed a big master agreement. And so incredibly fast.

难以置信地快。

incredibly fast.

你知道吗？他们能飞。我们每周工作 7 天。我的意思是，他们有好几家律师事务所。对于一个 200 多亿美元的交易，在 4 周半内完成是 exceptional 的。

You know what? They can fly. And we were working 7 days a week. I mean, they had several law firms. I mean it was a huge deal for a 20-plus billion dollar deal to do it in 4 and a half weeks was exceptional.

我实际上认为这是这个市场的一个疯狂特征，我以前从未亲身经历过，那就是每个人都在努力跟上需求。

I actually think that's like a crazy characteristic of this market that I've not personally experienced before which is everybody's trying to keep up with demand.

我认为你知道我和 Cognition 的人谈过，对吧？他们在一个周末内收购了 Wind Surf。对吧？我认为很多我们认为是光速的事情其实不是。对吧？可以做得更快。而且我认为，你知道，Elon 建造数据中心的速度。每个人都说：“哦，你不能那样做。”除非你是他，那样你就能。或者你不能在三天内收购一家 3 亿美元的公司。实际上，你可以。你不能在 24 天内完成这样的交易。但如果你每天工作 8 或 10 小时，你可以。我认为这种推动以我从未预料到的方式扩展了可能性的艺术。

I think you know I talked to the guys at Cognition, right? They bought Wind Surf over a weekend. Right? I think many of the things that we thought were speed of light weren't. Right? Could be done much faster. And I think, you know, the rate at which Elon has been able to build data centers. Everyone says, "Oh, you can't do it that way." Except if you're him, in which case you can. Or you can't buy a $300 million company in three Actually, you can. You can't do a deal like this in 24 days. But if you work on it every day, 8 or 10 hours a day, you can. And I think the art of the possible has been expanded by this push in a way I never would have expected.

而且我认为，如果你相信速度是可能的，那么拥有对速度的雄心是一个巨大的优势。

And I think it's a huge advantage to have the ambition for speed if you believe it is possible.

没错。我认为我们在市场上看到了一些非凡的运营者建造了令人惊叹的东西，对吧？我的意思是 Cursor 或 Cognition 的人。你看到了我们从未见过的增长。你不能增长那么快。嗯，实际上你可以。你不能建造数据中心。你不能做交易。这些只是被截断的愿望，这很有趣。

That's right. I think we have seen some extraordinary operators in this market build amazing things, right? I mean the guys at Cursor or Cognition. You see sort of growth we've never seen before. You can't grow that fast. Well, actually you can. You can't build data centers. You can't do deals. It just Those were sort of truncated aspirations which is interesting.

说到这些像 Cog 和 Cursor 之类的，开源生态系统的增长使一代公司能够做非常令人印象深刻的事情。比如

Speaking about these like Cog and Cursor and such the growth of the open source ecosystem has enabled a generation of companies to do really impressive things. Like

超级超级令人印象深刻。

Super super impressive.

你知道，Devin 在 Cerebrus 上是一种真正神奇的体验。在 Cerebrus 上编码就像大规模速度下的高性能，非常特别。

You know, Devin on Cerebrus is a really magical experience. Coding on Cerebrus is like high performance at massive speed is really special.

你如何看待开源和后训练工作负载，以及你对此未来的看法？

How do you think about open source and post-trained workloads and your perspective on that going forward?

它们滋养了这个市场。当闭源过于昂贵时，开源社区保持了兴趣的火种，让火焰持续燃烧。我认为这推动了闭源厂商。我们看到一些中国厂商的技术，让我们觉得‘哇，我们必须保持领先。我们不能躺在功劳簿上。不能依赖我们拥有更大的训练集群和更多数据的事实。’我认为这造就了一个异常活跃的生态系统。它激发了创造力，让创意生根发芽，真正产生有趣的结果。参与其中很有趣。看到别人的想法在你的硬件上实现有趣的事情，这很有趣。如果你不喜欢这一点，那你的基础设施就不适合你。你必须热爱别人的想法在你所建造的东西上起飞。

They have fed this market. When closed-source was too expensive, the open-source community kept the interest alive and kept the flame going. And I think that pushed the closed-source guys. The techniques we saw from some of the Chinese makers made us think, 'Whoa, we got to stay ahead of that. We can't rest on our laurels. We can't depend on the fact that we have bigger training clusters and more data.' I think that's made for an extraordinarily vibrant ecosystem. It's made for creativity and allowed creativity to take root and really produce interesting results. And that's fun to be in the mix of. It's fun to see other people's ideas do interesting things on your hardware. If you don't love that, your infrastructure's not right for you. You got to love other people's ideas taking flight on what you built.

当你想到你认为只有在 Cerebras 上才能实现的体验时，未来几年有什么让你兴奋的事情值得我们关注吗？

When you think about experiences you imagine will be possible only on Cerebras, is there anything you're excited about in a couple years from now that we should all look out for?

当我思考速度带来的影响时，它并不是让现有商业模式变得更好一点。Netflix 曾经用信封寄送 DVD，他们认为竞争对手是 Blockbuster。当互联网变快后，他们变成了一家电影制片厂。这就是速度带来的变化。他们并不是在递送 DVD 上逐步改进，而是开启了一个全新的业务，一些根本不同的东西。然后他们变成了电影制片厂，收购了现有的电影制片厂。我认为快速 AI 也是如此：它将呈现全新的商业模式。简单明显的做法是取代现有事物。当 PC 出现时，它取代了打字机和总账会计。但生产力的巨大飞跃发生在它重新组织我们工作方式的时候，你得到了云，有了云就有了 SaaS，有了 SaaS 我们就得到了以前负担不起的工具，因为对单个公司和小规模用户来说太贵了。然后你获得了生产力的巨大飞跃。我认为 AI 也是如此。现在我们正在取代每个人都能看到的东西，比如编码、设计、一些 SaaS 工具。但一旦我们开始围绕它进行根本性的重组，你就会看到新的商业模式和生产力的大幅提升。我对此充满期待。

When I think about what speed does, it doesn't make the existing business models a little better. Netflix used to deliver DVDs in envelopes, and they thought their competition was Blockbuster. When the internet got fast, they became a movie studio. That's what happens with speed. It wasn't that they got incrementally better at delivering DVDs; it opened up an entirely new business. Something fundamentally different. And then they became a movie studio. They bought existing movie studios. I think that's what fast AI does: it will present entirely new business models that are available. The easy and obvious is to replace existing. When the PC came in, it replaced typewriters and general ledger accounting. But the big jump in productivity was when it reorganized how we did work and you got the cloud, and with the cloud you got SaaS, and with SaaS we got tools that you previously couldn't afford because they were so expensive to the individual company and to the small number of seats. Then you got this massive jump in productivity. I think AI is the same way. Right now we're replacing things that everybody can see, like coding, design, some of the SaaS tools. But once we start fundamentally reorganizing around this, you're going to see new business models and fundamental jumps in productivity. I'm eager for that.

太酷了，非常令人兴奋。非常感谢你今天参加我们的节目。

That's so cool. Very exciting. Thank you so much for joining us today.

各位，非常感谢你们邀请我上节目。非常感激。

Guys, thank you so much for having me on your show. Really appreciate it.

恭喜。

Congratulations.

非常感谢。

Thank you so much.

在 Twitter 上关注我们 @No Priors Pod。如果你想看到我们的脸，请订阅我们的 YouTube 频道。在 Apple Podcasts、Spotify 或你收听的地方关注节目。这样你每周都能收到新剧集。注册邮件或在 no-priors.com 上查找每集的文字稿。

Find us on Twitter at No Priors Pod. Subscribe to our YouTube channel if you want to see our faces. Follow the show on Apple Podcasts, Spotify, or wherever you listen. That way you get a new episode every week. And sign up for emails or find transcripts for every episode at no-priors.com.