OpenAI与博通的芯片交易遭遇180亿美元融资障碍，微软削减Copilot冗余功能

欢迎各位来到《The Information》的TITV节目。

Welcome everyone to The Information's TITV.

我是阿卡什·帕什里卡。

My name is Akash Pasricha.

今天是5月8日，星期五。

It is Friday, May 8.

在正式开始之前，《The Information》独家报道指出，DeepSeek计划融资超过70亿美元，这将成为中国AI公司有史以来规模最大的一轮融资。

Before we get into the show, The Information has published exclusive reporting that DeepSeek plans to raise more than $7,000,000,000 in what would amount to the largest ever funding round by a Chinese AI company.

根据我们的报道，这笔融资正推动DeepSeek加速发布新模型，并加快其盈利计划的落地。

According to the information's reporting, the funding round is prompting DeepSeek to expedite both its model releases and also some of its plans to start generating revenue.

报道中还透露了下一代模型预计发布时间的细节。

The story also has details on when the next model is expected to be released.

建议大家去我们的网站上仔细阅读这篇报道。

I encourage you to check it out on our website.

今天节目的第一部分，《The Information》独家披露了OpenAI与博通在AI芯片上的合作，以及这项合作可能面临高达180亿美元的变数。之后我们会转向微软的记者，聊聊他们Copilot战略的调整。

First up on the show today, the information has exclusive reporting about OpenAI's AI chip deal with Broadcom and how it could hit a snag to the tune of $18,000,000,000 We'll then turn to our Microsoft reporter about the company's shifting Copilot strategy.

我们还会回顾一下ServiceNow本周的大型会议，看看它反映了企业软件买家的整体趋势。

We also have a recap of ServiceNow's big conference this week and what it tells us about enterprise software buyers at large.

我们会从火箭德鲁那里获得每日更新，他一直在法院旁听埃隆·马斯克对OpenAI的诉讼，最后我们会邀请编辑们来深入讨论这场诉讼，分析它对整个AI领域意味着什么。

We'll get our daily update from Rocket Drew, who has been at the courthouse for the Elon Musk OpenAI trial, and we'll end the show by bringing on our editors for a broader discussion on that trial, discussing the implications of that courtroom saga on the AI story at large.

今天这期节目会很精彩，咱们马上开始。

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

OpenAI和博通正在合作开发一款芯片，而融资是这个故事中最关键的部分之一。

OpenAI and Broadcom are working on a chip together, and financing is one of the most important parts of this story.

我的同事阿尼萨·加尔迪齐有最新报道，提到这个项目可能在融资方面遇到障碍。

My colleague, Anissa Gardizy, has some reporting on a possible hurdle that the project could hit in that respect.

我要不要请她上线，跟我们说说她了解到的情况？

Want I to bring her on to share with us what she knows.

阿尼萨，欢迎再次做客节目。

Anissa, welcome back to the show.

很高兴你今天能来。

It's great to have you here.

你能说说你了解到的关于OpenAI和博通这个芯片融资进展的情况吗？

Tell us a little bit about what you found with respect to how financing is going for this OpenAI Broadcom chip.

目前OpenAI和博通正处于一场相当紧张的谈判中，因为OpenAI希望把它的首款定制芯片交给台积电投产。

OpenAI and Broadcom are in the middle of pretty tense negotiations right now as OpenAI hopes to send its first custom chip to production at TSMC.

现在他们卡在一个关键点上：双方还没能就这笔180亿美元交易的供应商融资条款达成一致。

They're at a little bit of a roadblock right now because they're trying to agree on vendor financing terms for this initial $18,000,000,000 deal and that hasn't been figured out yet.

尽管这两家公司去年十月就宣布了合作，但融资问题一直没敲定，现在他们来回拉扯，赶着在博通必须通知台积电预留产能之前做出决定，好让芯片能赶在明年投产。

So even though these two companies announced their partnership last October, they did not have the financing figured out and they're both going back and forth, you know, kind of running at a deadline to when Broadcom needs to alert TSMC to save capacity for OpenAI's chip so that it's ready next year.

明白了。

Okay.

那我们来逐条梳理一下这些情况。

So let's unpack each of those elements here.

首先，再帮我们回顾一下，这款芯片到底要实现什么功能？

First, just remind us, what is this chip supposed to do exactly?

OpenAI正在设计一款定制芯片，专门用来运行其模型的推理任务。

OpenAI is designing a custom chip to help it run inference of OpenAI models.

关键是，如果OpenAI设计一款专门针对ChatGPT和其他产品运行方式的芯片，它的效率可能会超过英伟达的通用GPU。

And the idea is that if OpenAI designs a chip that is, you know, specifically modeled to how OpenAI runs ChatGPT and other products, it could be more efficient than a general purpose GPU from NVIDIA.

我们知道，博通本来就是做定制ASIC的。

So this is I mean, we know that Broadcom is in the business of custom ASICs.

他们为很多公司做过这类芯片，包括Meta。

They do these for many companies, including Meta.

这次是为OpenAI的工作负载定制的芯片，叫Jalapeno芯片。

This would be a custom chip for OpenAI's workload, and it's called jalapeno chip.

是这样吗？

Is that right?

对。

Yes.

第一款叫Jalapeno芯片，下一款叫Serano。

The first one is called the jalapeno chip, and then Serano is the next chip.

他们这个命名方式还挺有意思的。

So they're following an interesting naming convention.

没错，这在所有AI公司里都是常态。

Right, as is the case with all companies in AI.

所以这正是这个芯片要实现的功能。

Okay, so that's what the chip is supposed to do.

那你能说说，当时那个资金支持协议具体是怎么安排的吗？

Now tell us a little bit about what the vendor financing agreement in place was actually looking like?

我的意思是，这个安排实际是怎么运作的？

I mean, how would this have actually worked?

过去几周，各家公司在讨论的就是这个情况。

So here's what the companies have been discussing over the past couple weeks.

博通一直要求OpenAI每出一美元，他们就配套投入一美元来资助芯片研发。

Broadcom had been asking OpenAI to put up a dollar for every dollar Broadcom was going to finance the chip with.

也就是说，他们希望双方的投资比例是1比1。

So they were looking for a one to one ratio of investment.

但OpenAI其实希望博通能比自己投入更多前期资金。

But OpenAI actually wanted Broadcom to invest more upfront than itself.

最终，OpenAI 和 Broadcom 达成一致，由 Broadcom 提供更多前期资金。OpenAI 的高管在一份我们拿到的内部备忘录中表示，这对他们来说是个重大胜利。

And so OpenAI and Broadcom were able to agree that Broadcom would put up more capital upfront which OpenAI executives, said in an internal memo that we got that that was a huge win for them.

但这里有个前提：为了促成这笔交易，Broadcom 要求 OpenAI 确保微软会购买未来生产的 40% 芯片。

But there is a caveat there because, you know, to get this deal done Broadcom is asking OpenAI to ensure that Microsoft actually will buy 40% of the chips that are going to be produced.

正如我们报道的那样，OpenAI 在内部措辞非常谨慎，他们实际上是在说：我们还没确认微软一定会买 40% 的芯片，但我们可以先告诉 Broadcom，微软有这个意向。

And OpenAI, as we reported was being very careful in its wording internally and they're basically saying, look, we haven't confirmed that Microsoft is gonna buy 40% of the chip, but let's just tell Broadcom that Microsoft intends to do that.

但私下里，OpenAI 的高管很清楚，他们没法逼着微软必须买这些芯片。一旦微软不兑现承诺，他们就得赶紧找别的买家或资金方，因为不能再指望 Broadcom 了。

But behind the scenes, OpenAI executives know that they can't require Microsoft to buy the chip and that if Microsoft does not, you know, follow through, they're going to be stuck scrambling for another buyer or another person, a party to finance the chip because they might not be able to rely on Broadcom.

所以，微软到底会不会买这些芯片，现在成了一个巨大的悬念。

So there's a huge tension with whether Microsoft is going to buy the chip or not.

我想稍后再聊微软这部分，咱们先回头看看这个 1:1 的比例问题。

So I want to get to the Microsoft piece in a minute here, but let's go back to the one to one ratio here.

一开始，OpenAI 说，你们 Broadcom 投一美元，我们也投一美元。但后来的协议调整了，Broadcom 实际投入的比这个比例还要多。

So OpenAI initially said, We'll put in $1 if Broadcom, you put in $1 Then Broadcom Well, the revised agreement is that Broadcom actually puts more than that one to one ratio in.

为了澄清一下，是不是意味着 OpenAI 之后会分期偿还 Broadcom 的这笔投资？

And just to clarify here, is the idea that OpenAI would then pay back Broadcom over time for its investment?

没错。

Exactly.

正是如此。

Exactly.

这是供应商融资，博通会收取贷款利息。

It's vendor financing and Broadcom would interest on the loan.

所以这就是它的运作方式。

So that's how that would work.

明白了。

Okay.

那我们来谈谈可能和微软的这笔交易吧。

So then let's go to this possible deal, I guess, with Microsoft.

博通说，你们得确保这40%的芯片都由微软买走。

So Broadcom says, you you've got to make sure that 40% of these chips get bought by Microsoft.

根据你们的报道，这个协议具体是怎么说的？

And so according to your reporting, that agreement is what?

这只是一个口头协议，微软到目前为止并没有真正承诺这一点？

It's a handshake agreement and Microsoft hasn't really committed to that so far?

微软并没有承诺购买40%的芯片，私下里，OpenAI的高管们也讨论过，在没有微软任何正式承诺的情况下签署这样的协议所面临的风险。

Microsoft has not committed to buy 40% of the chip and privately OpenAI executives have discussed the risk involved in signing such an agreement without anything from from Microsoft.

他们私下里说，Broadcom并没有充分意识到这笔交易背后的风险。

And they say they're saying behind the scenes that Broadcom does not fully appreciate the risk involved in this deal.

风险就在于，万一微软不履约，OpenAI就得赶紧想个备选方案。

The risk being that Microsoft doesn't fall through Microsoft falls through and OpenAI has to come up with a plan B.

你觉得这有多大概率会发生？我知道很难量化，但既然他们都在谈这种可能性，说明这风险确实很真实。

How likely does it seem you know, if we're it's hard to put a number on it, but I guess, I mean, fact that they're acknowledging the possibility here suggests that it is a very real risk.

咱们往前推一推，看看情况。

I mean, let's move it forward a bit.

如果微软不参与，这笔交易会怎样？

What would happen then to this deal if Microsoft didn't come to the table?

那整个芯片交易会不会就此泡汤？

I mean, is this whole chip deal in jeopardy then?

我们的意思是这样吗？

Is that what we're saying?

这40%是因为OpenAI希望博通承担整个初期阶段的融资。

40% of it would be because OpenAI wants Broadcom to finance the entire initial phase.

金额高达180亿美元，容量大约13亿瓦。

That's $18,000,000,000 and around 1.3 gigawatts of capacity.

而且OpenAI希望这部分由别人来融资。

And OpenAI wants it to be financed.

所以，如果没有博通的融资，这部分交易很可能无法推进。

And so it probably wouldn't move forward with that portion of the deal without financing from Broadcom.

如果拿不到博通的融资，他们就得另找人来出资。

And if it doesn't get the financing from Broadcom, it's gonna need to find someone else to finance it.

但目前还不清楚到底有谁会愿意接手这件事。

But it's unclear who who exactly would really want to do that.

明白了。

Okay.

那TSMC这边呢？

Now what about the TSMC angle here?

OpenAI是怎么拿到TSMC的产能的？

How are they how is OpenAI getting capacity at TSMC?

所有和博通合作定制芯片的公司，都依赖博通和TSMC的关系来争取制造芯片所需的产能。

So all of the companies that work with Broadcom on Broadcom on custom chips rely on Broadcom's relationship with TSMC to get that capacity to actually manufacture the chip.

所以现在OpenAI和博通正试图达成一项附条件的协议，让博通能有信心，这个项目会推进下去，从而向TSMC下订单。

And so right now OpenAI and Broadcom are trying to reach a conditional agreement so that Broadcom can be feel confident that the program is going move forward and it can place that order at TSMC.

现在两家公司都在赶这个截止日期，好让芯片能在2027年如期投产，避免因为博通没能锁定TSMC的产能而进一步延误。

So that is the deadline that the companies are racing to reach right now so that the chip can be ready in 2027 and not face a further delay because Broadcom couldn't secure the right capacity with TSMC.

但博通不会在没有明确协议的情况下这么做，所以他们现在就在想办法解决这个问题。

But Broadcom's not going to do that without a solid agreement in place, so that is what they're trying to figure out.

对。

Right.

因为现在的产能实在太紧张了。

Because that capacity is very scarce right now.

如果博通要分配任何产能给别的项目，那他们必须确定这事儿值得做。

If Broadcom is going to allocate any of its capacity to anything, I mean, they have to know that it's going be worth it.

不然的话，他们早就把产能用在别的业务上了。

Otherwise, they'd allocate it for another part of its business.

安妮莎，我最后一个问题：咱们不妨退一步，整体看看这个芯片合作项目。

The last question I have for you, Anissa, is So let's just take a step back and look at this chip deal at large.

就像我们之前说的，OpenAI 的目标非常宏大，涉及很多不同的产品方向，对吧？

I mean, OpenAI, as we've talked about, it's pursuing a vast ambition across many different product vectors, right?

有软件，有数据中心，现在又加上了芯片。

There's the software, there's the data centers, and now there's the chip.

OpenAI 真的觉得自己需要自研芯片吗？

Mean, does OpenAI think that it needs its own chip?

这在他们的战略里是核心部分吗？是未来赚钱的关键吗？

Does this feel core to the strategy here and core to the ambition to eventually start generating cash?

还是说，这只是一块锦上添花的业务？

Or is it just a nice to have?

我认为，任何一家大型云服务消费者，一旦成为云支出排名前十的公司，就会开始考虑：我们是不是该自己开发硬件？毕竟一大笔钱都花在了硬件供应商身上。

I think every large consumer of cloud capacity once they start consuming once they're like one of the top 10 companies spending on cloud capacity they start to think should we develop our own hardware because a lot of our money is going to our hardware vendors.

亚马逊、谷歌、苹果都已经这么做了。

So Amazon has done this, Google has done this, Apple has done this.

所以OpenAI希望制定一套内部策略，既能降低成本，又能在与英伟达、AMD、AWS、Cerebras以及其他公司的合作谈判中掌握更多主动权。

And so OpenAI wants to make sure it has an in house strategy, to potentially lower its cost and give it negotiating leverage as it's signing deals with NVIDIA, AMD, AWS, Cerebras and all of the other companies.

他们其实早就开始布局这块芯片了，时间可以追溯到两年前以上。

And they've been working on this chip effort for quite a while, you know, dating back more than two years.

这款芯片对他们的整体算力战略至关重要。

And so this chip is central to their overall compute strategy.

非得明年就完成不可吗？我觉得不一定。

Does it have to happen next year?

但如果他们不得不大幅缩减这个项目，那确实会是个重大打击，毕竟他们已经和博通签了10吉瓦的订单。

I don't think so but it would be a major blow if they had to, you know, curtail this in a major way because they already announced a 10 gigawatt deal with Broadcom.

所以我们接下来会密切关注，他们到底能不能兑现这个承诺，还是说这又只是OpenAI众多宣布了却没真正落实的计划之一。

And so we'll be closely tracking if they rise to that level or if this is one of the many announcements OpenAI makes that it doesn't actually end up following through on.

没错。

Right.

安妮莎，非常感谢你来做客。

Well, Anissa, want to thank you for coming on.

这位是安妮莎·加迪兹，我们《信息》杂志负责云和计算的记者。

That is Anissa Gardizy, our cloud and compute reporter here at The Information.

微软正在重新评估其Copilot产品的战略。

Microsoft is reassessing its strategy on its Copilot products.

我的同事亚伦·霍姆斯对这件事做了深入报道。

My colleague Aaron Holmes had some inside reporting on that story.

我想请他来跟我们分享一下他了解到的情况。

I want to bring him on to share with us what he knows.

亚伦，欢迎来到节目。

Aaron, welcome to the show.

很高兴你再次回来。

It's great to have you back.

很高兴能来。

Happy to be here.

好的。

Okay.

微软和Copilot这边现在是什么情况？

What's going on in Microsoft and Copilot land?

我们看到的是全面铺开。

So what we're seeing is across the board.

简单说，自从微软在2023年拿到OpenAI的模型后，就开始到处塞Copilot，能加的地方全加了。

Basically, know, as soon as Microsoft got its hands on OpenAI's models in 2023, the company started basically putting Copilot's everywhere it could.

你瞧，Office和Windows里突然都有了Copilot，安全软件、Azure，几乎微软卖的每款软件都裹上了它。

You know there was suddenly Copilot's in office and Windows, but also in its security software and in Azure and essentially every single piece of software that Microsoft was selling.

他们还开始往每个程序里加一个Copilot聊天侧边栏。

It started adding, you know, a Copilot chatbot sidebar.

最近，不少客户反馈说，Copilot太多，有时候真让人烦，也用不上。

More recently, you know, there's been pretty heavy feedback from customers who find the number of Copilots sometimes annoying or unnecessary.

我认为微软对这种反馈很敏感，因为过去几个月，他们已经撤回了Windows中一些弹出的Copilot功能，就在上周或这周。

And I think Microsoft has shown that it's sensitive to that feedback because, in the last couple of months we've seen it roll back some of the copilots that show up in Windows and just last week or this week.

抱歉，Xbox的首席执行官阿卡什·帕什里卡说，他们打算移除Xbox PC应用里的游戏Copilot。

Sorry, Akash Pasricha, the CEO of Xbox, said that they were going to remove the gaming Copilot that lived in the Xbox PC app.

我觉得微软总体上意识到，现在大家对AI有点疲劳了，他们想专注在真正有人想用的AI功能上，而不是把AI硬塞进所有产品里，指望它能提升一切。

And I think the general idea is that Microsoft is kind of aware that there's this AI fatigue and wants to focus on actual products that people want to use AI for rather than simply adding it to everything and hoping it lifts all ships.

对。

Right.

你之前在报道里提到过一个客户的说法。

And you had this quote from a customer in your story.

你说有些客户觉得这些Copilot根本没用，压根就不碰它们。

You said some customers find these copilots to be functionally useless, meaning like literally, like, don't use any of it.

是的。

Yeah.

我接触过一位负责公司采购微软软件的首席信息官，他说他们测试过Copilot，结果发现有人让它总结一页文档，它却给出了两页的总结。

I spoke to one CIO who was in charge of buying Microsoft software for his company, and he said that they were testing Copilot, and some of their tests showed that you know they would ask it to summarize a one page document and it would return a two page summary.

还有人抱怨说，他们想用Copilot分析安全数据，结果它却自信满满地编造信息。

Somebody else was complaining that you know they were trying to use Copilot to analyze security data and it would just make stuff up with confidence.

然后他就问：‘你这个数据是从哪儿来的？’

And then you know, he would say, like, where did you get this data from?

Copilot居然回答：‘你说得对，是我瞎编的。’

And Copilot would say, you're right, I made it up.

这些只是几个具体的例子。

So I mean, those are just a few specific examples.

我认为微软正在积极改进Copilot，特别是在企业应用场景上。

And I think Microsoft is actively working on improving Copilot in especially the sort of enterprise use cases.

我们听说它的表现已经好很多了，但与此同时，他们也在撤回一些不必要的Copilot功能，先把它用在真正有用的地方。

We've heard it's gotten a lot better, but I think at the same time, they're kind of pulling back on unnecessary use cases for Copilot while they try to make it more useful.

如果他们正在撤回之前推出的一些Copilot功能，那我们之前聊过的定价策略，他们有没有调整呢？

So if they're pulling back some of the Copilot tools that they had rolled out I mean, we've talked about pricing, you and I, and how they're handling that part of the product.

他们有没有在改价格？

Are they changing the pricing at all?

换句话说，你们是在拿掉产品的一些功能。

Mean, conceivably, you're taking some parts of the product away.

我知道这些功能根本没人用，但从客户的角度来看，你可能会想：原本有十个Copilot功能，现在少了四个，为什么我还得付同样的钱？

I know it was never used, but from the customer's end, I imagine you could say, Well, if there were 10 Copilot features and four are gone, why am I paying the same price?

是的，我们看到微软整个产品线都在重新考虑运行Copilot的成本问题。

Yeah, I think what we're seeing across the entire, you know, Microsoft portfolio is that the company is starting to think more carefully about, you know, how much it costs to run Copilot.

最近几周被取消的那些功能，基本都是免费的Copilot，出现在Windows、Xbox这些消费级应用里，用户本来就没为这些功能额外付费。

Specifically, the ones that were rolled back in the last couple of weeks were all essentially free copilots that showed up in Windows or Xbox or these consumer applications where people were not paying extra for copilot.

与此同时，他们也开始提高那些更高级的Copilot产品的价格。

At the same time, have seen them start to increase prices for the more sophisticated copilots.

比如GitHub Copilot，运行成本很高，因为它用的是Anthropic和OpenAI的先进模型，微软不得不专门提高价格，针对那些由公司付费购买席位的开发者。

You know, GitHub Copilot, for example, was basically costing a lot to run because it uses advanced models from anthropic and OpenAI, and Microsoft was forced to hike the prices of that specifically for, you know, the software developers whose employers are paying for seats.

总的来说，微软似乎越来越谨慎了，开始更仔细地权衡运行这些AI工具的毛利问题。

So we are starting to see sort of across the board, I think Microsoft get a little bit more careful about how it manages the gross margins of running all of these AI tools.

这正是我想说的，如果那些臃肿的Copilot功能没带来额外收入，但计算资源的开销还在继续。

That's what I was going to say, that if they weren't incurring any additional revenue for those bloated Copilot features, then they're still spending on compute.

这样的话，对公司的利润会有正面影响。

So this would have beneficial repercussions then for its bottom line.

对。

Yeah.

我觉得现在有两件事在同时发生。

And I think like there's kind of two things happening.

他们也在探索更多使用低成本AI模型的方法，特别是针对那些比较基础的Copilot使用场景。

You know, they're also figuring out more ways to use cheaper AI models for especially the kind of more rudimentary Copilot use cases.

我听说他们会继续把免费版Copilot迁移到更轻量的模型上，这样能减少计算资源消耗。

And I'm hearing that you know they're going to continue to move the free versions of Copilot onto these smaller models that don't take up as much compute.

但与此同时，你知道吗？微软上周在财报中明确提到，GitHub Copilot的运行成本实在太高，严重拖累了云业务的毛利率。

But at the same time, know how Microsoft disclosed actually last week in earnings that specifically GitHub Copilot was so expensive to run that it was dragging down gross margins in their cloud unit.

我想这正是他们不得不调整定价、让客户支付更高费用的原因。

And that, I think, is why they had to basically rejigger the prices there and and make it a little bit more expensive for customers to use.

很好。

Great.

好了，亚伦，非常感谢你来做客。

Well, Aaron, want to thank you for coming on.

这位是亚伦·霍姆斯，我们《信息》杂志负责微软报道的记者。

That is Aaron Holmes, our Microsoft reporter, here at The Information.

这周，ServiceNow在拉斯维加斯举办了他们的年度盛会。

ServiceNow hosted its flagship conference in Las Vegas this week.

这家公司如今成了企业软件板块整体压力的风向标。

The company has become proxy for the pressure against enterprise software stocks at large.

接下来我请来UpperEdge公司软件合同谈判领域的负责人亚当·曼斯菲尔德，听听他从这次活动里观察到的第一手情况。

I want to bring on Adam Mansfield, a practice lead at software contract negotiation firm UpperEdge, to give us a view from the ground on what he learned from the event this week.

亚当，欢迎再次做客我们的节目。

Adam, welcome back to the show.

很高兴你能来。

It's great to have you here.

我也很高兴能来。

Great to be here.

确实很喜欢这种对话。

Certainly enjoy these kind of conversations for sure.

好的。

Okay.

所以是 Knowledge 2026。

So Knowledge twenty twenty six.

这是 ServiceNow 在拉斯维加斯举办的大型会议。

This was the Big Service Now conference in Vegas.

你今年去了。

You were there this year.

去年你也都去了。

You were also there last year.

今年公司传达的信息，跟去年有什么不同？

Tell me about how the message this year differed from last year from the company.

这是个好问题。

Yeah, mean, it's a great question.

我觉得区别在于，今年完全是AI主导，而去年是大部分内容涉及AI，对吧？

I would tell you it differed in the sense that there was maybe all AI, and where last time it was most AI, right?

所以这里还是有一点区别的。

So there's a little distinction there.

不过，当然也有一些共同点。

But you know, there was definitely some commonality.

我觉得最引人注目的，就算你当时有点走神也能注意到，那就是AI控制塔这个主题，几乎占了主题演讲前四十五分钟的内容，当然我可能有点夸张了。

I think the thing that really jumped off, and if you were maybe half asleep, would still have noticed this, which is AI control tower was pretty much the first forty five minutes, and maybe I'm exaggerating, of the keynote.

而且当时非常普遍，到处都是，对吧？

And it was prevalent, and it was all over the place, right?

那安全方面呢？

What about security?

那安全方面呢？

What about security?

这是他们关注的重点吗？

Was that a focus for them?

是的，安全是其中非常重要的一部分，而且这也和控制中心密切相关。

Yeah, security was a big part of that, and that ties to Control Tower as well.

他们必须重点关注这一点，因为客户显然很在意，也在向ServiceNow以及其他供应商施压。

I think they have to obviously focus there because their customer base is obviously paying attention to that, and they're pushing back on ServiceNow, as well as other vendors as well.

这可不是ServiceNow一家的事。

It's not like just the ServiceNow.

一提到AI，大家自然就会想到安全，想到那些正在考虑升级的机构。

And when you think AI, you think security, you think about these entities that are considering the jump in.

成本结构、运行方式这些因素当然一直都很重要，但安全永远是第一位的。

Cost profile, how it works, things like that always are going to be important, but security always comes to the top.

对吧？

Right?

那ServiceNow怎么才能在这方面领先呢？

So how does ServiceNow get ahead of that?

他们该怎么推销自己？

How do they sell themselves?

没错。

Right.

整个会议期间都是这样。

That was throughout the conference.

好的，明白了。

So, okay.

所以这就是他们的宣传重点。

So that was the pitch.

现在咱们聊聊你和客户的交流吧。

Now let's talk about your conversations with customers.

客户对这个宣传有什么反应？

How they react how do they react to the pitch?

我有个关键问题想问你，亚当：你看他们的股价，投资者对ServiceNow确实有顾虑。

And the big question I have for you, Adam, is I mean, look, you look at the stock price, investors have concerns about ServiceNow.

尽管财务状况依然相当强劲，至少比去年好多了。

Despite the financial profile still being pretty strong, certainly compared to where it was last year.

客户也是这么想的吗？他们怎么看？

Are customers feeling the same way, or what's their read?

是的。

Yeah.

是这样的。

So here's the thing.

对我来说幸运的是，我能和这些客户坦诚交流。

So the fortunate thing for me is I get to have candid conversations with these customers.

我和他们建立了关系。

I've built relationships there.

实际情况是这样。

So here's the deal.

我觉得他们确实有兴趣。

I think there's genuine interest.

他们理解ServiceNow的整体愿景和目标。

I think they understand the big picture and the story where ServiceNow is trying to bring these customers.

但问题就在这儿。

But here's the problem.

我接触的大多数客户一开始都会这么说。

Most of these customers that I talk to lead with this.

我仍然需要弄清楚这到底是什么东西。

I really still need to understand what this actually is.

不管是叫AI控制塔还是分析平台，名字不重要。

So AI control tower or analysis, doesn't matter what it is.

我得先把它搞明白。

I need to understand it better.

他们遇到的困难是，当想深入讨论时，告诉我的是，ServiceNow在很多方面似乎还在摸索中。

And the problem they're having is when they try to get into those deep dive discussions, what they're telling me is, it feels like ServiceNow in a lot of ways is still figuring this out.

他们还在摸索，但这不是批评。

They're still, and that's not a negative comment.

想想看，我们现在才刚起步。

Think about how early we are in this game.

这就让人停顿了一下。

And so that creates pause.

让他们对兴趣产生了一点阻力，对吧？

It creates a little bit of friction on their interest, right?

我要告诉你的第二点是，当他们思考自己的AI产品时，这当然没问题。

The second thing I will tell you is that when they think about their AI offerings, that's all well and good.

但总会出现这种情况。

But here's what always happens.

那我 already 采用的产品怎么办？

What about the products I've already adopted?

那我明明在花几百万美元在上面的那些呢？

What about the, or should be using, spending millions of dollars on?

我的核心ITSM、核心CSM怎么办？

What about my core ITSM, my core CSM?

我的核心HRSD又怎么办？

What about my core HRSD?

那既然已经用了这些产品，为什么不先想办法让它们发挥更大价值，而不是非得加新功能或者被推着接受什么AI打包方案呢？

What about getting more value out of those without needing to add on or be pushed into some new packaging that is AI embedded?

对吧？

Right?

AI这块我明白，但我的重点其实是怎么把现有的东西用得更充分。

The AI component, I get it to a degree, but I'm so focused on getting more out of what I already have.

你的精力到底放在哪儿了？

Like where is your time?

你把钱花在研发上，到底是为了什么？

Where are you spending your, my money on the R and D to that?

这些在大会上哪儿看得见？

And where is that in this conference?

你们安排了哪场讨论？

What session do you have?

主题演讲里提到过吗？

Where was it on the keynote?

这正是我 constantly 被问到的问题——我有没有答案。

And that's the thing I'm- literally, that's what I'm being asked all the time, is if I have the answer.

但这个问题确实很好。

I But it's a great question.

另外让我想到的是，由于创新速度太快，你们不断推出新产品，所有企业都在强调‘这是新功能’、‘这是新功能’。

Well, and the other thing that comes to mind is, look, because of the pace of innovation and the new products that you can offer to customers, I mean, look, the focus from all these businesses is here's what's new, here's what's new.

客户根本来不及消化，你们三个月前刚推出的东西，现在又出新的了。

You know, customers don't even have time to You just came out with this other thing three months ago.

我怀疑是不是因为你们总推最新最好的，而我还在琢磨你们两个半月前发布的东西呢。

I wonder if that's the lag, is that just because you're offering the latest and greatest, I'm still focused on understanding the thing that you put out to market two and half months ago.

完全没错。

100%.

这其实挺有意思的，对吧？

So it's really interesting, right?

我觉得这不光是ServiceNow的问题，很多其他供应商也都这样。

I think, and it's not just ServiceNow, it's a lot of other vendors.

你知道的，我刚才一直在听，关于Copilot这部分，我其实也能聊很多。

As you know, I was just listening in, the Copilot piece, I could talk a lot about that too certainly.

但关键是，这些厂商都在升级他们的AI功能，归根结底，他们不得不这么做。

But the point as these vendors, all of them, enhance their AI offerings, they're doing that because they ultimately have to.

而且大家对这个确实很感兴趣。

And there's interest in it.

问题在于，同一个公司内部，比如ServiceNow，就有太多不同的产品了。

The problem is, is that there's many within the same, ServiceNow has many.

所以当你参加一场主题演讲或研讨会时，他们会不断提到什么自主工作流、员工助手、Now Assist、AI控制塔、AI、AI、工作平台，一大堆术语。

So when you sit and listen to a keynote in a session, they say things autonomous workflow, employee words, Now Assist, AI control tower, AI, AI, Work Fabric, all this stuff.

信息量实在太大了。

It's a lot.

然后客户就被问：这玩意儿对我们来说具体怎么用？

Then the customer is asked, can we talk about how it would work for them?

客户一听就懵了：等等，这些功能真能整合到一块儿吗？

And they're like, woah, woah, is that all gonna come together?

更别提了，我自己的项目里还用着其他AI工具。

And not to mention, I have other AIs on my initiative.

Copilot、Anthropic，随便挑一个都行。

Copilot, Anthropic, pick your favorite one.

你却让我从这一堆乱七八糟的东西里理出头绪。

And so you're asking me to look through all of this noise.

那ServiceNow，你们怎么跟别人区分开呢？

So ServiceNow, how are you distinging?

你们打算怎么跟别人不一样？

How are you gonna distinguish yourselves from it?

问题就在于，当他们真有机会展示的时候——而且这不光是他们的问题。

Then the problem becomes when they give the opportunity, and again, it's not just them.

当机会来了，产品团队、销售团队一进来，

But when they're given that opportunity, and then the people come in, the product teams, the sales teams, and

他们其实并不真正

they're not really

清晰明了，这才是问题所在。

crisp, that's a problem.

更别说，授权机制也让人一头雾水。

On top of it, licensing component is also confusing.

这就形成了一个恶性循环，带来困扰，同时我又在其他地方花了上百万。

And so it creates this cycle, which then creates anguish, which then says, but I'm spending millions on other things too.

他们因为推出太多没经过充分验证的功能，反而伤了自己，想抓住创新反而适得其反。

So they're almost hurting themselves by putting so much out there that hasn't been fully vetted to catch innovation.

这还干扰了客户，让他们在执行ServiceNow期望的操作时处处碰壁。

But that's also disrupting customers in the sense that it's creating friction in these customers to actually go do what ServiceNow wants.

上次请你们来的时候，你们提到希望这些公司能更透明地说明，他们到底能在前六个月实现什么。

So last time we had you on, you talked about how you needed more transparency from these companies insofar as what they can achieve in the

前六个月。

first six months.

那是你们当时提出的要求。

That was something you asked for.

这可不是六个月前的事了。

Look, that wasn't six months ago.

那大概是两个月前的事。

That was more like two months ago.

但你有没有注意到，我谈的不只是ServiceNow。

But have you seen I'm not just talking about ServiceNow.

我是在说所有这些公司。

I'm talking about all these companies.

现在它们对客户在前六个月能实现什么，有更透明了吗？

Are they any more transparent now with respect to what customers can achieve in the first six months?

多是相对的，对吧？

More is relative, right?

如果你从零开始，现在到了二，那确实算多了。

So if you start start at zero, and you get to two, like I guess it's more.

不过，答案是肯定的。

But yes, the answer is yes.

但我想问你的是，他们现在达到应有的水平了吗？

But what I would tell you is, are they where they need to be?

没有。

No.

对吧？

Right?

我给你举个例子。

So I'll give you an example.

我接触的大多数客户，不管是ServiceNow还是别的公司，都很明确。

Most of the customers that I talk to, whether it's ServiceNow or others, and they are very specific.

我希望你们ServiceNow进来，帮我搞清楚这五件事——到底哪里变好了，我需要掌握哪些信息才能做决策。

I need you, ServiceNow, to come in, here are the five things I need to understand, how it's gotten better, how I, what I need to know and to make decisions.

他们现在有没有更清楚一点？

Are they coming in with a little bit more clarity?

确实有进步，但整体还是显得很笼统，没针对他们具体的情况。

Sure, but it still feels very much being figured out general, not specific to them.

这对我有什么影响？

How is it going to impact me?

我的投资回报率是多少？什么时候能看到？

What are my ROIs and how soon?

你说 rollout 要一百天，McDermott 还到处宣传这些数字。

You say a hundred days to roll out and McDermott's excited and throwing those numbers out there.

但说真的，这可能吗？

But the point is, really?

这到底是怎么实现的？

Like, how does that actually happen?

要是没做到呢？

What if it doesn't?

你们能给我哪些具体的承诺？

What commitments are you gonna give me?

他们对实际内容说得很少，反而把那些我都不想说的营销包装吹得天花乱坠。

So they're very light on the actual meat of the matter, and they're very enhanced on the, and I hate to say this, the fluffer marketing piece.

我觉得他们把这层包装抬得太高了。

I think they've gotten that elevated.

所以事情到底是怎么发生的？我一再说，他们为了争取到这些会议，拼命往里挤，结果却忘了：进了会议室，你得说些能让客户产生共鸣的话。

And so what's happening again, and I keep saying this, it's almost as if, as they've tried to push themselves in the room to get these meetings, they're forgetting about when you're in the meetings, you actually have to say something that resonates with the customer.

你得真正展示产品，得拿出路线图，得说明它到底能带来什么价值，这才是问题的关键。

You actually have to present a product, you have to show a roadmap, you have to show how it's going to be valuable, and that's the disconnect.

还不如放慢脚步，先专注把现有的东西做好，边做边完善，再慢慢推进。

It'd be better off to slow down, focus on what they already have, build as you do it, and then push.

还有其他

And there's other

事情。让我问你一下，

things So let me ask that you this.

根据你和客户的交流，

You're Based on your conversations

你看看

with customers, you look at

ServiceNow的股价跌了，去年跌了差不多一半吧？

the stock price, ServiceNow, it's down, I don't know what it is, 50% in the last year, something like that?

根据你听到的消息，你觉得这股价是不是被低估了？

Do you think it's oversold based on what you're hearing?

从我的角度来看，我可以这么说：

Here's what I can tell you, right, from my lens.

关于ServiceNow，我可以肯定地告诉你，我接触的所有客户，没有一个跟我说要撤掉这个系统。

The thing that ServiceNow I can definitively tell you is that all these customers that I work with, I have no customers that are coming to me and saying, Hey, I'm taking it out.

他们没打算退出。

I'm moving off.

这不在我的规划里。

It's not on my roadmap.

我也没考虑再增加使用。

I'm not considering doing more.

Yes AI、Control Tower，还有Assist Pro Plus，这些现在还不行，也许以后会考虑。

A Yes AI, Control Tower, now Assist Pro Plus, yes, not now, maybe later.

就像这种事要是真存在的话。

Like if that exists.

但其他我负责的厂商就不一样了，咱们之前也聊过，我确实听到客户说打算退出Marketing Cloud。

Whereas with other vendors that I cover, and you and I have talked about this, I am having conversations of, I think I'm going to move off of Marketing Cloud.

我觉得可能会从Tableau转到Power BI。

I think I might move from Tableau to Power BI.

这种事正在发生。

Like that's happening.

你大概能猜到我说的是哪家厂商，但ServiceNow这边，我真没听过这种话。

And you could probably know what vendor I'm talking about, but with ServiceNow, I don't have that.

客户一般拿到平台后，都能看到它的价值，也理解它的好处。

So they generally get the platform, they see the value, they understand it.

当然总有小问题，但还没到让人想彻底放弃的地步。

There's always issues, but it's not to the point where it's like, I'm out.

更常见的是：我想要更多我 already 在用的功能，因为我真的很喜欢。

It's more like, give me more of what I already have because I like it.

别一下子推得太快。

Stop pushing more too fast.

我跟你说这些，是因为我觉得现在到了一个关键点，市场可能期待他们提供更多东西。

So when I the reason I tell you that is, I think it's an inflection point where they need the market probably wants more from them.

客户还在摸索他们到底需要什么，真正会用哪些功能。

The customer- They're figuring out what they need, yeah, what they're actually going to use.

没错。

Exactly.

我觉得，说到底，客户现在依然和ServiceNow站在一起，还会继续用下去。

I think, you know, listen, I think part of this is customers are still there with them, and they're going to keep going with ServiceNow.

只是他们扩展的范围和速度，会不会赶上Salesforce希望他们达到的程度而已。

It's just a matter of how wide they're going to go as fast as Salesforce and ServiceNow rather wants them to.

在你走之前，我最后问你一个问题。

Let me ask you one last question before you go.

我们之前聊到的另一件事是，一些企业软件公司现在对员工访问他们所存储的数据开始设限了。

So the other thing that we've been covering here is the tolls that some of these enterprise software companies are now putting on agents accessing the data that they are sitting on, storing.

这还是客户的数据。

Again, it's customer data.

大家一直在争论，客户是不是永远拥有自己的数据，对吧？

There's a big debate around customers will always we own the data, right?

这周我们在《应用人工智能》通讯里提到过，ServiceNow 现在已经开始推出这类工具了。

But we covered this this week in our Applied AI newsletter about ServiceNow is one of the companies that is putting those tools up now.

说实话，这家公司的理由是，你用得越多，我们就收得越多。

And look, the argument the company makes is that you're using the product more, so we're going to charge more.

就这么回事。

That's what it is.

你的客户对这些工具有什么反馈？

What are your clients telling you about those tools?

他们是怎么反应的？

What's their reaction

是的。

Yeah.

这真是个好问题。

To here's the It's a really good question.

我尽量简洁地回答，但这个问题有两个方面。

So I'll answer as crisply as I can, but there's two pieces to it.

首先，我觉得客户现在对模型到底是什么了解得更多了，对吧？

The first is, I think customers have become more informed about what the model actually is, right?

ServiceNow 主要是混合模式，他们自己在财报电话会上也这么说。

So ServiceNow is predominantly hybrid based, and they'll say this on the earnings calls themselves.

你以为你买的是一个单位，但实际上你用的是多次辅助服务，或者消耗了一定数量的令牌，对吧？

So you think it's unit, but then you get a number of assists or you get a number of tokens, a number, right, you draw down on.

你其实想说的是，这种按使用量计费的部分。

So that's what you're getting at, the consumption based component of this.

问题是，客户一看就犯嘀咕：我不太清楚，或者担心以后的开销会失控。

The problem is, is that customers look at this and go, I don't really know, or I'm concerned about future spend.

这完全取决于一个计费指标，可我对这个指标的具体依据也说不准。

That's just going to be based on a meter, which again, I don't really know the foundations of the meter itself.