AMD 营收增长 38%，Anthropic 与谷歌达成 2000 亿美元协议，OpenAI 总裁格雷格·布罗克曼谈埃隆·马斯克

欢迎各位收看Informations TITV。

Welcome everyone to the Informations TITV.

我是阿卡什·帕什里卡。

My name is Akash Pasricha.

今天是5月6日，星期三。

It is Wednesday, May 6.

今天第一个消息，AMD昨天发布了季度财报，股价随之上涨。

First up today, AMD shares are rising after their quarterly results yesterday.

该公司报告称，得益于数据中心业务的推动，营收增长了38%。

The company reported a 38% jump in revenue bolstered by its data center segment.

我们马上详细解读这些数据。

We'll get into the results in just a moment.

此外，Information独家报道了新数据，揭示了Anthropic与谷歌业务之间日益紧密的联系。

Plus, the information published exclusive reporting with new numbers that highlight the growing link between Anthropic and Google's businesses.

我们请来了揭露这一消息的记者。

We'll bring on the reporters who broke that story.

我们还会听听火箭德鲁的每日更新，他一直在报道埃隆·马斯克对OpenAI的庭审。

We'll also get our daily check-in from Rocket Drew, who was covering the Elon Musk OpenAI trial.

接着我会采访Tessera Labs的首席执行官，他们刚刚宣布了新一轮数千万美元的融资，最后我们会分享来自Information订户调查的洞察，看看在ChatGPT、Claude和Gemini这三者之间的竞争中，谁更占优势。

I'll then talk with the CEO of Tessera Labs, which just announced a new multimillion dollar funding round, and we'll close out the show with some insights from the information subscriber survey highlighting who is winning in the three way race between ChatGPT, Claude, and Gemini.

今天这期节目内容很精彩，咱们马上开始。

It's going to be a fun show, so let's get right on into it.

AMD又交出了一份亮眼的季度财报。

AMD posted another big quarter.

第一季度营收增长了38%，达到103亿美元，主要得益于数据中心业务的强劲表现。

Revenue jumped 38% to $10,300,000,000 in Q1, driven by its data center segment.

我们请来创意策略公司的首席分析师本·巴贾林，帮我们详细解读一下。

I want to bring on the CEO of Creative Strategies, Ben Bajarin, to help us break it all down.

本，欢迎再次做客我们的节目。

Ben, welcome back to the show.

很高兴你能来。

It's great to have you here.

是的

Yeah.

很高兴回来

Good to be back.

谢谢邀请我

Thanks for having me.

好的

Okay.

AMD的业绩上涨了

So AMD is up.

从这些数据里，你最关注的是什么？

What stood out to you from the results?

对

Yeah.

其实我们也没怎么感到意外

I mean, we weren't terribly shocked.

我觉得，早就有不少迹象表明，数据中心CPU的需求趋势已经很明显了，所以大家普遍觉得，AMD的情况应该和英特尔差不多，就是需求远远超过了供应。

I think, you know, there'd been enough signal around what we'll just say, like, data center CPU as a demand trend that I think everybody sort of assumed they would be in a similar spot as we saw from Intel in terms of just out you know, outpacing the the demand with or outspike outpacing supply with demand.

这其中很大一部分原因，是基础设施和智能推理这些新变化带来的，而这些在一年前根本还不存在。

And part of that's really just being driven obviously by changes in infrastructure, agentic inference that's happening, which, you know, was not a dynamic we had a year ago.

现在行业里冒出好多新东西，很明显，AMD是这些变化的最大受益者。

And so there's a number of new things coming to the industry where, obviously, AMD is a is is a huge beneficiary of that.

所以，大家的关注点完全集中在CPU的增长上。

And so really really the spotlight was was on CPU growth.

他们更新了总可用市场预估，这和我们对AI驱动下CPU市场规模的预测基本吻合。

You know, they updated their TAM, which sort of aligns with some of our forecasts around the CPU TAM because of agentic.

说实话，我觉得现在整个故事的核心就在这里。

And really, I think overall, right, this is the bulk of the story right now.

当然，GPU也不是不吸引人。

Not that GPUs aren't, you know, sexy.

显卡和加速器依然重要，但眼下真正的焦点是CPU，光是这一条，就足以推动这一块市场的走势了。

They and accelerators are still relevant, but big spotlight on CPUs, and that story alone seems to be enough to be driving this part of the market.

所以当你说可服务市场在这里扩大时，具体是指什么？

So when you talk about the total addressable market expanding here, the idea here is what?

代理式推理意味着在这个环境下需要更多的CPU，对吧？

That agentic inference, I mean, that requires more CPUs in this environment, right?

是的。

Yeah.

我觉得最好的理解方式是，在代理式推理出现之前，我们通常说的云原生CPU是什么情况。

And I think the best way to look at it, right, prior to agentic, so if we were just talking about, you know, the framing that we use, which is cloud native CPUs.

在那之前，AMD、ARM，还有英特尔都会说，他们主要把这些芯片卖给超大规模云服务商。

So prior to this moment, AMD, even ARM, right, Intel would talk about, you know, we're mostly just selling these to hyperscalers.

这些客户在运行企业级软件。

They're running enterprise software.

对吧？

Right?

他们在运行自己的电商平台。

They're running their e commerce.

他们还在运行邮件服务。

They're running their email.

比如在托管云上的工作流。

Like, they're hosting cloud workflows.

那个市场之前并没有增长。

And that market wasn't growing.

每年大约是两千五百亿美元，一直没什么变化，CPU也没多大吸引力。

It was roughly a $25,000,000,000 market a year and staying relatively flat, and CPUs were not super interesting.

现在有人开始说，这个数字已经发生了巨大变化，就连Lisa Sue去年都提到，他们把CPU的总潜在市场规模上调到了六千亿美元。

Now people are coming out and saying that number is drastically changing as of even, you know, Lisa Sue said last year, they changed their CPU TAM to going up to 60,000,000,000.

从两千五百亿涨到六千亿，光是这一项就相当惊人了，对吧？

So from 25 to 60, which alone at the time was like, that's a pretty big jump, right?

现在又说，智能代理还能再带来四千多亿的增量，甚至他们直接喊出了一万两千五百亿美元的数字。

And so now saying, well, agentic drives another 40 plus billion, or in their case, than that because they're saying 125,000,000,000.

我们最初估算时大概是一万一千亿，这跟过去常年两千五百亿、几乎不增长的市场相比，简直是天翻地覆的变化。

You know, we were around 110 at at our first shot with this, is again a drastic increase from what it was typically a $25,000,000,000 market and not growing.

所以我认为在市场层面，大家现在都看出来了，总市场规模正在扩大。

And so that's why I think in the market context, people are seeing that and realizing that the TAM's expanding.

以前我们谈AMD切入英特尔或ARM的数据中心市场时，说的是抢占份额。

It used to be a share taking environment when we talked about AMD encroaching on a you know, on Intel in data centers or even ARM.

但现在不再是抢占份额了。

It's not share taking anymore.

而是直接抢夺，就像丽莎说的，年增长率达到了35%左右。

It's share grabbing in, you know, what is essentially, as Lisa said, a 35%, you know, year over year growth.

单说智能代理这一块，它的年增长率就达到了65%。

And just for agentic, we have that growing 65% year over year for just agentic

系统。

systems.

为什么现在是抢夺份额，而不是抢占份额了呢？

Share why is it share grabbing now instead of share taking?

在你看来，这两者之间到底有什么区别？

And what what is the difference there in your mind?

是的

Yeah.

对

Yeah.

我的意思是，当这个市场还只有250亿美元的时候，大家都知道每年能卖多少台设备。

I mean, when, again, when it was a $25,000,000,000 market, right, and growing, we know how many units that moved every year.

那时候，无论是AMD、英特尔，还是之前大家谈ARM潜力时，能增长全靠从别人手里抢份额，因为市场根本没怎么扩大。

It was their ability to grow for AMD or Intel or even when you thought about, like, arms potential, right, prior to this was purely on take share because it wasn't growing.

但现在市场体量大幅扩张了，无论是出货量还是单机价格，都涨得厉害，CPU在智能代理和新数据中心领域，其实都是在开疆拓土。

Now that the market is is moving meaningfully to a drastically larger size, both in unit volume, ASP, you know, as a dollar amount, it's really just greenfield growth for CPUs in agentic for new data centers.

所以从某种程度上说，就像我们之前看GPU的增长故事——NVIDIA虽然占主导，但整个AI加速器市场冲到万亿美元规模，我这还没算上CPU呢，其实是大家都受益。

So in a lot of ways, like we thought about, right, that the GPU expansion story, so NVIDIA, while being dominant, right, moving to a trillion dollar market for just AI accelerators, I'm not even including CPUs in this point, was a everybody benefits, right?

这波增长带来的红利，不光是NVIDIA的。

It's not just dollars there for NVIDIA.

AMD也能分到一杯羹。

There's dollars for AMD.

其他竞争对手也能分到钱。

There's dollars for other competitors.

还有那些自己研发定制芯片的公司，也都有钱可赚。

There's dollars for, you know, the guys, everybody making their own custom silicon.

所以从市场总规模扩张的角度来看，这完全是新出现的商机。

So in TAM expansion, like we're talking about, it's brand new dollars to go after.

不是说我要从竞争对手手里抢钱。

It's not, I have to take money from my competitor.

对。

Right.

我有个问题，既然市场在扩大，那你觉得AMD有多大把握能真正增长自己的份额？

I guess my question is, so in that story though, I mean, confident are you in AMD's prospect to then grow its market?

市场在变大，这点我同意。

The market's expanding, I'm with you.

但说到提升市场份额，目前的策略能持续多久？能真的抢到更多市场比例吗？

But in terms of growing the market share, how sustainable is the current strategy in terms of trying to grab a higher percent of the market?

非常有把握。

Very.

我的意思是，再强调一遍，所有声称在数据中心做CPU的公司，其实都在夸口。

I mean, I think, you know, again, and I'll make this point for everyone in CPUs, right, that's making a claim that they're doing CPUs in the data center.

他们瞄准的是一个巨大的市场扩张机会。

This is a massive expansion that they're going after.

从AMD的角度来看，有意思的是，他们一向很擅长与超大规模客户深度定制和合作。

From AMD's standpoint, right, what's interesting is they have typically done a lot of customization and collaboration with their hyperscale customers.

比如谷歌、亚马逊、微软这些巨头。

So, you know, the the Googles, the Amazons, the Microsofts of the world.

但说实话，这种合作模式还得延伸到新兴云服务商才行。

And this will have to extend to Neo Clouds, to be honest with you.

这正是我一直在跟那些新兴云信徒争论的地方。

Like, this is something I've been arguing with the neo clan cultists.

我知道他们不爱听我说这话，但那些圈子里的人，他们的基础设施基本都还停留在训练时代。

I know they hate it when I say that, but, like, the folks in in in those communities is that most of their infrastructure has been in the training era.

对吧？

Right?

如果你只是在部署英伟达的方案，比如Blackwell那些，它们主要用在训练和部分推理上。

If you're just installing NVIDIA solutions, right, Blackwells and whatnot, those were training and some inference.

大家都认同，推理的基础设施和训练完全不同，这时候CPU机架就派上用场了。

Everybody's agreeing that inference infrastructure looks drastically different than training, and that's where CPU racks come in.

所以，数据中心里需要部署的CPU数量庞大得惊人。

So there's a ton of data center CPUs that have to get played out.

AMD一直有个策略：我们可以调整产品型号。

And AMD has always had a strategy of saying, we can change our SKUs.

我们有一套可扩展的架构，能对CPU做针对性优化，或者保留它来跑云原生的工作负载。

We have a one architecture that scales, and we can do agentic tweaks to our CPU, or we can keep them for cloud native workloads within their pipeline.

而且，他们和那些超大规模云厂商的合作非常深入。

So and again, they are deep customers with the hyperscalers.

现在CPU的需求大到超乎想象，比一年前我们预估的还要多得多。

And right now, there's so much CPU demand, like, really more than we ever thought a year ago was going to come up.

光是这一点就说明，只要你能在数据中心CPU领域占有一席之地，市场就会明显扩大。

And that alone, it just means if you're relevant in data center CPU, like, you're gonna see market expansion.

没错。

Correct.

关于AMD的整机架系统，以及它和NVIDIA的竞争，我们从结果中看到了什么？

What did we get from the results around AMD's full rack system and how that competes with NVIDIA?

目前进展到什么程度了？你有多大的把握认为它真能构成威胁？

Where is that at, and how confident are you that that will actually pose a challenge?

是的。

Yeah.

但目前还谈不上太多进展。

I mean, not a lot yet.

对吧？

Right?

他们七月有个叫‘Advancing AI’的活动，到时候我们就能深入了解Helios，还有他们称作MI555的下一代产品。

They have an event coming up in July called Advancing AI where, you know, we're gonna learn a whole lot about Helios as well as the the successor to Helios, which they call MI555.

你们在电话里就一直想打探这事。

You People tried to prowl this on the call.

大家都想知道他们之前的MI355到底是什么样子。

Everybody wants to know what their prior, so MI355 has looked like.

但说实话，规模并没有特别大。

Again, it hasn't been gigantic.

市场的重心已经转向了NVIDIA的加速器和定制ASIC，这些产品依然主导着整个讨论。

The role of the market has moved, again, largely to NVIDIA's accelerators as well as custom ASICs, which which continue to dominate this conversation.

但我认为，真正的增长会出现在今年下半年。

But I think the second half of the year is where this ramp is.

我这里做个概括性的说法，不过这对AMD来说也很关键。

And I'll just sort of lump this as a broad statement, but I think this is relevant for AMD too.

我们现在在AI算力上缺口太大了，实际上，只要是可行的算力方案，都会被快速采用。

We are at such a compute deficit for AI that the reality is almost any viable compute will get adopted.

所以你们才看到他们跟OpenAI这些公司签了合作。

And that's why you saw them do deals with, you know, OpenAI, right, and others.

我觉得你们还会看到他们做更多这样的合作，因为单靠一家供应商的算力根本不够用。

And I expect you'll see them do a whole lot more because you just can't get enough compute for one vendor.

你至少得找两家，可能还得三家，甚至四家供应商，才能满足眼下所有人看到的这种需求。

You need two, you need three, you may even need four vendors to have enough to meet this demand, right, that everybody's seeing.

所以这里面有大量的机会和增长空间。

And so there's a lot of opportunity and growth side.

我觉得下半年才是关键，就像丽莎和管理层说的，那时候才会真正开始加速。

And I think the second half of the year is what, you know, Lisa and management has said this starts to ramp.

等我们看到 MI455 真实表现，还有在 Advancing AI 大会上展示的 Helios 之后，大家对它未来的发展，以及到 2027 和 2028 年能做多大，才会更有信心。

And I think once we see what MI four fifty five looks like, so Helios at the Advancing AI Day and beyond, people will have a little bit more confidence about where that goes and how much they can grow that business into 'twenty seven and 'twenty eight.

对。

Right.

最后一个问题。

Last question for you.

你觉得得发生什么，投资者才会对 AMD 失去兴趣？

So what do you think it would take for investors to fall out of love with AMD?

公司现在显然处于高速发展阶段，但我们也看到过一些芯片公司经历过繁荣期后，增长却没能持续。在你看来，估值倍数什么时候会开始回落？

This is obviously This is boom times for the company, but now we've seen a couple chip companies go through their boom period and then keep booming, but when does the multiple start to contract in your opinion?

是的，关键问题其实是你能把产能扩张到什么程度。

Yeah, I think the debate really is how much capacity, right, can you grow and scale.

如果你一直处在供应受限的环境中，大多数人就会根据他们对你的未来增长预期来建模。

And if you're just constantly in a supply constrained environment, then most people will model in what they believe, right, your forward projections can be in terms of growth.

所以如果产能是瓶颈，那唯一的上升空间就是你还能不能增加更多产能。

And so if that's your constraint, then the only upside is can you get more capacity?

你上哪儿去搞更多的晶圆？

Like where are you gonna get more wafers?

你上哪儿去增加更多的制造能力？

Where are you gonna get more manufacturing capacity?

说实话，我觉得现在 NVIDIA 正面临这样的问题——大家觉得，我们已经很清楚你能增长多少了，或者至少能根据你的供应链情况估算出来，这些预期已经反映在股价里了。

And that's, to be honest with you, a dynamic that I think is hitting NVIDIA is that most people just have said, we know how much you can grow, or we think we know how much you can grow based on your supply chain, and that's what we're pricing in.

我认为，这正是现在所有人建模时都在考虑的核心问题。

And that's really, I think, the dynamic that everybody's working on in their models.

所以 downside 就是他们的增长已经到头了。

So the downside is, right, that they've capped.

大家都清楚他们已经到上限了，模型里也已经反映了这一点。

They're capped and everybody knows it and they've modeled that.

那他们能不能打破这个预期呢？

And can they, you know, defeat that narrative?

如果真有什么转折点，大概就在这里了。

But that would be the stale point, you know, if there's anything.

CPU 领域的势头，依然在持续上升。

The momentum in CPUs, right, remains an upward trajectory.

他们的业务还会增长，但究竟能增长多少，就成了模型关注的核心。

Their business will grow, but how much can they grow really becomes then, you know, where the model gets focused on.

没错。

Right.

很好。

Great.

本，非常感谢你今天来做客。

Well, Ben, I wanna thank you for coming on.

这位是来自Creative Strategies的本·巴贾林，欢迎来到TITV。

That is Ben Bajarin from Creative Strategies here on TITV.

大家都知道，Anthropic和谷歌这些年来业务一直同步增长，但我的同事Sri Muppidi、Erin Woo和Amir Afrati最新独家报道揭示了这种依赖关系如今已经到了何等惊人的地步。

It is no secret that Anthropic and Google's businesses have grown in tandem over the years, but new exclusive reporting from my colleagues, Sri Muppidi, Erin Woo, and Amir Afrati reveals the great extent to which that is now the case.

我想请Sri和Erin来详细聊聊他们发现了什么。

I want to bring on Sri and Erin to talk more about what they found.

Sri，我先从你开始。

Sri, I want to start with you.

关于Anthropic对谷歌的依赖，你们发现了什么？

What did we find out about Anthropic's reliance on Google?

我们发现，Anthropic在过去五年里，向谷歌承诺的计算资源和芯片采购总额超过了两千亿美元。

We found out that Anthropic committed more than or roughly 200,000,000,000 in over five years to Google for computing power and chips.

这一点特别关键，因为Anthropic的营收增长速度实在太快了。

And so that's particularly important just because Anthropic has been scaling its revenue really quickly.

比如，这家公司上个月的年化收入已经突破了300亿美元，而年底时还只有90亿美元。

So for example, the company crossed more than 30,000,000,000 in annualized revenue last month, up from 9,000,000,000 at the end of the year.

随着公司持续扩张、不断新增客户，对算力的需求变得极为迫切。

And so as the company continues to scale and add on more customers, it needs computing power desperately.

因此，谷歌作为出色的云服务商，正好能满足安特ropic的这一需求。

And so Google is an excellent cloud provider that can serve that need for Anthropic.

这就是安特ropic对谷歌的依赖。

So that's Anthropic's reliance on Google.

你们报道中另一个特别有意思的部分，是你们实际上分析了OpenAI和安特ropic对各大云服务商的依赖情况。

The other part of your story that was really interesting is you actually looked at the reliance that both OpenAI and Anthropic have on all the hyperscalers.

能跟我们说说你们在这方面的发现吗？

Tell us a little bit about what you learned there.

当然。

Totally.

除了安特ropic和谷歌签下的这笔协议，我们还报道过谷歌与亚马逊达成了一笔价值1000亿美元的算力合作。

So along with Anthropic signing that deal with Google, we also had written about how Google had signed a $100,000,000,000 computing deal with Amazon.

亚马逊也和OpenAI签了协议，OpenAI承诺投入大约一千亿美元，这为亚马逊带来了约3640亿美元的收入待确认额。

Amazon also has a deal with OpenAI where OpenAI committed about $100,000,000,000 And so that results in about a $3.64 revenue backlog for Amazon.

同样，谷歌也有大约四千六百亿美元的收入待确认额，这指的是已经签约但尚未实现的未来客户收入。

And similarly with the Google backlog, which basically means the revenue contracted for future customer revenue, Google has about a $460,000,000,000 revenue backlog.

对谷歌来说，Anthropic在这笔待确认收入中占了大约百分之四十。

And so Anthropic for them represents about 40% of that backlog.

你可以看到，这些云服务商和AI实验室之间的关系有多紧密：谷歌、亚马逊、微软向这些实验室投资，而这些实验室反过来又把钱花回给他们。

And so you just essentially see how intertwined a lot of these cloud providers and AI labs are because as Google and Anthrop or sorry, Google, Amazon, Microsoft invest money into these labs, the labs, and then turn back and actually spend money on them as well.

当然，还有一点要提的是微软，它和OpenAI长期合作，现在也和Anthropic建立了合作关系。

And so the other one to mention of course is Microsoft, which has a longtime relationship with OpenAI and now also has one with Anthropic as well.

没错。

Right.

文章里有个很棒的图表，我建议大家去看看。如果我没记错的话，它显示的是，这些云服务商公布的收入待确认额中，大约一半——差不多一半——都来自OpenAI和Anthropic，对吧？

And there's a great chart in the story that I would encourage people to go look at, but it basically shows, if I'm recalling correctly, of the revenue backlogs that all of these companies have told us that they have, the cloud companies, roughly half of it, give or take, for I think most of these hyperscalers is coming from OpenAI and Anthropic, right?

我理解得对吗

Do I have that right,

没错。

Totally.

这些公司，也就是云服务商，预计的大部分收入其实都来自这些AI实验室。

So much of that revenue that these companies or these cloud providers are projecting will actually come from these AI labs.

随着AI热潮持续，投资者了解这一点非常重要。

And so I think that's quite important for investors to know as the AI boom continues on.

明白了。

Okay.

那么，埃琳，我们现在换个角度看看。

So Erin, let's look at it from the other side now.

谷歌在Anthropic身上花了多少钱？

So how much is Google spending on Anthropic?

是的。

Yeah.

谷歌最近表示，他们正在向Anthropic投资100亿美元，如果达到某些里程碑，还计划再追加300亿美元。

So Google has most recently said that it's investing $10,000,000,000 into Anthropic and their plan is to invest another $30,000,000,000 if certain milestones are met.

这还只是最近一次的投资金额。

And that's just the most recent amount of investment.

明白了。

Okay.

埃琳，我们换个角度想想，斯里提到的这种相互关联，你怎么看？

Erin, taking a step back here, what do you make of the interconnectedness here that Sri pointed out?

我们之前在各种场合都讨论过这种循环融资。

We've talked about this circular financing in all sorts of different circles.

这正是其中一个循环，我甚至能想象出把这些重叠的圈画在纸上的情景。

This is one of said circles, and I'm sort of imagining drawing all these overlapping circles on a piece of paper.

但你跟业内人士聊的时候，觉得这种局面是必然的吗？

But when you talk to people, was this inevitable?

他们对此感到担忧吗？

Is it something they're concerned about?

你个人对此有什么看法？

What's your reaction to it?

是的，我觉得这未必是必然的。

Yeah, I mean, don't necessarily think it's inevitable.

我觉得这又回到了大家一直在问的大问题：AI是不是一个泡沫？

I think this gets back to the big question that everyone's asking, which is, is AI a bubble?

这个泡沫会不会破？

Is the bubble going to burst?

比如这一周，谷歌这些主要云服务商的财报都挺不错，表现很好。

And so I think you have a week of pretty good earnings for all of these four major hyper sales like Google, for example, did very well.

但你仔细看数据就会发现，很多增长其实来自Anthropic。

And then you dig into the numbers a little bit and you see that a lot of this growth is coming from Anthropic.

其他一些服务商的增长，很大一部分也来自OpenAI。

A lot of this growth is coming from OpenAI for some of the other providers.

而且，除了我们这里提到的具体数字，比如企业跟这些服务商签的云服务合同之外，

And even beyond the specific numbers that we have in this that represent like the cloud deals that companies have signed with these providers.

还有额外的收入来源，比如谷歌云的部分增长并不是来自计算资源，而是来自转售Anthropic的模型。

Like, there's also additional revenue because some of the growth in Google Cloud's revenue that's not about compute, like, comes from the resale of Anthropic models.

所以这些很大程度上都跟这些公司绑在一起了。

So a lot of this is, like, bound up in these into companies.

这正是让人担心会形成泡沫的原因。

And so that's what's stoking fears of a bubble.

关键问题是，到底有没有人真正在用这些AI技术？

And the question is obviously, is anyone actually using this AI?

李，甲骨文当然也在这场游戏中，但他们的情况有点不一样。

Sri, Oracle in a Oracle is certainly in this equation, but they have sort of a slightly different case here.

甲骨文和这些公司的合作有什么特别之处？

What is different about Oracle's relationship with these companies?

甲骨文实际上正在为OpenAI建设这些数据中心。

So Oracle is actually working with OpenAI to build out these data centers on behalf of OpenAI.

OpenAI已经承诺在这项合作上投入大约三千亿美元。

And so OpenAI has committed about $300,000,000,000 towards that relationship.

所以很多投资者也在关注甲骨文的数据中心建设到底会怎么发展，因为这直接关系到OpenAI能否获得足够的算力。

And so I think a lot of investors are also sort of seeing how the data center build out on Oracle will actually play out because it definitely impacts OpenAI in terms of having that compute available.

所以我认为，Oracle 对微软、谷歌和亚马逊来说处境不太一样，因为它正在转型为一家主要的云服务商，专门承接这些大型合同。

And so I think that Oracle is perhaps in a different position for Microsoft, Google, and Amazon because it's reinventing itself as a major cloud provider to actually take on some of these huge contracts.

因此，Oracle 和 OpenAI 之间这种更紧密的合作关系很有意思，因为 Oracle 明白，只有搭上 OpenAI 这班车，才能进入这个市场。

And so I think that it's quite interesting in that more interconnected relationship that Oracle has with OpenAI because Oracle definitely realized in OpenAI to be able to get on that business.

对。

Right.

不过，艾琳，如果我理解得不对，请纠正我。

But Erin, correct me if I'm wrong.

他们的模式其实不太一样，因为他们不像谷歌和微软那样直接转售模型，对吧？

I mean, is slightly different in that they're not reselling the way that Google and Microsoft are, right?

是的。

Yeah.

谷歌、微软和亚马逊都能通过向客户转售模型来获得额外收入。

So Google and Microsoft and Amazon, they all make additional revenue reselling the models to customers.

比如谷歌，就有一个叫模型花园的服务。

So like Google, for example, has this thing called the Model Garden.

他们不仅提供谷歌自家的模型，还接入了Anthropic的模型以及大量开源模型。

So they offer Google's proprietary models, but they also offer Anthropic models and a lot of different open source models.

这又带来了一笔额外的收入。

So there's that additional revenue stream.

Erin，接下来你打算从哪些角度深入报道这个故事？有什么想法吗？

And Erin, as you look at the reporting questions that you have for this story going forward, what comes to mind?

我说过，其中一个关键问题是，企业到底在怎么使用人工智能。

Like I said, I think one of the big questions is companies are actually using AI.

坦白说，确实有不少企业已经看到了AI带来的巨大收益，比如Anthropic从企业级AI业务中获得的快速增长。

And I mean, certainly to some extent companies are like we've seen the huge revenue growth that Anthropic has that comes from enterprise AI.

我昨天刚报道过一条新闻，谷歌正和几家私募股权公司洽谈合作。

Another story that I reported yesterday was that Google and private equity firms are in talks to form a new partnership.

这些公司包括黑石、凯雷和EQT。

So this is Blackstone, KKR, EQT.

这个合作旨在把人工智能技术推广到它们的投资组合公司中。

And this would help distribute AI to the portfolio companies.

我最近写过一些文章，谈到有些公司在采用人工智能时遇到了不少困难。

I mean, one of the trends that I've written about recently is the difficulties that some companies are having adopting AI.

这对谷歌来说是有利的，因为这些被投公司都会开始使用它的服务。

And so this would be beneficial for Google, for example, because all of these portfolio companies would start using it.

只要你相信人工智能对被投公司有帮助，那对它们和私募股权公司来说也都是好事。

And as long as you believe that using AI is also beneficial for the portfolio companies, it's also good for them and private equity firms in that way.

没错。

Right.

确实，你和Sri最近不断有新闻、公告和报道，都在讲私募股权公司如何积极推动旗下企业应用AI来提升管理。

Yeah, it certainly seems like a steady drumbeat of news or announcements or stories from you, and Sri, mind you, talking about how the private equity firms are really leaning into this to get their portfolios in check.

这值得持续关注。

It's one to watch.

谢谢你们两位来参加节目。

I want to thank you both for coming on.

今天我们请到的是《The Information》新闻团队的Sri Muppidi和Erin Woo。

That is Sri Muppidi and Erin Woo from our newsroom here at The Information.

《The Information》的罗克特·德鲁本周全程在法庭现场报道埃隆·马斯克诉OpenAI的庭审。

The Information's Rocket Drew is covering the Elon Musk OpenAI trial from the courtroom all week long.

这是他关于OpenAI总裁格雷格·布罗克曼周二第二日作证的最新进展。

Here is his latest update for us on OpenAI President Greg Brockman's second day of testimony in the court on Tuesday.

格雷格·布罗克曼在法庭上的表现比之前好了一些。

Greg Brockman had a slightly better day in court.

之前，马斯克一方的律师对他进行了严厉质询，涉及他的个人财务状况以及在OpenAI的持股情况，而这次，布罗克曼得到了OpenAI律师较为友善的提问机会。

After previously facing some really rough questioning from a lawyer for Elon Musk who asked about Brockman's personal finances and his stake in OpenAI, Brockman had the chance to answer some friendly questioning from a lawyer for OpenAI.

这让他有机会解释自己私人日记的背景。

That gave Brockman the opportunity to explain some of the context around his personal journal.

这份数字日记被马斯克团队拿来当作关键证据，声称其中包含布罗克曼意图欺骗马斯克、隐瞒OpenAI未来规划和公司结构的自白。

This is the digital diary that Elon Musk's side has presented as though it contains a smoking gun, an admission of guilt from Brockman that he was intending to lie to Elon Musk or deceive him about his plans for OpenAI's future and for its corporate structure.

特别是其中一段，布罗克曼曾考虑过一条他本人认为道德上无法接受的路径。

In particular, there was one passage where Brockman contemplated a path forward that he said would be morally bankrupt for him to pursue.

在OpenAI律师的追问下，他说明，这条路径最终他们并没有选择。

Under questioning from OpenAI's lawyer, he explained that that was a path they never ended up taking.

那条路意味着把埃隆从OpenAI董事会踢出去，然后单独成立一家营利性子公司，不让他参与。

That was a path that would have involved firing Elon Musk from OpenAI's board and then creating a for profit subsidiary without him.

实际上，他们并没有解雇埃隆，而且在成立营利性子公司时，证词显示，布罗克曼和其他人一直在想办法让他加入，甚至考虑给他股份。

In fact, they didn't fire Elon, and when they did create a for profit subsidiary, the questioning revealed that Brockman and others were looking for ways to include Elon in the plan, were looking for ways to offer Elon equity.

根据布罗克曼的证词，只有当埃隆无法掌控这家营利性实体时，他才彻底放弃这个项目，离开了OpenAI。

Under Brockman's testimony, it was only once Elon was unable to have control over that for profit entity that Elon gave up on the project and left OpenAI.

埃隆一方声称，他最初本会有控制权，但这份控制权会随着新董事加入和新投资者进入而逐步减弱。

Elon's side says that he would have had control initially, but that control would have been intended to dissolve or or reduce over time as further board members were added and new investors were brought on board.

布罗克曼今天证词中最引人注目的地方之一，是他直言不讳地说，他认为埃隆并不懂人工智能。

One of the most interesting exchanges during Brockman's testimony today was that Brockman said he doesn't think Elon is very good at AI.

他说，埃隆懂汽车。

He said he knows cars.

他也懂火箭。

He knows rockets.

没错。

Sure.

但他对人工智能其实了解不多。

But he doesn't know too much about AI.

为了说明这一点，他举了个例子：当初有人给埃隆演示了后来支撑ChatGPT的早期技术时。

And to illustrate that, he gave an example of when Elon was given a demo of an early iteration of the technology that would later power ChatGPT.

布罗克曼说，团队其他人都看出了这项技术的潜力，但埃隆觉得这玩意儿挺蠢的。

And Brockman says while the rest of the team understood the promise of the technology, Elon thought it seemed kind of stupid.

他觉得网上随便一个小孩跟你聊天都比这强。

He thought kids talking to you on the Internet could do a better job.

他根本没当回事，结果研发这技术的科学家们都很受打击。

He was not impressed, and the researchers who worked on that technology ended up feeling disheartened.

所以布罗克曼说，这件事让他很早就意识到，就AI技术本身而言，他对埃隆的判断并不信服。

So Brockman said for him, this was an early illustration that he didn't really Elon's judgment on matters of AI as far as the technology goes.

这跟布罗克曼 earlier 提到的另一件事形成了对比：当时埃隆请了一群OpenAI的研究员去特斯拉，帮忙优化他们的技术和流程。

That was kind of contrasted with an anecdote that Brockman gave earlier in his testimony of a time when Elon asked a bunch of OpenAI researchers to come visit Tesla and improve some of the technology and the processes at Tesla.

在布罗克曼的描述里，伊利亚·苏茨克弗——也是OpenAI的联合创始人，被公认为顶尖AI科学家——去了特斯拉，给了他们一些建议：只要你们能收集一万个特斯拉自动驾驶经常出问题的烂路图像，我敢保证AI一定能解决这个问题。

In Brockman's telling, Ilya Sutskeber, who is also a cofounder of OpenAI and is regarded as a brilliant AI scientist, went to Tesla and gave them some encouraging words of wisdom, saying that if you can collect, say, 10,000 images of roads with poor conditions where Tesla cars currently struggle with self driving, then I guarantee you that AI will be able to handle the problem.

所以这多少说明了，真正对人工智能有天然直觉、熟悉它潜力的人是什么样子。

So that was kind of an illustration of what maybe it does look like to have a a natural intuition or a familiarity with, with AI and a sense of the promises that it has.

这就是布罗克曼告诉我们的。

So that was what we heard from Brockman.

接下来是马斯克诉奥特曼案。

Up next in Musk v.

接下来我们将听到西奥班·齐勒斯的证词，她是埃隆·马斯克四个孩子的母亲，也曾是OpenAI董事会成员。

Altman, we're expected to hear from Siobhan Zilles, who is a mother of four of Elon Musk's children, but was also on the OpenAI board.

问题是，她在董事会期间，究竟是恪守了职责，还是在私下向马斯克传递了内部信息。

The question for her is whether she was sort of observing and honoring her commitments on the board or whether she was also funneling information back to Musk in a personal capacity during that time.

接下来还是马斯克诉奥特曼案。

So that's what's up next in the Musk v.

的庭审。

Altman trial.

AI平台Tessera Labs宣布完成了一轮由安德森·霍洛维茨领投的6000万美元融资。

AI platform Tessera Labs announced $60,000,000 in a new funding round led by Andreessen Horowitz.

现在我旁边的是这家公司的首席执行官卡比尔·纳格雷查。

I'm joined now by the company's CEO, Kabir Nagrecha.

卡比尔，欢迎来到节目。

Kabir, welcome to the show.

很高兴你能来。

It's great to have you here.

能来我也很高兴。

Great to be here.

非常感谢，阿卡什。

Thanks so much, Akash.

好的。

Okay.

我很想聊聊这家公司，但得先问你一个问题。

So I am excited to talk about the company, but I have to ask you the question.

你13岁就上大学了？

You started college at age 13?

这到底是怎么做到的？

How does one even do that?

说来话长，经历挺曲折的，

It's a long story, long journey,

其实也没那么久吧。

I mean, it wasn't that long.

你就是从13岁到17岁，读了四年大学。

It was four you did the same four years.

你从13岁一直读到17岁。

Just you were 13 to 17 years old.

没错。

That's right.

四年本科，三年博士，加起来大概20岁就搞定了。

Four years of college, three years of PhD, ended up wrapping it all up around 20.

明白了。

Okay.

然后呢？

And what?

你是说，你12岁的时候就直接去申请了？

You just, like, you just walked you just applied at age 12?

后来我参加了一些加州的项目，跳过了高中，走了这条路。

Ended up skipping high school with some programs for it out in California, so went down that route.

明白了。

Okay.

这条路径我挺想以后多了解一下的。

Well, that's a route I would like to hear more about later on.

咱们聊聊你这周的新闻吧。

Let's talk about your news this week.

Tessera Labs 正在筹集六千万美元，这家公司是做什么的？这笔钱你打算怎么用？

So Tessera Labs is raising $60,000,000 What does the company do, and what do you plan to do with the money?

是的。

Yeah.

我们主要做的是围绕IT转型的人工智能解决方案。

So what we do is really AI around IT transformation work.

所有那些复杂的传统工作——比如ERP迁移、并购、整合，还有私募股权公司对投资组合企业的系统升级——这些全球范围内的IT转型任务，目前主要靠人工服务来完成。

All of the heavy legacy work around ERP migrations activities, mergers and acquisitions, consolidations, revamps of portfolio companies for private equity firms, IT transformation work all around the world, currently done primarily by services activity.

而我们想用人工智能来推动这些流程。

We instead try to drive via AI.

这笔资金将主要用于支持我们的扩张。

The money, it's really gonna help drive our expansion.

到目前为止进展不错，已经拿到了几家顶级财富500强企业作为首批客户，现在正准备拓展到更广泛的市场。

We've been on a good rip so far, getting our first few blue chip Fortune 500s, and now pushing out to that broader market.

我们非常感谢安德森霍洛维茨的支持与信任，他们领投了这轮融资，推动了这项业务的发展。

We really appreciate the support and confidence from Andreessen to help lead our round and drive this activity.

既然进展不错，那你们目前的营收大概是多少？

So on a good rip so far, how much revenue are you guys generating?

具体数字我不方便透露，但我可以说的是，我们瞄准的项目都是数十亿美元规模的。

I don't know if I can mention specific numbers, but what I can say is that the programs we're going after are multi billion dollar programs.

我见过最低的项目也有两亿美元，这些都是全球顶级的服务项目，而我们正尽力把成本降低好几个数量级。

I think the lowest one I've ever seen was $200,000,000 These are the heavyweight services engagements of the world, and we're doing what we can to cut down those costs by orders of magnitude.

你们融资时的估值是多少？

And what was the valuation that you guys raised at?

具体估值我不太方便透露，但可以肯定的是，它反映了我们到目前为止为这些企业创造的显著价值。

The valuation we raised at, I'm not sure if I'm able to disclose, but it reflects the substantial value we've been able to bring to those enterprises so far.

很好。

Great.

我想再稍微了解一下。

I just want to understand this a little bit.

你们说的企业软件转型，是指那种老式的数字化转型吗？比如公司把IT系统从本地部署迁移到云端？

When you talk about these software transformations going on in enterprises, I mean, are we talking the old school digital transformations, like companies trying to get their IT systems from on prem to the cloud?

那种转变确实发生过。

Because there was certainly that transition.

而现在，AI带来的变革完全是另一回事。

Now there's the whole transition of using AI, which is a whole separate story.

但我的理解是对的吧？你们其实还是在专注过去二十年里常见的那种传统数字化转型？

But am I correct in understanding that you are actually focused on sort of the more traditional digital transformations that we have been doing for the last twenty years?

没错，正是如此。

Very much so.

我们普遍认为，很多所谓的数字化转型其实早就完成了，早就结束了。

I think we tend to believe that a lot of these digital transformations have already come to fruition, have already ended.

我可以肯定的是，从本地部署迁移到云端这个过程，现在依然在进行中。

What I can say is even the on prem to cloud transition is very much still alive.

我们依然看到很多客户在使用本地的数据中心。

We still see it in many, many customers out there that they're running on prem data centers.

他们的整个业务还在依赖八九十年代的大型机系统运行。

Their entire businesses are still being run through mainframes through systems from the eighties, the nineties.

我们总在谈论AI转型、转向AI应用，但现实是——

The reality is, we talk about this AI transition and moving to AI usage.

很多时候，这个AI的推进被卡在了第一步：先完成那场基础的数字化转型。

Oftentimes, this is gatekept and bottlenecked behind making that first digital transformation.

所以我们关注的不只是旅程终点的AI，而是把AI融入整个数字化转型过程，来解决AI落地前的那些基础问题。

So our view is not just focusing on the AI at the end of the journey, but actually bringing AI into that digital transformation journey to solve the prerequisites for AI.

那你们具体是怎么用AI来推动从本地部署到云端的转型呢？

And so how are you using AI then tactically to help with that on prem to cloud transformation?

我们主要是帮客户解决那些传统上耗时极长的环节，原本可能要花六个月、八个月，甚至好几年才能搞定的事。

We're essentially helping to solve out all of those traditionally long pole items, the ones that would have taken six, eight months, multiple years to drive.

比如代码重构、数据清理、新系统的配置，还有识别业务流程中的缺口。

Code transformation, data cleanup activities, configuration of new systems, identification of business process gaps.

如果我是一家《财富》500强企业的负责人，我该怎么确保供应链绝对不中断？

If I'm running a Fortune 500, how do I ensure my supply chain is never disrupted?

我又该怎么保障制造运营的稳定性？

How do I ensure resiliency of my manufacturing operations?

IT系统是这些企业运转的根基。

IT is the foundation for these businesses.

如果IT系统撑不住，代码有问题，数据不准确，其他一切都无从谈起。

And if the IT systems aren't there to support that, if your code isn't right, if your data isn't right, nothing else can be done.

我们正是想在这里创造价值。

That's where we're trying to bring the value in.

那你为什么会对这个问题这么着迷呢？

And why did you get so addicted to this particular problem?

当年你才十七岁，刚上大学，现在当然不是了，但你依然是最年轻的创始人之一。

Of all the problems 17 that year old college grad you're not 17 now, but you're one of the younger founders out there.

那为什么偏偏选这个问题？

So why this problem?

这可是个老问题了。

This is like an ancient problem.

我的意思是，你本可以去解决任何其他问题。

I mean, you could have solved anything.

其实我最早接触这个领域的时候，年纪还很小。

Well, my first exposure to the space actually came pretty early.

我父亲在IT转型和数字化转型领域干了很多年，我从小就在听那些高管会议，亲眼见证了整个过程。

My father used to work in the IT transformation, digital transformation world for many, many years, and I grew up listening to the steering committee meetings and seeing what it's like firsthand, essentially.

后来，当我真正深入这个领域后，说实话，看到自己能为真实企业带来的价值，真的让人上瘾。

Later on, once I got into the space hands on myself, I mean, it is is addicting in some ways to see the value you can bring to real companies is what I would say.

比如我看到一罐可口可乐，或者路上一辆货运卡车，我就会想，嘿。

When I see a can of Coca Cola, let's say, or a shipping truck out there, I like to think, hey.

你知道那辆货运卡车吗？

You know that shipping truck?

它配送路线的规划，是我们帮忙搭建的系统做的。

Its route was configured by some kind of IT application we helped set.

那罐可口可乐的生产流程，也是我们参与优化的。

That Coca Cola can flow through a manufacturing system we helped optimize.

我脑子里经常想着这些事。

Those are the ideas I like to have in mind.

顺便说一句，这些都只是举例，我可没提任何客户的名称。

By the way, those are all just examples there, not any customer names I'm listing.

对。

Right.

最后一个问题给你。

So, last question for you.

那些传统上帮助企业从本地部署转向云端、推动数字化转型的系统集成商和咨询公司，

The system integrators, the consulting firms that traditionally have helped with this on prem to cloud transition, digital transformation.

现在它们正从企业那里拿到大笔钱，去帮它们完成向人工智能的转型。

Now they're getting all the money from companies to help with the AI transition.

那你们为什么不做咨询公司呢？

Why are you not a consulting firm?

在我看来，像财富500强这样的客户，已经不再需要第三方来介入或主导他们的AI战略了。

To me, customers, Fortune 500s, they're not looking for third party engagement and third party ownership of their AI strategy anymore.

财富500强的首席信息官们，更希望把这件事掌握在自己手里。

Fortune 500s, CIOs, they want the leverage of owning this in house.

他们想自己掌控IT中心，想拥有自己的AI平台，让团队在里面受训，推动转型，真正掌握自己的AI发展路径，而不是交给外部咨询公司来代管。

They want to own their IT centers, they want to own the AI platform that they can train their folks on to drive these transformation activities, to have that leverage, to own their own AI journey, not to have it managed by a consulting party, third party.

对。

Right.