Block裁员40%，奈飞放弃收购华纳兄弟，Anthropic与五角大楼对峙，亚马逊与OpenAI达成500亿美元交易

欢迎各位来到Informations TITB。

Welcome everyone to the Informations TITB.

我叫阿卡什·帕什拉查。

My name is Akash Pashracha.

今天是2月27日，星期五。

It is Friday, February 27.

今天我们为大家准备了一场内容丰富的节目。

We have got a jam packed show for you today.

我们将拆解代码块、40%的裁员，以及尝试回答这个问题：是AI的原因，是臃肿的原因，还是两者都有？

We are unpacking blocks, 40% layoffs, and we're gonna try to get an answer to the question, was it AI, was it bloat, or was it both?

我们会邀请一位分析师和一位投资者来讨论这个问题。

We'll bring on an analyst and an investor to talk about that.

我们还将深入探讨今天早上新披露的关于OpenAI高达1100亿美元融资的细节。

We'll also get to the new details that have emerged this morning about OpenAI's $110,000,000,000 funding round.

这将对亚马逊和微软产生重大影响，我们也会就此展开分析，同时独家报道谷歌与Meta价值数十亿美元的AI芯片合作项目。

This has big implications for Amazon and Microsoft that we're going to get to, along with our exclusive reporting about a Google and Meta AI chip deal that is worth billions of dollars.

接下来，我们将邀请两位资深编辑，深入解析Netflix退出华纳兄弟探索公司竞购战的决定。

We'll then bring on two of our top editors to break down Netflix's move to drop out of the Warner Bros.

以及本周其他重大新闻头条。

Discovery bidding war and the other big headlines of the week.

我们还将进一步探讨答案引擎优化领域，并与该领域最新的独角兽公司进行对话。

We're also wading deeper into the world of answer engine optimization and chat with the newest unicorn on that block.

最后，我将与开放爪基金会首位独立董事会成员戴夫·摩尔坐下来，探讨该项目的下一步计划，以及他打算如何履行其监督职责。

And finally, I am sitting down with Dave Moore, the first independent board member of the Open Claw Foundation to talk about what is next for the project and how he plans to approach his oversight role.

这将是一场精彩的节目，让我们直接开始吧。

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

拥有Square、Afterpay和Cash App的Block公司，昨晚宣布将裁减其约1万名员工中的40%，这一消息震动了硅谷。

Block, the company that owns Square and Afterpay and Cash App, shook Silicon Valley with its news last night that it is laying off roughly 40% of its 10,000 employees.

首席执行官杰克·多尔西就他的决定发表了一篇长文，但许多人仍在争论这些裁员的根本原因是什么，以及其中有多少是人工智能的影响。

CEO Jack Dorsey made a long post about his decision, but a lot of people are still debating what the root causes of these layoffs are and how much of it is AI.

我想邀请沃尔夫研究公司的董事总经理达伦·皮勒，以及海勒豪斯管理合伙人马塞洛·利马，来谈谈他们对这一切的看法。

I wanna bring on Darren Pillar, managing director at Wolfe Research, and Marcelo Lima, managing partner of Heller House, for their thoughts on all of this.

两位先生，欢迎你们的到来。

Gentlemen, welcome to you both.

很高兴你们能来。

It's great to have you here.

谢谢，阿卡什。

Thanks, Akash.

谢谢。

Thank you.

达伦，我想先从你开始。

Darren, I want to start with you.

你对这个决定怎么看？

What did you make of this decision?

说实话，我们一直认为这家公司运营得比应有的效率要低一些。

Look, I think for some time we've been saying this company was operating a little bit more inefficiently than they should have been.

我们实际上在2025年5月就发布过一封致董事会的公开信，差不多一年前了，指出他们在多个方面支出过高。

We actually published an open letter to the board back in May 2025, so almost a year ago already, saying there's a lot of areas they're overspending.

如果你看看他们的研发支出，比如，他们在研发上的投入占毛利润的百分比比行业平均水平高出约11个百分点。

If you look at how much in R and D for example, they're about 11 percentage points higher than the industry in how much they spend on R and D as a percentage of gross profit.

因此，从我们的角度来看，当时确实需要进行一次必要的削减。

And so from our perspective, there was an overdue cut needed.

我们会进一步讨论这个问题，我敢肯定马塞洛也有类似的看法要分享。

You know, we'll talk more about this and I'm sure Marcelo has similar views to share.

但考虑到他们裁减的员工数量，正常的调整轨迹本应让他们现在拥有大约7000到8000名全职员工，到2025年可能只有7000人。

But considering the amount of heads they're cutting, a normal trajectory would have allowed them to be about 7,000 to 8,000 FTEs right now, maybe 7,000 by the '25.

他们正在从10000人削减到6000人。

They're cutting down from 10,000 to 6,000.

我们估计，大部分裁员都是为了消除长期以来管理不善造成的冗余和浪费。

We estimate most of that is coming from just overall bloat and expense that was mismanaged.

而现在，他们又因人工智能的推动再削减约1000人。

And now they're catching up with another maybe 1,000 being AI driven.

马塞洛，你对此有什么看法？

Marcello, what are your thoughts here?

是的，我同意达伦的观点。

Yeah, I agree with Darren.

我多年来一直关注这家公司，并在2022年、2023年和2024年与管理团队沟通，指出他们的利润率过低。

I've been following this company for many years and have been communicating with the management team in 2022, 2023, and 2024 that their margins were way too low.

因此，如果你看看全球同行的情况，以及他们产生的毛利润中有多少能转化为营业利润率，即使在此次裁员和2026年的指引之后，Block的业绩仍然严重落后。

So if you look at global comps and how much they generate in gross profits and how much they generate, how much of that drops down to the operating margin line, Block is still very much underperforming, even after this reduction in force in their guidance for 2026.

马塞洛，当你与管理层沟通时，他们的回应是什么？

What was the response from management, Marcello, when you tried communicating with them?

你们是股东，但你们卖掉了股票。

You were shareholders, you sold your shares.

他们给了你们什么答复？

What was the answer that they gave you?

很遗憾，基本上是沉默。

Mostly crickets, unfortunately.

所以

So

互动非常少。

very little engagement.

在我看来，杰克是个杰出的产品专家，显然很有远见，但长期关注这家公司的人都知道，它和推特实际上并不是为股东利益而运营的。

It seems to me, Jack is a brilliant product guy, obviously visionary, but people who have been following this company know that it was and Twitter, frankly, was not really run for the benefit of shareholders.

再多谈谈这一点。

Talk more about that.

你是什么意思？

What do you mean?

正如你所知，埃隆来了之后裁掉了推特80%的员工，还发现了一些奇怪的事情，比如旧金山的餐饮费用高达每餐400美元，但根本没人来吃。

Well, as you know, Elon came and laid off 80% of Twitter staff and found a bunch of things like the food, the catering in San Francisco was $400 a plate because nobody showed up.

他们把家具存放在甲级写字楼里。

They had furniture stored in class A office buildings.

还有许多关于他们订阅的SaaS产品的故事，这些产品他们根本从未登录使用过。

Those are a number of stories of SaaS products that they had signed up for and never actually even logged into.

所以他们在为所有没用的软件付费。

So they were paying for all the software they weren't using.

而且我认为最近在Block举办的一场派对，花费高达六千八百万美元左右。

And I think there was a party at Block now recently that was what, 68,000,000 or something like that.

所以杰克实际上为他的员工举办了一场派对，然后解雇了其中40%的人。

So Jack in effect threw a party for his staff and then fired 40% of it.

我可以提一个问题吗？

Could So jump in a question.

好的，达伦，请说。

Yeah, Darren, go ahead.

我再补充一点。

I'll just add on to that.

你们知道，当我们发表那封致董事会和杰克的公开信，指出他们可以找到一些效率提升空间时。

You know, when we published that open letter to the board and to Jack basically saying there's areas of efficiencies you can find.

他们对我们的回应是：我们的工程师薪资高于普通公司。

Their response to us was, look, we have engineers that are more expensive than the average company.

我们做了一个图表，收录在我们发布的报告中，显示在所有我们覆盖的金融科技和支付公司中，他们在人均运营支出上排名第二。

We put a chart together, which, is in a note we published that shows that of all of the names in FinTech and payments that at least we cover, they're the second highest spender on an OpEx per employee basis.

他们平均每位员工的支出超过55万美元，大约是该行业中位数的两倍。

They spend over $550,000 on an average per employee, which is somewhere around two times the median for the space.

他们的回应是：这些员工薪资很高，他们是工程师。

And their response was They're expensive employees, they're engineers.

但你知道，处理高成本人力有很多方法，比如离岸外包，现在或许还可以借助人工智能。

But, you know, there's a lot of ways to handle high cost headcount, specifically offshoring and maybe now with AI as well.

所以，达伦，我们来谈谈人工智能这个方面。

So, Darren, let's talk about the AI element here.

这就是争论的焦点。

This is the debate.

是人工智能吗？

Was it AI?

是人员冗余吗？

Was it bloat?

还是两者都有？

Is it both?

我觉得我的观点居中，也许AI能帮他们达成一些目标，但确实存在过度招聘的问题。

I think I sort of land somewhere in the middle where maybe AI could have gotten you somewhere, but there was all this over hiring.

不过，达伦，你知道吗，这或许回到你之前说的，如果公司按正常轨迹发展的话。

But, Darren, you know, maybe this goes back to what you were saying, the normal trajectory if the company was going along this.

你知道，如果用AI裁员，会是什么样子？

You know, what would an AI cut have looked like?

这对公司能节省多少成本？

How much savings could that have been for the company?

我的意思是，他在推文中给出了这个数字，杰克，说我们目标是每人创造200万美元的毛利润。

I mean, he's putting out this figure, Jack, in his tweet saying we're targeting $2,000,000 per person in gross profit.

这就是我们追求的目标。

That's the target we're going towards.

你知道，AI能帮他们实现这个目标吗？

You know, was AI going to get them there?

你怎么看？

What do you think?

我认为这一点我之前稍微提到过，但我觉得，如果一家正常成熟的公司按照其正常的人力增长轨迹发展，到今天员工人数应该在7000人左右，而不是现在的10000人，这意味着在总共4000人的裁员中，大约有3000人是因为公司本应更高效运营，正如马塞洛所暗示的那样。

I think that, and I did touch on this a little bit before, but I do think that the trajectory of their headcount growth in a normal mature company matter would have gotten them to around somewhere in the ballpark of around 7,000 FTEs by today, instead of the 10,000 that they have now, suggesting that let's call it three of the 4,000 cuts really were because the company should have been more efficient, as Marcelo was alluding to all, you know, as well.

我认为那额外的1000人，也就是使裁员总数达到4000人的部分，很可能与公司现在试图用人工智能解决的一些低效问题有关。

And I think that the final 1,000, the incremental 1,000 above that, getting them to the 4,000 they're cutting, probably is related to some inefficiencies they're now fighting using AI.

我们将看到这会对业务产生影响。

We'll see this impacts the business.

他们现在已经在测试了，无论是他们自己的Goose产品、项目，还是正在内部开发的Autobot或其他机器人，都是为了提供帮助。

They've been testing this now, whether it's their own Goose products, or programs, or it's, you know, Autobot or bots they're working on right now internally, to help.

但归根结底，我认为其中三分之二到四分之三的裁员原因都是过度支出造成的。

But when it comes down to it, think it was probably two thirds to three quarters related to overspending.

马塞拉，你认为这会为其他公司树立先例，导致它们也进行如此大规模的裁员吗？

Marcella, do you think this sets any kind of a precedent for other companies to make similar cuts of this size?

我的意思是，达伦刚刚为我们详细分析了他的数据，对吧？

By that, I mean So Darren just walked us through his numbers, right?

所以，事实是一回事，但叙事又是另一回事，对吧？

And so, I mean, there is the truth of the matter, but then there's the narrative, right?

这意味着公司可以进行这些裁员。

Is that companies can do these cuts.

Marcella，你认为其他公司会采纳这种说法吗？

Marcella, do you think other companies will latch onto this narrative?

我认为他们已经在这么做了。

I think they already are.

我最近看到C3AI和eBay都在进行裁员。

Think I saw recently C3AI and eBay doing reductions in force.

我认为现实是，如果你听经济学家谈论工作，他们会说工作是由一系列任务组成的，而人们会做许多不同的任务。

I think the reality is, if you listen to how economists talk about jobs, they are bundles of tasks, And so people do a number of different tasks.

其中一些任务正被人工智能自动化。

Some of those tasks are being automated by AI.

这一点毫无疑问。

There's no question about that.

我整天都在使用Cloud Code。

I use Cloud Code all day.

我用云功能在Excel里。

I use Cloud in Excel.

太棒了。

It's fantastic.

它真的大大加快了我的工作进度。

It really accelerates my work.

但这意味着我可以做以前没做的其他类型的任务。

But that means I can do other types of tasks that I wasn't doing before.

我可以加快其他一些任务，更高效地完成它们。

I can accelerate some of my other tasks and do them more efficiently.

我认为，知识经济和白领领域中的每个人现在都必须利用这些工具，提升自己的生产力。

Everybody, I think, in the knowledge economy and the white collar space has to now harness these tools and improve their productivity.

否则，我认为公司现在就有了政治上的借口来实施这类裁员。

Otherwise, I think companies will have a now political cover to do these types of reductions in force.

达伦，你怎么看？

Darren, what do you think?

我认为这取决于你所谈论的公司类型。

I think it depends on the types of companies that you're talking about.

事实上，我认为会发现一些效率提升。

Truth is I think there are gonna be efficiencies found.

正如马塞洛所说，我们看到其他一些公司虽然没有达到如此大规模的裁员，但已经指出能在支出上节省低个位数到高个位数百分比的费用。

We're seeing it, as Marcelo said, in other names we talk about that are already calling out nothing to the degree of this headcount reduction, but, you know, call it anywhere from low single to high single digit percentages of expenses being found, where they can take it out.

话虽如此，我也认为，这种效率提升还帮助这些公司在竞争中与其他公司拉平了差距。

That said, I also think that, you know, the efficiencies that it's driving is also allowing these companies to level the playing field from a competitive standpoint with many others.

老实说，这还得看行业。

Frankly, depends on the industry though.

支付和金融科技领域——比如Block所在的行业——将从节省成本中受益。

Payments and fintech, which where Block sits, it's gonna benefit from saving money.

但我并不认为这个领域会出现太多颠覆，因为涉及大量监管资金流动，这些是AI难以替代的。

But I don't see a lot of disruption in terms of the actual category given how much regulatory money movement is required, that really can't be replaced by AI.

因此，市场实际上把许多行业一概而论，全都打压了一番。

And so the market's really kind of beaten up quite a few of the sectors all as one.

我认为应该更具体一些，指出支付领域中一些公司被过度抛售了。

And I think it should be a little more specific than that, with some names within payments having been oversold and everything.

达伦，简短问一下，昨晚最大的新闻就是裁员。

Darren, very quickly, I mean, this was the big news last night, reduction in headcount.

显然，他们也发布了季度业绩。

Obviously, they released their quarterly results as well.

我想请你快速谈谈Square、Afterpay和Cash App这三项业务各自的状况。

And I just want to get a quick picture from you on how Square, Afterpay and Cash App, how each of those businesses are doing.

如果你看他们的总收入或净利润，基本上已经持平了。

I mean, if you look at their top or their net revenue, that's basically flatlined.

毛利润却在上升。

The gross profit has been coming up.

帮我理解一下，像这样的公司怎么可能做到这一点？

Help me understand how that's possible for a company like

这样。

this.

是的

Yeah.

你得记住，他们的净收入包含了加密货币交易，而且有大量的转嫁成本。

You got to remember their net revenue incorporates crypto trading, and there's a lot of pass through costs.

由于加密货币交易量下降，这将影响净收入。

And with volume on crypto down, that's going to impact net revenue.

但当你看毛利润时，这才是我们真正衡量他们主营业务的方式。

But by the time you get to gross profit, that's really their true way we look at their top line.

而毛利润一直增长良好。

And that's been growing well.

所以你看，Cash App 的增长超过了 30%。

So look, Cash App grew over 30%.

这其中很大一部分要归功于 Cash App 借贷业务。

A lot of that has to do with Cash App Borrower.

他们提供了更多的信贷，但在用户增长方面依然表现非常出色。

They're giving more credit out, but it's still doing very well in terms of user growth.

越来越多的人正在使用他们以及其他像Chime这样的数字银行服务。

And more and more people are using them and other companies like Chime for neobanking.

Square的商户业务方面，他们新增了不少销售人员，这些人员将用于重新投资之前通过AI或整体人力节省下来的资源，以推动销售和市场拓展。

And the Square, the merchant side, they've been adding quite a few salespeople, which is where they're gonna reinvest some of these AI savings or just headcount savings overall into sales and go to market.

这项业务正在加速发展，预计交易量将增长12%，高于第一季度的10%。

That business is accelerating and it's expected to grow volume by 12%, up from 10% in Q1.

因此，至少在收入端，他们整体表现相当不错。

So they're showing pretty good traction overall on the top line at least.

达伦，在这次调整之后，你对利润端有多少信心？

And any confidence on the bottom line after this move, Darren?

听我说，目前投资者向我们抛出了大量关于潜在增长幅度的数据。

Listen, there's a lot of numbers that investors are throwing out to us right now in terms of the magnitude of upside.

他们实际上将净利润目标，即运营收入，提高了5亿美元，从27亿美元提升至约32亿美元，利润率现在约为25%，而去年大约为20%。

They effectively raised their bottom line, their operating income targets by 500,000,000 So it went from 2,700,000,000.0 to $3,200,000,000 about a 25% margin now, up from what was around a 20% type margin last year.

实际情况是，如果他们在这些人员成本上节省得更多，还可能存在更大的增长空间。

The reality is there could be more upside depending on how much they're paying a lot of these heads.

到2026年底甚至进入2027年时，你可能会看到比现在高出五亿到六亿美元，甚至高达十亿美元的数字。

You could see numbers that are $500 to $600,000,000 even a billion dollars better than that by the time we get to the end of '26 and into '27.

因此，基于我们之前讨论的定价原因，以及整体收入增长的原因，这种情况确实可能发生，也理应发生。

So there really could be and should be for that reason, for pricing reasons we've talked about, and for just overall top line growth reasons.

马塞洛，

Marcello,

要让你重新投资这家公司，需要什么条件？

what would it take for you to get back into the company?

你现在会考虑买入吗？

Would you at all consider buying now?

关键问题是，他们是否真的醒悟了，并改变了对创造股东价值的看法？

You know, the big question is, have they really found religion and changed their thinking on creating shareholder value?

另一个问题是，这家公司的最终估值倍数是多少？

And then the other question is, what's the terminal multiple of this business?

差不多两年前的今天，我给他们发了一封邮件，当时我在推特上发布了这封邮件，我用25倍的市盈率作为我的退出倍数，并注意到，如果他们能实现这些每股收益目标，从现在到那时的内部收益率是多少。

I sent an email to them almost two years ago to the date which I posted on Twitter, and I was using a 25x earnings multiple for my exit multiple, and I noticed this is what the IRR is between here and then, assuming that you deliver on these EPS numbers.

现在，他们为2026年所承诺的数字，略低于我之前在2026年高预期中所设定的水平。

Now, what they've promised for 2026 is just shy of what I had outlined in my high takes for 2026.

我当时预测的调整后每股收益是4美元。

I had $4 in adjusted EPS.

他们目前实现的，我认为是3.66美元。

They're delivering, I think, 3.66.

正如达伦所说，这里确实还有很大的空间可以进一步提升。

There is a lot more room, as Darren said, to expand that for sure.

但问题是，它的估值会参考贝宝的倍数，导致股价大幅下跌，还是会以更高的倍数交易？

But then the question is, does this trade at a PayPal multiple, in which case the stock could go a lot lower, or does it trade at a higher multiple?

我认为这是这里的关键问题。

I think that's a big question here.

那么，你认为呢？

And so what do you, what do you think?

这种改变是否足以扭转他们的理念，马塞洛？

Has the rel is it enough to change the religion, Marcello?

你认为呢？

What do you think?

我个人很难相信，这个为股东服务了这么多年的管理团队会突然改变作风。

Personally, I, I, I have a hard time believing that the management team that has been doing this for so many years to shareholders has suddenly changed their stripes.

我目前还不信。

I'm not buying it yet.

很好。

Great.

好了，感谢达伦和马塞洛的到来。

Well, Darren and Marcello, want to thank you for coming on.

这位是来自沃尔夫研究的达伦和来自海勒之家的马塞洛，欢迎来到TI TV。

That is Darren from Wolfe Research and Marcello from Heller House here on TI TV.

今天早上，OpenAI宣布了其110亿美元的融资轮次。

OpenAI announced its 110,000,000,000 funding round this morning.

这一消息我们曾在《信息》上做过广泛报道。

That was news we have reported on extensively at The Information.

这个故事更有趣的部分在于，OpenAI 与参与本轮融资的不同云服务商所达成的云协议的具体构成。

The more interesting part of the story is the makeup of the different cloud agreements that OpenAI has struck with the different cloud providers that are participating in the funding round.

现在邀请我的是我们的云与计算记者安妮莎·加尔德西，以及我们的微软记者亚伦·霍姆斯，来讨论这一切。

Joining me now to discuss all of this is Anissa Gardesi, our cloud and compute reporter, and Aaron Holmes, our Microsoft reporter.

欢迎你们两位。

Welcome to you both.

很高兴你们再次回来。

It's great to have you back.

谢谢。

Thank you.

好的。

Okay.

亚伦，我想先从你开始。

Aaron, I want to start with you.

今天早上我们得到了一个大新闻。

We got big news this morning.

我们有融资轮次，有云协议，涉及OpenAI、微软和亚马逊。

We've got the funding round, we've got the cloud agreements, there's OpenAI, Microsoft, Amazon.

给我们详细解释一下。

Break it all down for us.

我们需要了解哪些关键信息？

What do we need to know?

是的。

Yeah.

所以，今天的大新闻是，OpenAI和亚马逊宣布，亚马逊将向OpenAI投资500亿美元，这是总额达1100亿美元的更大融资轮的一部分，其中还包括英伟达和软银的参与。

So I mean, today, the the big news is that OpenAI and Amazon announced that Amazon will be committing $50,000,000,000 in funding to OpenAI, is part of this bigger 110,000,000,000 funding round that also, you know, has NVIDIA and SoftBank participating.

特别是，我们看到亚马逊和OpenAI的关系正变得更加紧密。

And specifically, we're seeing that Amazon and OpenAI as part of this are getting even closer.

他们宣布OpenAI将使用亚马逊的Trainium芯片来训练其模型。

They announced that OpenAI is going to be using Amazon's Trainium chips for training its models.

他们还宣布，双方正在合作开发一系列新产品，这些产品将带有OpenAI品牌，并运行在亚马逊云上，这与以往OpenAI仅在微软Azure云上提供服务的情况略有不同。

And they also announced that there's going to be a bunch of new products that they're working on together that will have, you know, OpenAI branding, and that will run on Amazon's cloud, which is a bit of a departure from, you know, the status quo up to now, in which OpenAI has only been available on Microsoft Azure cloud.

这实际上如何运作的细节相当复杂。

The the nuances of how that's actually going to work are pretty complicated.

而且这些模型本身仍然运行在Azure云上。

And the models themselves are still running on Azure's cloud.

但如今会有OpenAI产品出现在Amazon上，这一点在某种程度上对Amazon来说是个重大胜利，尤其是从市场营销的角度来看。

But the fact that there's now going to be, you know, OpenAI products on Amazon is, in some ways, a big win from Amazon, especially from a marketing perspective.

所以我不想深入所有细节，但有些部分可以谈谈。

So I don't want to get into all the details, but some of them.

所以模型运行在Azure上，但它们会带有AWS的品牌吗？

So the models are running on Azure, but they're going to be branded with AWS?

是的，这其实是个非常复杂的话题。

Yeah, so it's it's a very kind of complex topic.

本质上，微软在合同上拥有在Azure上运行OpenAI模型的权利，这些模型本身不能运行在其他云上。

Essentially, Microsoft contractually has the rights to run OpenAI's models on Azure, and the models themselves cannot run on any other cloud.

因此，我们看到的是亚马逊和OpenAI现在表示，他们将合作开发一些产品，这些产品看起来像是AI代理或应用程序，能够记住特定客户的上下文，比如他们的姓名、公司名称和数据，这些上下文信息将存储在AWS上。

So what we're looking at is Amazon and open AI are now saying they're going to collaborate on products that look like you know, AI agents or applications that could remember a specific customer's context, like their name and the name of their company, their data, that context will sit on AWS.

但当应用程序需要实际使用AI模型进行思考和执行AI模型的所有功能时，它会再次调用Azure，同时为Azure计费。

But when the application needs to actually use the AI model to think and to, know, do all of the things that AI models do, it'll be making calls back to Azure and essentially spinning the meter for Azure at the same time.

所以操作的不同部分分布在不同的云上。

So different parts of the operation are on different clouds.

我的问题想问一下亚伦，然后是阿尼萨，我会接着问你，你们觉得他们为什么和亚马逊达成这笔交易？

My question for you, Aaron, and then Anisa, I'm going to come to you, is why do you think they did this deal with Amazon?

我的意思是，这是个更好的方案吗？

I mean, is this a better offering?

他们是想分散风险吗？

You know, is it them trying to diversify their bets?

你觉得这里的逻辑是什么？

What do you think the logic here is?

所以，OpenAI需要从亚马逊获得一些东西。

So, OpenAI needs a few things from Amazon.

他们肯定需要亚马逊愿意提供的资金。

They definitely need the cash that Amazon is willing to contribute.

据那里的消息来源称，微软目前尚未决定是否参与这一轮投资。

Microsoft, meanwhile, still has not decided if it's going to participate in this round, according to a source there.

但他们需要资金，同时也需要更多的计算能力来训练和运行他们的模型。

But they need the cash, they also, you know, need more computing power to train and run their models.

而亚马逊将通过训练芯片提供这一算力。

And Amazon is going to provide that capacity with with training, training them chips.

而亚马逊从中获得的是，可以将OpenAI的标志放在AWS上，宣称我们是一家与OpenAI合作的云服务商，我们的云上运行着OpenAI的应用，即使模型本身并不在我们的云上。

And then what Amazon gets out of it is essentially, they get to, you know, put the OpenAI banner on AWS and say, we are a cloud provider that works with OpenAI, and we have OpenAI applications running on our cloud, even if the models themselves don't.

所以我认为，从这个角度看，各方都能从中获益。

So I think in that way, of the companies are getting something out of this.

所以，Anisa，你曾对亚马逊的芯片Tranium进行过广泛报道。

So Anisa, you've reported extensively on Tranium, which is Amazon's chip.

我想知道，这是否是对Tranium能力的一种认可？

And I'm curious, is this a vote of confidence for what Tranium can do?

它在处理这些工作负载时是否比NVIDIA更出色？或者说，我们稍后会谈到TPU，但Trainium在这里有什么特别的优势吗？

Is it better at handling any of these workloads you know, NVIDIA or, I mean, we're going to talk about TPUs in a second, but is there something really good about Trainium here?

这就是亚马逊希望客户对Trainium芯片产生的想法，这也是我认为Trainium芯片成为此次公告核心部分的原因。

So that's exactly what Amazon wants customers to think about Trainium chips, which is why I think the Tranium chips were such a big part of this announcement.

我想让大家知道的是，OpenAI使用Trainium芯片，这从一开始就是亚马逊在谈判这笔对OpenAI的大额投资时就考虑在内的事情。

I think the thing for people to know about, you know, OpenAI using Tranium is that that was always on the table for Amazon as they were negotiating this large investment into OpenAI.

实际上，当我们首次报道亚马逊可能向OpenAI投资超过100亿美元时，我们就已经知道Trainium芯片是当时讨论的一部分。

Actually when we first reported about Amazon potentially investing more than $10,000,000,000 into OpenAI, we knew back then that Trainium chips were part of the conversation.

所以我认为，安迪·贾西在对OpenAI进行如此重大的战略投资时，不可能不包含Trainium的成分。

So I mean, I think there was no way that Andy Jassy was going to make a large strategic investment into OpenAI and not have a Tranium component.

当云服务商宣布重大云交易时，让客户使用其定制芯片是非常常见的做法。

This is very common for cloud providers to do when they announce big cloud deals is get the customer to use their custom silicon.

现在，英伟达也参与了这笔庞大的OpenAI融资，OpenAI将在亚马逊的云上大量使用英伟达的GPU。

Now NVIDIA also is in this giant OpenAI round and OpenAI will be using a lot of GPUs, from NVIDIA on Amazon's cloud.

因此，这将是检验Trainium芯片性能的一个良好机会，但我确实认为，从亚马逊的角度来看，这并不令人意外。

So, you know, it'll be a good test of the Trainium chips, but I do think that that is something that was not surprising from the Amazon side.

对。

Right.

换句话说，这不是一个非此即彼的选择，而更像是一种并存关系。Aaron，这对微软意味着什么？

In other words, it's not an or, it's more of an and type of Aaron, what does this mean for Microsoft now?

嗯，微软今天发布了一份声明，本质上是在提醒世界：我们是唯一能够为开发者提供OpenAI模型的云平台。

Well, saw Microsoft come out with a statement today essentially reminding the world like we are the only cloud that can serve OpenAI's models to developers.

这一点没有改变。

That hasn't changed.

他们还表示，即使亚马逊和OpenAI推出了基于AWS的OpenAI前沿应用版本，Azure上依然会有一个版本可用。

And also saying that, you know, even if Amazon and OpenAI build this AWS version of OpenAI's frontier app, there's still going be a version available on Azure.

所以我认为，从微软的角度来看，他们仍然能够抓住那些希望构建使用OpenAI模型的应用程序的开发者市场，这至今仍是他们一项非常盈利的数十亿美元业务。

So I think, you know, Microsoft from their perspective, they are still able to capture the market for developers who want to build applications that use OpenAI's models, which has been a very lucrative, you know, multibillion dollar business for them so far.

与此同时，他们确实失去了一点以往的独家光环——过去任何与OpenAI相关的东西，都必须通过Azure才能使用。

At the same time, they do lose a little bit of the sheen of exclusivity that, know, in the past, anything OpenAI related whatsoever, you had to go through Azure.

而现在，至少这部分技术栈可以部署在AWS上，只是模型本身不行。

And now you can have at least parts of that stack live on AWS, just not the models themselves.

因此，我认为这对他们来说是轻微的损失，主要体现在品牌层面，但他们依然能获得所有的模型收入。

So it's a slight loss from them, I think mostly from a branding perspective, but they do still get to capture all of that model revenue.

阿尼莎，我想转到另一个相关的故事。

Anisa, I want to pivot to a different story, a related story.

但你这周和我们在亚洲分局的同事报道了，Meta和谷歌达成了他们自己的数十亿美元交易。

But you reported this week, along with our colleagues in the Asia Bureau, that Meta and Google have signed their own multibillion dollar deal.

跟我们说说你发现了什么。

Tell us about what you found.

当然。

Sure.

所以，是的，当OpenAI正与亚马逊达成协议并讨论Tranium芯片时，我们这周报道了Meta与谷歌签署了一项巨额协议，Meta将使用谷歌的AI芯片，即张量处理单元（TPU）。

So yeah, while OpenAI is signing this deal with Amazon and talking about Tranium chips, we reported this week that Meta just signed a very large deal with Google in which Meta will be using Google's AI chips which are called tensor processing units or TPUs.

我们早就知道Meta和谷歌一直在就一项大型TPU协议进行谈判，而你也知道，Meta此前也与英伟达和AMD发布了重大合作消息。

So we had known that Meta and Google were talking for a while about a big TPU deal and as you know Meta also has done big announcements with Nvidia and AMD.

但这项协议特别有趣，因为谷歌一直试图将其TPU推向市场。

But this one is really interesting because Google has been trying to get its TPUs into the market.

他们已经拥有Anthropic这样的重要AI客户，现在Meta也将大规模使用TPU。

They already have Anthropic as a big AI customer, and now Meta is going to be using the TPUs in a big way.

你刚刚向我们解释了Trainium芯片的用途。

And now you just explained to us what the Trainium chip was good for.

在工作负载方面，TPU是否有什么特别擅长的地方，使得Meta会转向使用它们？

Is there something here on the workload side that TPUs are really good for that Meta would turn to them for?

人们对于TPU的主要兴奋点在于，它们是Nvidia的一个可行替代方案。

People are really excited about the TPUs mainly as a viable Nvidia alternative.

因此，大家对TPU在某些AI工作负载中可能比Nvidia显卡更便宜感到非常期待。

So there's been a lot of excitement around TPUs potentially being less expensive than Nvidia GPUs at certain AI workloads.

问题是，你知道，谷歌和Nvidia都面临供应限制，这显然是谷歌将TPU推向市场的一大障碍。

The thing is, you know, Google and Nvidia both have to deal with supply constraints so that is obviously, you know, a major constraint on Google's end in getting TPUs into the market.

但Meta在这里采取的是一种多芯片策略。

But Meta is kind of taking a multi chip strategy here.

所以它们并不只依赖Nvidia，显然通过与AMD的合作，它们也在寻找其他选择。

So they're not exclusive to Nvidia and obviously with the AMD deal they're looking around for other options.

但谷歌的宣传是，TPU和Nvidia芯片一样优秀，而且可能更便宜，当然，我们还得看看这是否属实，以及实际能有多少折扣。

But the way Google pitches it is that TPUs are just as good as Nvidia chips and they could be cheaper, which you know, we'll have to see exactly if that's true and what discounts people have of course.

凯伦，我想稍微退一步，看看这些连续发布的公告有何重要意义。

Karen, I want to step back here quickly and just look at the significance of all of these announcements in succession.

我的意思是，我们一直在讨论这个话题。

I mean, this is a story we've been talking about.

人们什么时候才会从英伟达转向其他选择？

When will people diversify away from NVIDIA?

还有哪些其他的替代方案？

What are the other alternatives?

超大规模云服务商的芯片努力最终会取得回报吗？

Will the hyperscalers chip efforts finally pay off?

这似乎终于到了一个关键时刻——亚马逊和谷歌的芯片终于开始大规模应用了。

This seems like finally a point where, I mean, you've got Amazon and Google, their chips are finally getting used in a big way.

我不认为有人会说这对英伟达是坏消息，因为我们刚刚看到他们本周发布了惊人的财报。

I don't know that anyone would argue this is bad news for NVIDIA because we just saw that they posted monster earnings this week.

但这整体上告诉了我们关于芯片行业的什么信息？

But what does this tell you about the chip story overall?

我认为这些芯片交易中有一件有趣的事情是，除了谷歌和Meta之外，它们通常都伴随着融资轮次。

I think that one thing that's interesting about all of these chip deals is that they do tend to With the exception of Google and Meta, they tend to come in tandem with funding rounds.

我认为OpenAI特别是一个例子，微软最初通过一次性投资数十亿美元赢得了它作为客户。

I think OpenAI specifically is one customer that, you know, Microsoft won originally as a customer by investing billions of dollars at once into.

我们现在看到，当OpenAI的业务与500亿美元的投资同步进行时，亚马逊也在这样做。

We're now seeing Amazon when OpenAI's business is in tandem with a $50,000,000,000 investment.

我认为我们已经看到亚马逊和微软的投资者和股东质疑这种投资是否可持续，以及OpenAI是否会在这些投资成为过去后，仍然作为一个持久的客户留下来。

And I think we have seen, you know, investors and shareholders in Amazon and Microsoft questioning whether that is sustainable, and whether you know, OpenAI is a customer that will be durable and will stick around even after these, you know, investments are long in the rearview mirror.

我认为有趣的是，我们开始看到像Meta这样的公司转向谷歌。

I do think it's really interesting that we're starting to also see companies like Meta turn to Google.

我认为整个云行业面临的一个大问题是，AI实验室和构建AI模型的公司对训练任务的需求加速会持续多久。

I think that the big question for the entire cloud industry is just you know, how long this acceleration in demand for training jobs among like the AI labs, the companies building AI AI models will persist.

如果这种需求继续多样化，到目前为止，情况看起来是肯定的：需求足够旺盛，而且不同芯片之间的竞争足够激烈，使得这些新芯片开始以有意义的方式崛起。

And if that'll continue to diversify, so far, it's looking like yes, there is enough demand and, you know, enough competition between different chips that these newer chips are starting to take off in meaningful ways.

而且，阿尼萨，我觉得另一件有趣的事情是，这也让我们看到了哪些公司的芯片努力真正取得了成效，因为本周我们还有另一个消息，称Meta也在考虑设计自己的芯片。

And, Anisa, I mean, you know, one thing that is also interesting to me is it also shows us whose chips efforts are actually working because we had another story this week that Meta was also thinking about designing its own chip.

这个项目在某些方面似乎已经失败了。

Really that effort seems to have fallen apart in some ways.

没错。

Exactly.

是的。

Yeah.

我们的同事JOT Wayne和Channer撰文指出，Meta已搁置了其先进AI训练芯片的计划。

Our colleagues, JOT Wayne and Channer wrote about how Meta is shelving plans on its advanced AI training chip.

据行业消息来源称，Meta的训练芯片项目从一开始就被认为可能是首个不会成功的内部硅项目。

That, you know, I can say like from industry sources, the Meta training chip program was always seen as maybe the first in house silicon project that, you know, maybe wouldn't be successful.

但我确实认为，这确实是。

But I do think this is yeah.

当你看到像英伟达这样的巨头公司每年将其芯片性能提升数个数量级时，所有这些公告都在给内部芯片开发项目施加压力。

You know, all of these announcements are putting pressure on in house chip development efforts when you have giant companies like Nvidia making their chips you know orders of magnitude better year on year.

在内部开发一款具有竞争力的芯片非常困难。

It's very hard to have a competing chip internally.

人们认为，内部芯片在特定的Meta工作负载中会非常有用，比如Meta可以为与其核心业务紧密相关的任务定制芯片，而不是制造像NVIDIA显卡那样通用的AI训练芯片。

Think, you know people think that in house chips are going to be very useful on, maybe specific Meta workloads where Meta can, you know, make a chip for a workload that is very, core to their business versus Meta making an AI training chip, that is as ubiquitous as an NVIDIA GPU.

对。

Right.

好了，Anissa和Aaron，感谢你们的到来。

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

这位是我们的云计算与计算记者Anissa Gardizi，以及我们的微软记者Aaron Holmes，他们都来自《The Information》。

That is Anissa Gardizi, our cloud and compute reporter, and Aaron Holmes, our Microsoft reporter here at The Information.

Netflix已退出了对华纳兄弟的竞购战。

Netflix has dropped out of the bidding war for Warner Bros.

Discovery在与Paramount经过数月的激烈竞争后，终于退出了竞购。

Discovery after duking it out with Paramount for months after it made its initial bid.

本周在《编辑精选》节目中，我想深入剖析这一新闻，以及本周其他几则重大消息。

This week on The Editor's Cut, I want to unpack that news and also some of the other big stories of the week.

我们邀请了我们的联合执行主编Martin Pearce和专题编辑Nick Wingfield加入。

We are bringing on Martin Pearce, our co executive editor, and Nick Wingfield, our features editor.

欢迎你们两位。

Welcome to you both.

很高兴你们再次回来。

It's great to have you back.

嘿，卡什。

Hey, Kash.

嘿，谢谢你们邀请我们。

Hey, thanks for having us.

好的，马丁。

Okay, Martin.

所以，你之前就猜到了。

So, you kind of called this.

你说过是Netflix。

You said Netflix.

这真的是一次认真的竞标吗？

Was this really a serious bid at all?

他们打算退出吗？

Were they going to drop out?

给我们详细讲讲你的分析过程。

Walk us through your calculus here.

我觉得这是认真的。

I think it was serious.

我不认为他们会花这么大的力气，仅仅是为了当一个搅局者，给派拉蒙制造麻烦，但我觉得他们确实做出了正确的决定。

Don't think they would have gone through this much trouble just to sort of be a stalking horse and just to make life difficult for Paramount, but I think they definitely made the right decision.

我认为现在故事的重点会转移到派拉蒙身上。

I think now the story will move to Paramount.

今天早上我看了派拉蒙本周发布的第四季度财报。

I was looking this morning at the Paramount fourth quarter numbers that were released this week.

这家公司去年的收入下降了。

That company's revenue declined last year.

所以这是一家正在管理一个衰退业务的公司。

So here's a company which is overseeing a declining business.

它即将承担巨额债务来收购另一家也于本周公布业绩下滑的公司。

It's about to take on a massive amount of debt to buy another business which also reported this week that they were declining.

埃里森家族正在投入数十亿美元的股权资金。

The Ellison family is putting in tens of billions in equity.

他们已经向派拉蒙投入了数十亿美元。

They've already committed several billion to Paramount.

他们将这两家业绩下滑的公司合并所承担的风险是巨大的。

The risk that they are taking by combining these two declining companies is enormous.

所以我认为奈飞真是运气不错，你可以通过股价看出这一点——就在奈飞退出此事的消息公布后，股价立即上涨了。

So I think Netflix has really lucked out and you can see that by looking at the stock price, which has just moved up on news that Netflix is out of this thing.

尼克，我记得几个月前、几周前，我们三个人在这里讨论时，曾谈到这对两家公司意味着什么。

Nick, I seem to recall when the three of us were on here a couple months ago, weeks ago, I don't know, we were talking about what it would mean to both companies.

我认为你似乎更倾向于认为，派拉蒙的合并或许更有道理，如果真有什么道理的话。

I think you seem to be more in the camp of Paramount might make a little more sense, if any sense at all.

嗯，是的。

Well, yeah.

我认为只要他们不插手那些核心业务，这才是关键所在，我昨晚在HBO上看了相关内容。

Think that as long as they don't mess with the pits, that's really what this comes down to, which I was watching last night on HBO.

归根结底，问题就在这里。

That's what it's all about at the end of the day.

我的意思是，我们走着瞧吧。

I mean, we'll see.

这两家公司之间有很多重叠之处。

There's a lot more overlap between these two companies.

所以我认为这对好莱坞的影响将会非常值得观察。

So I think the implications for Hollywood are going to be really interesting to watch.

Netflix当然辩称，他们在院线发行等方面重叠较少。

Netflix's argument of course was that there was less overlap in terms of theatrical and things like that.

马丁对这方面更专业，但我认为，如果这笔交易最终达成，合并后的公司很可能在人员方面面临相当严重的调整。

Mean, Martin is more of an expert on this than I am, but I think we're likely to see some, if this deal goes through, some pretty serious implications from a workforce standpoint for the combined company.

我可以告诉你，记者们会特别关注的一个领域是，CBS和CNN最终会被合并，因为记者们最喜欢写关于自己行业的新闻。

One area which I can tell you reporters will obsess about because reporters like nothing more than to write about their own industry is that CBS and CNN will end up being combined.

我向你保证，这将引发海量报道，远超其他任何事件。

So I guarantee you that this is going to produce reams of coverage that will dwarf anything else.

这只是整个事件中相对较小的一部分，但它很重要，而且对派拉蒙获得批准可能很关键，因为我相信特朗普希望CNN由他认为对他更友好的人拥有。

That's a fairly small part of this whole thing, but it will be important and it's probably important for Paramount's ability to get this approved because I'm sure Trump would like CNN to be owned by people that he considers friendlier to him.

马丁，谈谈这对整个媒体格局意味着什么。

Martin, just talk a little bit about what that means for the media landscape at large.

我的意思是，如果这两个网络合并，这说明了媒体的什么趋势？

I mean, have these two networks, if they come together, what does that say about the media?

这是不可避免的吗？

Was it inevitable?

当

When

你说媒体时，你指的是CBS和CNN吗？

you say media, you're asking about CBS and CNN?

是的，CBS和CNN，没错。

Yeah, CBS and CNN, yes.

我的意思是，你看，这两个都是正在衰退的资产，所以把两个衰退的业务合并起来，希望能削减一些成本并提高盈利能力，这是很合理的。

Well, I mean, look, those are both declining assets, so it makes sense that you would put together businesses that are declining and hopefully you can cut some costs and make it more profitable.

这对新闻行业来说并不理想，但这就是当前的发展趋势。

It's not ideal for the news industry, but that is the way things are moving.

马丁，关于这个话题，我最后问你一个问题。

Last question for you on this topic, Martin.

就反垄断这个部分而言，政府审查会有多大风险导致这笔交易被阻止？或者这取决于他们的计划吗？

Just on the antitrust component here, the government review, how much of a risk is there that this gets blocked or does it matter on what the plans are?

我觉得风险不大。

Don't think there's much risk.

派拉蒙前几天宣布，他们已经获得了所需审批中的其中一项。

Paramount announced the other day they've already secured one of the approvals that they need.

所以我认为这笔交易会通过。

So my guess is this is going to go through.

我现在真正关心的问题是，派拉蒙如何应对成本问题。

I I think the real question now is how Paramount can handle the cost.

因为正如我所说，他们必须承担巨额债务。

Because as I said, they have to take on a huge amount of debt.

将会有很多削减措施发生。

There's going be a lot of cutting that goes on.

而华纳在过去四五年里一直在经历这些，所以这不会轻松。

And Warner has already been going through that over the last four or five years, so it's not going to be fun.

尼克，本周另一个重大新闻——虽然还有很多其他新闻——但Anthropic与政府之间的对峙一直是所有人关注的焦点。

Nick, one of the other big stories this week, and there were many, but the standoff between anthropic and the government has been a story that everyone has been watching.

请你给我们简单介绍一下你关注的这个事件，以及对你来说最有趣的部分是什么。

Walk us through a little bit of what you've been paying attention to in this story and what the most interesting parts are for you.

在过去一两周里，我们目睹了这场戏剧性对峙：一方面，国防部（误称为‘战争部’）要求Anthropic交出其AI技术，且不对 Pentagon 使用该技术进行战争施加任何限制。

Well, we've kind of seen this drama play out between the Department of War, the Pentagon on one hand, and Anthropic over the past, I don't know, one to two weeks, where the Department of War is demanding that Anthropic hand over its AI technology with no restrictions on the Pentagon's ability to use it in warfare.

Anthropic 是众多AI实验室中，对安全措施最为直言不讳的一家。

Anthropic is among the AI labs out there, is the most outspoken, shall we say, on safeguards.

他们不希望自己的技术被用于大规模国内监控或自主武器系统。

They don't want their technology used in mass domestic surveillance or for autonomous weapons.

现在，国防部为Anthropic设定了今晚这个最后期限，要求其同意其要求。

Now, there's a deadline tonight that the DOW has established for Anthropic to kind of agree to its demands.

如果Anthropic不答应，就会威胁采取各种极端后果，比如禁止Anthropic的技术，或将其列为供应链风险，这将带来相当严重的后果。

And if it doesn't, it's threatening all sorts of crazy implications, including that basically Anthropics technology would be banned or declared a supply chain risk, which would have some pretty serious implications.

最新出炉的消息是，OpenAI也正在划下一条红线，表示他们同样对技术的使用方式存在担忧。

The news hot off the presses here is that OpenAI also is basically drawing drawing a red line in the sand and saying they too have some concerns about how their technology would be used.

正如我们今天刚刚发布的一篇报道中指出的，不仅OpenAI对此表示担忧，谷歌可能也会有类似的顾虑。

As we report today in a story that just published, not only is OpenAI concerned about it, but Google probably will also have some concerns.

当然，公司内部正在展开讨论，员工们也在呼吁公司采取与Anthropic类似的立场。

Certainly there's discussion internally and employees advocating that the company take a stand similar to the one that Anthropic's taken.

你觉得，当其他公司也开始表达担忧时，这件事最终会如何发展？你认为政府会如何回应？

How do you think this inevitably plays out that now that other players, other than Anthropic, are starting to voice their concerns, what's your best guess on how this plays out and what the reaction is from the government?

这很难说。

I mean, it's tough to say.

一方面，Anthropic的技术确实非常出色，而国防部正想要它。

I mean, the one hand, anthropics technology seems really good and the Department of War wants it.