英伟达 GPU 短缺波及微软，ChatGPT-5.5 评测，Meta 与 AWS 芯片合作

欢迎各位收看《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4月24日，星期五。

It is Friday, April 24.

今天第一个话题，《The Information》独家报道了当前GPU短缺的现状，还披露了微软和通用催化剂公司对此的应对措施。

First up today, The Information published exclusive reporting about the current state of the GPU crunch with some inside reporting on how Microsoft and general catalyst are responding.

接下来，我们会深入探讨GPT-5.5，邀请了Harvey公司一位实际使用过这个模型的顶尖研究员来分享。

We'll then dig into GPT 5.5 with a top researcher at Harvey who has used the model.

我们还会分析Meta公司10%的裁员计划，我们的联合执行主编马丁·皮尔斯认为，这次和以往科技巨头的裁员可能有所不同。

We're also digging into Meta's 10% layoffs and why our co executive editor Martin Peers thinks this time could be different than previous rounds of cuts that big tech companies have made.

我们也会聊聊Meta和亚马逊新达成的芯片合作，顺便简要谈谈英特尔的财报情况。

We'll also get into Meta's new chip deal with Amazon, and we'll talk briefly about Intel earnings.

最后，我们会独家揭秘Cursor公司与SpaceX合作背后的内幕。

We'll then wrap the show with an exclusive look behind the scenes of Cursor's deal with SpaceX.

这期节目会很精彩，咱们直接开始吧。

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

显卡依然一卡难求，芯片短缺在硅谷引发了连锁反应。

GPUs are still hard to come by, and the supply crunch has caused reverberations throughout Silicon Valley.

今天早上，我的同事亚伦·霍姆斯、阿尼萨·加尔迪齐和斯蒂芬妮·帕拉佐洛发布了一篇深度报道，里面包含了关于微软、通用催化剂以及快速成长的初创公司的内部消息。

My colleagues Aaron Holmes, Anissa Gardizy, and Stephanie Palazzolo published a deep dive on that dynamic this morning with inside reporting on Microsoft general catalyst and fast growing startups.

我想请亚伦和阿尼萨来分享更多细节。

I want to bring on Aaron and Anissa to share more.

欢迎两位。

Welcome to the both of you.

阿尼萨，我先从你开始。

Anissa, I want to start with you.

现在芯片圈里到底发生了什么？

What is going on in Chipland?

在芯片领域，情况跟2023年很像，那时候我们不断听到初创公司和他们的投资人说，想搞到英伟达的显卡简直难如登天。

In chip land, things are very reminiscent of 2023 when we were constantly hearing from startups and their investors that it was extremely difficult for companies to find access to NVIDIA GPUs.

和2023年类似，原因是一些大型云服务商把大量GPU囤起来，留给自己的团队和像Anthropic、OpenAI这样的大客户。

And the reason much like 2023 is that some of the large cloud providers are, you know, hoarding some of these GPUs for their own internal teams and their largest customers such as Anthropic and OpenAI.

所以对那些急需GPU的初创公司来说，眼下这情况真不怎么样。

So it's not that great out there for startups who are looking for for GPUs.

我们接触的业内人士都说，今年他们遇到的最大瓶颈就是这个。

The people that we talk to said that this is their their biggest bottleneck that they're facing this year.

今年啊。

Oh, this year.

明白了。

Okay.

没2023年那么糟，但也是最近几年最紧张的时候了。

So it's it's not quite as bad as 2023, but it's it's as bad as it's been lately.

我也想听听Aaron的看法，但据我观察，情况可能比2023年还糟，因为AI的商业模式更清晰了，OpenAI和Anthropic花的钱比以前更多。

I mean, I'd love to hear what Aaron thinks, but from what I can tell, it sounds a little bit worse than 2023 because, the the business model around AI is a little bit more clear and OpenAI and Anthropic are spending even more money than they were spending back in 2023.

所以我觉得，情况更严峻了。

So it seems it seems worse to me.

亚伦，你怎么看？

Aaron, what do you think?

我觉得，2023年的时候，ChatGPT刚出来，那时候每家公司都想试试AI，看看怎么用到自己的业务里。

Yeah, I think to me what stands out is, you know, 2023 was this moment right after chat GPT came out where every company wanted to experiment with AI and see where it would fit in to, you know, their business.

但现在不一样了，很多公司已经很清楚自己需要GPU，尤其是像Anthropic和OpenAI这样的大客户，他们正面临AI相关计算需求的激增。

And I think what's different now is that there's a lot of companies that know they need to use GPUs, especially you know, that this sort of mega customers like Anthropic and OpenAI who are seeing this huge AI coding demand boom.

结果就是，GPU的价格被推高了，几乎演变成了一场竞标战。

But as a result, is kind of leading to this almost bidding war where prices of GPUs are going up.

比如在微软Azure云上，你根本没法租到少量最新的GPU，大多数情况下，你必须一次性预订至少一千个NVIDIA Blackwell芯片的核心。

And you know, I'm hearing that at Microsoft Azure cloud, for example, you know, you actually can't even reserve a small number of newer GPUs, you have to commit to reserving, in most cases, 1,000 cores or more of the Nvidia Blackwell chips.

这种预订通常要持续一到三年，总费用高达数千万美元。

And that costs, you know, 10s of millions of dollars over the time period that you have to reserve it, which is like one to three years.

云服务商在GPU销售上这么强硬的做法，我以前还真没见过。

So that's kind of a bolder sell from the cloud providers than I think we've ever seen before when it comes to GPUs.

阿妮萨，亚伦在说价格在上涨。

So Anissa, Aaron is talking about prices going up.

如果你是个初创公司，现在还有议价能力吗？

If you're a startup, do you have any bargaining power here?

你只能接受租这些芯片的价格比以前高很多了吗？

Do you just accept that the price of renting these chips is higher than it would have been?

你们是怎么应对的？

How are you coping?

如果你需要几千块GPU，现在根本没什么议价空间。

If you're looking for maybe a thousand to a couple thousand GPUs, you really don't have any bargaining power right now.

一年前或许还有机会，但现在想从云服务商那里拿到好价格，几乎不可能。

Maybe you did a year ago but these days it's going to be really tough for you to try to get a good price from your cloud provider.

我们认识一家初创公司，去年秋天他们签了个不错的合同，每小时每块GPU不到3美元租了六个月。

We talked to one startup in particular who was able to get a good contract last fall for six months renting GPUs for just under $3 per GPU per hour.

今年早些时候，他们再想租GPU时，之前争着抢他们生意的那些供应商，连回个电话都不愿意。

And earlier this year when they went back on the market looking for GPUs, some of the same people who were competing for their business weren't even giving them a call back.

这充分说明了市场变化有多快。

So that's just a really good example of how quickly things have changed.

我们还了解到，一些大型风投机构，比如 General Catalyst、Founders Fund 和 Sequoia 的投资组合公司，也就是那些初创企业，现在很难拿到 GPU。

And we also learned that portfolio companies of some of the largest VC firms, General Catalyst, Founders Fund, Sequoia, startups are struggling to get GPUs.

所以最近 General Catalyst 给旗下的所有创始人发了一封邮件，问大家获取 GPU 的能力怎么样。意思大概是，我们可能会帮你们搞定 GPU，代表你们去谈判，这样你们就能以更大的采购量去争取更好的价格。

And so recently General Catalyst sent out a note to all of their founders asking them, you know, what is your ability to get GPUs sort of in the vein of saying we might end up helping you guys all get GPUs and negotiate on your behalf so that you guys can be basically a larger customer trying to get a better price.

所以如果你现在需要几千张 GPU，那确实挺难的，这和几个月前的情况完全不同了。

So if you're looking for a couple thousand GPUs, right now it's pretty tough and that's different from just a couple months ago.

Aaron，让我们回到之前你们聊到的话题，关于云服务商囤积这些 GPU 的事。

Aaron, let's go back to what you guys were talking about earlier in terms of the cloud providers hoarding these GPUs.

你负责微软这边。

You cover Microsoft.

能不能具体讲讲，微软在地面上是怎么应对算力短缺的，以及它是怎么平衡自身算力和客户算力需求的？

Tell me a little bit tactically about how Microsoft has been dealing with the supply crunch on the ground and how it's balancing its own compute demands with the compute demands of its customers.

对。

Yeah.

也就是说，微软对此非常关注，而且公司也坦承，由于需要预留大量GPU用于自身开发和运行Copilot，Azure的收入增长受到了限制。

So I mean, this is something that Microsoft is looking at really closely, and the company has actually been pretty open about the fact that their Azure revenue growth is being constrained essentially by how many GPUs they need to reserve for their own, you know, serving and developing Copilot.

另外，我在做这个报道时发现，Azure 上能拿到 GPU 的客户其实分好几层。

And then on top of that, you know, what I learned in reporting for this story is that, you know, there's sort of different tiers of customers that can get GPUs on Azure.

我觉得最顶层的是 OpenAI 和 Anthropic 这类公司，Azure 正在为它们大规模新建 GPU 集群。

I think, you know, at the top, there's the OpenAI and Anthropic, which Azure is actively building massive new GPU clusters for.

再往下是第一梯队，比如那些已经在 Azure 上花大钱的财富 500 强企业，它们能优先拿到 GPU。

And then, you know, beneath that, there's sort of tier one, which is like the the Fortune 500 companies that have already a huge spending footprint on Azure and can kind of get first dibs on GPUs.

再往下是一些规模小一点的公司，虽然没那么多预算，但和 Azure 的关系还不错。

Below that you might have some smaller firms that don't have quite as much, know, revenue to throw around, but that still have good relationships with Azure.

至于其他人，就完全靠抢了——想用 GPU，得等上几周甚至几个月才有资源。

And then everyone else is kind of in the Wild West where, you know, if you want GPUs, you have to essentially wait weeks or months to get capacity.

而且就算你终于抢到了，据我听说，Azure 对使用率要求特别严，必须 24 小时满负荷运行。

And then once you get capacity, Azure is, from what I'm hearing being, you know, very kind of stringent about if you aren't using the GPUs and running them around the clock.

哪怕只是周五晚上停几个钟头，都会被认定为资源闲置，直接踢出集群，重新排到下个月的长队末尾。

And even if they're down for just a couple hours on a Friday night, then you will essentially be kicked off your GPU cluster for underutilization and kicked back to the back of the month's long queue.

这就导致很多小公司觉得在 Azure 上根本没法稳定用上 GPU，干脆考虑自己买显卡、换别的云平台，或者找新兴云服务商合作。

So that's kind of leading to a situation where smaller companies feel like they can't rely on having steady GPU access at Azure, which is, you know, leading to them making decisions like maybe they should just buy GPUs and run them somewhere else or, you know, look for other deals with with neo clouds.

这种快速变化的格局，确实让这些AI初创公司对自己的GPU获取能力感到不安。

It's it's definitely this rapidly shifting landscape that is leaving these AI startups feeling a little bit insecure about their ability to get GPUs right now.

所以，Anissa，Aaron跟我们讲了微软的情况。

So Anissa, Aaron's told us about Microsoft.

你有没有听说过，其他云服务商也在玩同样的把戏？就是这种‘用不了就取消’的政策？

Do you have any sense if other cloud providers are playing the same game here in terms of this use it or lose it policy?

还有那些新兴云服务商，他们是不是也在采取类似的做法？

And I mean, are the neo clouds for that matter, are they adopting sort of similar approaches at all?

说到新兴云服务商，这个话题其实挺有意思的。Aaron和我2023年一起做过这方面的报道，那时候很多初创公司都往新兴云服务商那边跑，大家觉得这正是新兴云服务商存在的理由。

The neo clouds are really interesting when it comes to this topic because Aaron and I actually reported this together in 2023, but at that time a lot of the startups were going to the neo clouds and this was seen this entire problem was seen as a reason that neo clouds should exist.

你知道的，亚马逊、微软、谷歌可能觉得你不够重要，不给你配大量GPU，但像CoreWeave这样的公司就大力宣传自己是解决GPU短缺问题的最佳选择，他们把自己定位成专门为拿不到GPU的公司服务的平台。

You know, Amazon, Microsoft, Google might not think that you're important enough to give you an allocation of a large amount of GPUs but companies like CoreWeave were really marketing themselves as the place to go if you couldn't get GPUs and they saw themselves as this was one of the big reasons that they should exist.

但过去三年我们看到，很多新兴云服务商其实已经转向优先服务最大的客户了。

But as we've seen over the past three years, a lot of the neo clouds have actually just prioritized serving the largest customers out there.

所以他们并没有像以前那样，默默在旁边帮这些初创公司。

So they're not really sort of waiting in the wings helping these startups.

所以现在的情况跟以前大不相同了，也许我们会看到新的新兴云服务商出现。

And so it's a much different time and you know maybe we'll see new neo clouds.

我觉得一些推理服务提供商正试图切入初创公司市场，但跟2023年已经不一样了，那时候很多初创公司还以为自己能靠新兴云服务商解决算力问题。

I think some of the inference providers are hoping to tackle the startup space, but it's a lot different than 2023 when some of these startups thought that they could go to the NEO clouds in this situation.

Anissa，最后一个问题。

Anissa, last question for you.

我记得2023年的时候，算力短缺不只是因为需求大，供应链本身也还没完全恢复，那时候想把芯片送到客户手里真的很难。

You know, I'm thinking back to 2023 and I'm recalling that part of that supply crunch was not just demand, it was also sort of I mean, the supply chain itself, it was still coming out of that big environment where it was tough to get these things in people's hands.

我不太确定现在是不是还存在同样的问题，但我想问的是，有没有什么迹象表明情况正在好转？

Not clear to me if that is as much of an issue today, but my question is, is there any signal that this is going to get better or that things might be improving?

短期内，初创公司看不到任何改善的迹象，因为他们每天醒来，都看到OpenAI和Anthropic这些公司宣布要花几百亿美元采购算力，而这些供应商正是他们自己也在争取的。

There's not much signal that this will improve in the short term for startups because they're waking up every day just like us and seeing these massive announcements from OpenAI, Anthropic about spending, you know, hundreds of billions of dollars on compute from the same people they're trying to get compute from.

所以除非那些大公司预订的芯片最终用不上，云服务商能把这些资源收回来再转给小公司，否则从我的角度看，短期内根本看不到出路。

So I think unless the biggest companies that have reserved the chips don't end up needing them and the cloud providers end up reclaiming them and redirecting them to smaller companies, there's not really a good end in sight from my perspective.

明白了。

Right.

谢谢你们两位，Aaron 和 Anissa，来参加这次对话。

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

这位是我们的微软记者 Aaron Holmes，这位是我们在《The Information》的云计算与计算方向记者 Anissa Gardizy。

That is Aaron Holmes, our Microsoft Reporter, and Anissa Gardizy, our Cloud and Compute Reporter here at The Information.

OpenAI 最新发布了 GPT-5.5 模型，进一步加剧了AI模型之间的竞争。

OpenAI released its GPT 5.5 model in the latest addition to the AI model rivalry.

我想请一位长期使用这个模型的人来分享一下他的初步体验。

I want to bring on someone who has been using it extensively for an early take on how the model is performing.

Niko Gruppen 是人工智能法律初创公司 Harvey 的应用研究负责人。

Niko Gruppen is the head of applied research at AI Legal Startup Harvey.

Niko，早上好。

Niko, good morning to you.

很高兴你再次回来。

It's great to have you back.

你一直在用 5.5.5，是吗？

You have been using 5.5.5, is that right?

没错，是的。

That's correct, Yeah.

那评价怎么样？

And what's the review?

给我们说说吧。

Give What us the

一周时间，变化真大。

a difference a week makes.

新的模型已经进场了，Akash。

New models have entered the arena, Akash.

上周是 Opus 4.7，这周就变成了 GPT 5.5。

Last week it was open, Opus 4.7, this week it's GPT 5.5.

所以 SPUD 终于正式登场了。

So the SPUD has officially landed.

社交媒体上关于这个模型的讨论很多，从我们目前看到的情况来看，这种热议是实至名归的。

There's a lot of buzz in the ether on social media about this model, from everything we've seen thus far, that buzz is warranted.

在我们内部的Big Law Bench基准测试中，它的得分达到了91.7，是历史上最高的分数之一。

So on our internal benchmarks, Big Law Bench, it posted one of the all time best scores at 91.7.

它在交易类和诉讼类法律工作这两方面都表现得非常出色。

It did really well across both transactional and litigation focused legal work.

在其他外部基准测试中，它的成绩也很亮眼。

And it posted great scores on other external benchmarks.

更重要的是，这些成绩并不仅仅集中在代码生成上。

Importantly, not only those focused on code generation.

所以，质量确实是有的。

So quality seems to be there.

我本来想专门问问你关于代码生成的事，毕竟这个能力现在这么关键。

I was going ask you specifically about code generation, just given how important cogeneration seems to be as a capability.

他们自己都说，GPT 5.5是迄今为止最强的智能编码模型，尤其是在OpenAI自家产品里，但跟其他公司的模型比起来，它的编程能力到底排第几？

I mean, they're calling it well, they said GPT 5.5 is our strongest agentic coding model to date, strongest certainly in terms of the OpenAI family, but how does it rank against other companies' models in terms of coding capabilities?

是的，我觉得至少过去三个月，甚至从年初开始，这些模型之间就明显拉开了差距。

Yeah, I think for the better part of three months now, and I think starting from the beginning of the year, there was a tangible gap between these models.

我认为，从OPUS 4.5到4.6，再到4.7，这一系列更新让Claude Code的性能一下子拉开了明显差距。

I think everyone recognized that OPUS 4.5, 4.6, and then through to 4.7 was driving Claude Code in a way that created a real gap in distance.

从我的角度看，这个差距现在正在慢慢缩小。

I think, from my perspective, that gap is starting to close.

而有了GPT 5.5，我觉得它可能又重回领先位置了。

And with GPT 5.5, I think it may be a back in pole position.

对吧？

Right?

我之所以说‘可能’，是因为有几点需要留意。

The reason that I say maybe here is that there's a couple things to watch out for, right?

首先，它现在还是个研究预览版。

First, it's research preview model.

目前主要只在编码环境中可用。

It's primarily available in codecs.

等它通过API全面开放，大家都能用上之后，真正的差异才会真正显现出来。

As it becomes more generally available via API and we get more widespread access, that's where you see these differences really start to show publicly.

其次，背后还是有那种神话般的期待在暗中影响，对吧？

And then secondly, we still have mythos sort of lurking in the background, right?

所以你可以做一时的观察，但做决定时得靠预测事情的发展趋势。

And so you can make point in time observations, but you have to make decisions based on projections of where things are going.

你说背后有神话般的期待？

You say mythos lurking in the background.

作为研究人员，你有没有想过自己哪天能用上 Mythos？

I mean, are you, as a researcher, are you expecting to ever be able to use mythos?

我的理解是——先不管那些泄露的事，毕竟大家都知道，那不是官方允许的使用方式。

Because my understanding of it and let's put leaks aside for the moment, because we know that you know, that's not the official way people are supposed to be able to use mythos.

对吧？

Right?

所以关键在于，他们到底要不要把 Mythos 对公众开放。

So, I mean, there's a decision to make whether or not they're gonna make it available to the public or not.

我觉得普遍的看法是，把它公开可能并不是个明智的选择。

I think the consensus has been it might not be the smartest idea to make it available to the public.

所以你说它潜伏在幕后，你是希望有一天能用上它吗？

So when you say it's lurking in the background, are you expecting at some point to be able to use it?

我觉得各实验室会继续提升性能和效率，毕竟这背后有很强的驱动力，对吧？

My expectation is that the labs will continue to push on performance and on efficiency, And the incentives are there to push both of those, right?

所以我希望它们能持续发布这些技术，让更多人用得上。

So my hope is that they'll continue to release these things and make them generally available.

哪怕因为安全问题、算力限制，或者其他原因暂时没法开放，我依然觉得有足够动力去使用它们。

And whether they're not available because of security issues or compute constraints or for whatever reason, I think the incentives will be there to use them.

如果市场上出现另一个性能相当甚至更强的模型，用户自然会把流量转向那个模型。

If there's another equally performant or higher performance model on the market, the incentives for the consumers of models will be to switch traffic over to those.

所以我预计，OpenAI、Anthropic 和 DeepMind 会继续推动模型智能的前沿，不断推出新成果。

And so I expect the labs, OpenAI, Anthropic, and DeepMind to continue to push and release the frontier of model intelligence.

我想问一下，你们有没有担心过这些模型的成本问题？现在大家的假设是，很多技术其实都在被补贴。

Can I ask, are you guys concerned at all about the costs of these models and the hypothesis being, hey, a lot of this technology is subsidized right now?

等它们开始按实际成本收费，应用层的公司恐怕就很难负担得起这些了。

Once they start to charge what they need to in order to turn a profit, it's going to be really hard for application layer companies to afford all this.

你在实际操作中会考虑到这一点吗？

Is that something you think about on the ground?

是的，毫无疑问。

Yeah, absolutely.

GPT 5.5 这次发布的一个亮点，正是这种效率提升。

And this is actually one of the high points for the GPT 5.5 model release, is exactly that efficiency.

过去两年，我们一直处在追求质量优先、忽略延迟和成本的阶段。

So I think we've been living in a world for the last two years where quality has taken precedent over things like latency and cost.

这么做的原因是为了推广普及。

The reason for that is adoption.

对吧？

Right?

如果质量不够好，用户就不会用。

If quality is lower, adoption is lower.

我们还处在行业的早期阶段。

We're in a nascent industry.

采用率和市场接受度才是关键。

Adoption and traction are key.

但随着企业以及像我们这样的公司逐渐清醒，对令牌使用成本有了更成熟的看法。

But as I think enterprises and companies like ourselves are wisening up or developing a more mature stance with respect to token spend.

所以我认为，像GPT 5.5这样的模型，虽然比GPT 5.4质量更高，但在一些实验中却能减少50%的推理令牌，效率大幅提升。

And so I think the idea that a model like GPT 5.5 is a higher quality model than GPT 5.4, but in some experiments has shown 50% reduction in reasoning tokens, much greater efficiency.

这些因素变得至关重要。

Those things become really important.

因此，我们开始不再只关注模型能提供多高的极致质量，而是更多地考虑每美元或每个令牌能换来多少质量。

So we're starting to think about things less through the lens of what is the maximum quality that we can get from a model, and more so from the perspective of what is the quality per dollar spent, or the quality per token used.

最后一个问题，尼科。

Last question for you, Niko.

本周Anthropic宣布要更深入布局法律领域，我想大家对此早有预期。

So Anthropic came out this week with some news that they are pushing deeper into the legal space as well, and I think people probably saw this coming.

大型实验室开始转向具体应用场景了。

The big labs starting to move into applications.

哈维在法律领域可是个响当当的名字。

Obviously, Harvey is a big name in the legal space.

你怎么看待他们的这次发布？和Anthropic这样的大公司相比，你们的优势在哪里？

How are you thinking about that release and what your differentiation is against a giant company like Anthropic?

是啊，他们最近确实挺忙的，对吧？

Yeah, they've certainly been busy, haven't they?

过去十天里，他们已经涉足了设计、法律、金融、网络安全这些领域。

I think it was design, cogen, finance, cybersecurity, and legal, all verticals they've touched in the last ten days.

那边的动作真不少。

A lot that's going on there.

我们的重点始终是让哈维成为法律团队的首选，确保律师需要的工具和服务都做到顶尖水平。

Look, our focus will always be on ensuring that Harvey is the best bet for legal teams, that the tools and services that lawyers need to be first class will be made first class.

这里有个重要的趋势值得注意：什么时候他们谈模型能力，什么时候又谈产品差异？

And I think there's an important pattern to pay attention to here: when are the labs talking about model capabilities, and when are they talking about product differentiation?

如果你在模型能力上领先，或者即将发布新模型，自然会更强调评测指标、性能数据。

If you're ahead on model capability, or in your upswing about to release a new model, you're going to be emphasizing benchmarks, evaluations, model performance.

而当你处于低谷期，或者在两次发布之间，就会更侧重讲产品的功能。

And when you're in a downswing or when you're in between launches, let's say, talking about product capabilities.

所以这也不奇怪，Anthropic 其实很擅长向企业市场宣传自己，把自己打造成企业级 AI 解决方案。

And so no surprise here, Anthropic has done a great job actually of marketing themselves to the enterprise, to be the AI enterprise sort of offering.

因此，他们把产品重点推给那些市场规模最大的领域，比如金融、法律和 cogen，也就不足为奇了。

And so it's no surprise that they're marketing their product to the highest TAM verticals: financial services, legal services, and cogen being amongst three of the biggest.

这其实挺有意思的，我以前从来没这么想过，但你是不是在暗示——我可没曲解你的意思——如果你去研究一下模型在各种基准测试中的排名变化，再看看这些波动和产品发布之间的关系，会不会发现它们之间存在某种关联呢？

So this is kind of interesting, because I never thought about this, but are you maybe suggesting I'm not putting words in your mouth, but if you were to maybe study where the models rank on the benchmarks, and the oscillations between who's on top, if you were to sort of maybe look for a correlation between that and then press releases for applications, maybe I mean, what we're suggesting here is there could be a correlation there, right?

或者也可能是反向关联。

Or an inverse correlation.

当你排名靠后的时候，反而更爱谈应用场景。

When you're on the bottom, that's when you're talking more about your applications.

是的，我们自己也经常这么想。

Yeah, I think about this from our side too, right?

每个模型发布后都会达到一个特定的性能水平，但在下一次发布之前，还有很多工作要做。

There is a certain state of performance that's reached by any given model release, and then there's a lot that you need to do between model releases, right?

赛道上也是同样的情况。

And the same is true for the laps.

你有一段很长的预训练周期在持续进行。

You have this long kind of pre training clock that's running.

而训练完成后的周期则短得多。

You have a much shorter post training clock that's running.

但在两次模型发布之间，仍然有六到八周的时间需要填补。

But there are still six to eight weeks of time to spill between model launches.

总得做点什么。

Got to do something.

很好。

Great.

明白了。

Okay.

尼科，非常感谢你来参加这次对话。

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

一如既往，这位是来自Harvey应用研究部的负责人Niko Gruppen，正在TITV为您带来报道。

As always, that is Niko Gruppen, Head of Applied Research at Harvey here on TITV.

Meta计划裁员10%，大约涉及8000人。

Meta is cutting 10% of its headcount, which amounts to around 8,000 people.

本周《编辑精选》栏目中，为我们解读这一新闻的是我们的联合执行主编Peers。

Here to break down the news in this week's edition of The Editor's Cut is our co executive editor, Peers.

Martin，欢迎再次回到节目。

Martin, welcome back to the show.

很高兴你今天能来。

It was great to have you here.

你昨晚在简报里写的那篇关于Meta裁员的文章提到，这一轮裁员可能和以往不太一样。

You wrote this column last night in the briefing about Meta's cuts, and you suggested that maybe this round of layoffs could be different than past rounds of layoffs.

我的意思是，过去几年，包括谷歌在内的多家公司，每次裁员都被宣传为重大的成本削减行动。

Well, the point I was making was that the past rounds by all of the companies, Google as well as various other companies, were always reported as and the companies made this point, they were major cost cutting exercises.

但如果你仔细追踪这些公司裁员之后的员工人数——我说的是2022年开始、一直持续到2024年的那几轮裁员——你会发现，很多公司的员工规模后来又回升到了裁员前的水平。

But in fact, if you actually track their employee numbers in the years after those rounds, and I'm talking about the rounds of cuts that started in 2022 and went through 2024, if you track the employee numbers after that, in many cases, the workforces went back to at least the point where they had been before the cuts.

而且并不是每次都立刻反弹，比如谷歌，去年的员工总数比历史上任何时期都多。

And not always right back up, but in some cases, Google, for instance, ended last year with more people than they have had at any point in their history.

所以我不确定这次会不会再出现同样的情况。

So I'm not sure that will happen again this time.

以前之所以这样，是因为这些公司一边裁员，一边招进更专业的人才，尤其是搞AI的。

One of the reasons that was happening was that these companies were replacing the people that they had and bringing in more specialized talent, possibly for AI.

但现在，他们裁员更像是为了腾出工程师的薪资预算，好把钱投到AI上。

Now, I think they are replacing people in order or freeing up the money they spend on engineers so they can spend more on AI.

这一点还不太明确。

It's not clear.

我只是想说，公司刚宣布裁员的时候，我们根本不知道背后真实的情况，得过一段时间才能看清实际影响。

I'm just saying that you don't really know at the time they make these announcements what is actually going on and you have to track things over time to see the real impact.

你说的‘加大AI投入’，是指那些数据中心、算力设备之类的物理建设吧？

And when you say spend more on AI, these are all those investments in data centers, compute, all the physical stuff.

确实包括这些，但还有AI工具本身的开支，那也是一笔巨款。

There is that part of it, but there's also the spending on the AI tools, which is very expensive.

这些公司采用AI工具后，就得想办法在其他地方省钱。

As these companies adopt the AI tools, they have to find a way to reduce spending elsewhere.

最直接的办法就是裁员。

The obvious way is to cut the people.

很明显，这些AI工具部分目的就是替代一些人类的工作。

And obviously the AI tools are aimed partly at replacing what some humans do.

这正是大家普遍认为正在发生的事。

That's what everyone is thinking is happening.

我只是想说，过去的经验告诉我们，事情往往不会像宣布时看起来那样发展。

I'm just saying that, well, the history is that things don't always work out the way that they appear at the time of the announcement.

如果我们回头看以往的裁员潮，其实很多人过一段时间后又慢慢回来了，有些情况甚至如此。

So, I mean, if we just look back at previous rounds of cuts, I mean, idea that they do these cuts and then the people kind of trickle back over time or in Same some cases

人员。

people,

当然。

Obviously.

不同的人。

Different people.

是啊。

Oh yeah.

有时候是不同的人。

Sometimes different people.

不过我们确实听过不少这样的故事。

I mean, we've heard the stories though.

有时候有些人又被重新雇用了。

Sometimes some people just got hired back.

我

I

我想他们雇的不是原来那些人。

guess it's that they're not hiring back the same people.

偶尔会有，但大部分情况都是新招的人。

I mean, every now and then, but I think it's mostly different people.

但我是说，这些裁员是因为我今天早上和昨天看了Meta的股价。

But I mean, the cuts are because I was looking at Meta's share price this morning and yesterday.

我的意思是，股价根本没什么动静，通常投资者一兴奋，股价都会有个反弹，对吧？

I mean, the share price didn't really move, and oftentimes you see some sort of a pop as investors get excited, right?

关于这一点，有几点要说。

Well, a couple of things about that.

第一，这些裁员的消息早就传出来了，其实早就在进行中了。

One is that these cuts have been reported for a while as they were actually on the way.

没人感到意外。

No one is surprised.

另一点是，大家都清楚Meta在AI上砸了巨资，不管是资本支出还是运营成本，都在大笔招人。

The other thing is that everyone knows Meta is spending a fortune on AI both in CapEx and operating expenses in terms of hiring people.

他们一招人，花的钱就多得吓人。

And when they're hiring people, they're spending a fortune on hiring them.

所以华尔街可能心里想：这有什么好在意的？

So think Wall Street probably is a little bit like, Who cares?

另外，关于Meta的其他新闻，我想听听你对今天早上公布的Meta和AWS合作的看法。

Now, in other Meta news, wanted to get your read on the Meta AWS deal that was announced this morning, I believe.

Meta打算使用AWS的一些芯片，就是Graviton芯片。

So Meta is going be using some of AWS's chips, the Graviton chip.

我觉得这件事对AWS的影响可能比对Meta更大，但我想知道你怎么看。

My sense on this was this is actually more of a story that would impact AWS than Meta, but I wondered what you thought of it.

我觉得你说得对。

I think that's right.

现在这个阶段，这个领域的每家公司都在宣布各种合作项目。

I think we're in a period now of every company in this area is making announcements about deals that they're doing.

但大多数公告其实都是在重复他们之前已经说过的事情。

Most of these announcements fit into what they've already announced.

Meta早就说过，他们要在资本支出和芯片上投入巨额资金。

So Meta has already announced that they're going to spend an absolute huge amount of money on CapEx, on chips.

他们正在自主研发芯片。

They are developing their own chips.

他们也在从其他公司采购这些芯片。

They're also buying those from others.

这都只是整体计划的一部分。

This is all just part of that.

所以我觉得这对他们来说没那么重要。

So I don't think it's that meaningful for them.

但对亚马逊来说意义重大，他们正努力向外界证明，市场对AWS业务、包括他们自研芯片的需求是真实存在的。

I think it is meaningful for Amazon, which is trying to prove to the world that there is demand for its AWS business, including its own internally designed chips.

所以，你说得对。

So yeah, I think you are right.

而且，如果这就是目的，那确实说明了市场需求的存在。

Well, and if that's the point, it's showing that there is demand.

我的意思是，AWS这波操作像是后来居上——大概十八个月前，我们还觉得他们在AI领域有点落后。

I mean, it does feel like AWS has been another come from behind story in the sense that it feels like, what, eighteen months ago, we were sort of talking about them as a bit of a laggard in the AI game.

但自从有了对OpenAI的投资，再加上这一系列芯片合作，AWS似乎正逐渐重回话题中心。

But between the OpenAI investment and then all these chip deals, it feels like AWS is finding their way into the conversation more and more.

我的意思就是，这些公告正是我想说的。

I mean, that's what I mean about the announcements.

他们确实在慢慢融入进去，但真的能翻身吗？

They're finding their way into that, but are they actually coming back?

现在还说不准。

We don't know.

我觉得现在下结论还太早。

I think it's a bit early to say.

我觉得这个市场非常大。

I think this is a very big market.

每个人都有不少机会。

There's lots of opportunities for everyone.

AWS当然是大公司，肯定会有自己的位置，但它们真能追上其他人吗？

I mean, obviously AWS is a very big firm and they will have a role to play, but are they catching up to the others?

我觉得现在还真的看不清楚。

I just don't think that we know yet.

别太受这些公告影响，它们其实就是公关宣传。

I wouldn't be overly influenced by these announcements, which they're just PR.

我的意思是，谁知道呢？

I mean, who knows?

好的。

Okay.

咱们赶紧聊聊英特尔的其他芯片动态。

Now let's talk about Intel quickly in other chip news.

英特尔刚公布了季度财报，市场份额是

So Intel reported its quarterly results Share is

25%。

25%.

25%。

25%.

准确说是22%。

Well, 22%

今天早上。

this morning.

哦，已经平稳下来了。

Oh, it's already it's calmed down.

没事吧？

Is it okay?

之前还涨了25%呢。

Well, it was up 25 earlier.

今年到现在，股价已经涨了大约125%。

So far this year, stock is up 125% or so.

因为当初大家都以为英特尔完蛋了，现在却证明它还活着。

And that's because Intel, was left for dead, is now showing it's alive.

英特尔在CPU领域，我觉得依然是它真正占据优势的地方，它依然保有这个优势。

Intel is I think the CPU, which is where Intel is really still they still have

一个

a

在这个市场里有很强的存在感。

big presence in that market.

我不太确定该怎么称呼这种芯片类别，但现在AI领域越来越重要，英特尔也开始被大家讨论起来了，你懂的。

I'm not sure what you call it, the category of chip is becoming more important in AI and so Intel is finding its way into the discussion, as you might say.

我觉得现在说它能不能重回巅峰还为时过早，可能不太现实，但至少英特尔活过来了，这对美国和英特尔都是好事。

I think again, it's a bit early to say whether it will ever reclaim the position it was in, probably not, but Intel is alive and that's good for America and it's good for Intel.

从股东角度来说，这确实对美国也有好处。

Well, it's literally good for America as a shareholder in the company.

而且这对美国自主制造芯片的能力也很重要。

It's also good for American ability to make its own tips, which is important.

没错，马丁。

Well, Martin, Right.

谢谢你今天来做客。

I want to thank you for coming on.

这位是马丁·皮尔斯，我们《信息》杂志的联合执行主编。

That is Martin Peers, our co executive editor here at The Information.

Cursor和SpaceX的这笔交易，让硅谷很多人都大吃一惊。

Cursor's deal with SpaceX took Silicon Valley by surprise in many ways.

SpaceX可能以600亿美元收购这家编程初创公司，更让人意外的是，Cursor一路走到这一步的背后故事。

SpaceX could acquire the coding startup for $60,000,000,000 and equally as surprising is the backstory behind Cursor's road to that deal.

我的同事科里·温伯格、朱莉娅·霍恩斯坦、Aaron Wu和凯蒂·鲁夫做了一篇深度报道，披露了大量内幕，我想请朱莉娅来跟我们说说她发现了什么。

My colleagues, Corey Weinberg, Julia Hornstein, Aaron Wu, and Katie Roof wrote a deep dive with inside details about that, and I want to bring on Julia to tell us more about what she learned.

朱莉娅，欢迎再次做客我们的节目。

Julia, welcome back to the show.

很高兴你今天能来。

It's great to have you here.

我也很高兴能来。

Great to be here.

当你看到Cursor和SpaceX可能要合作的新闻时，

So you see this announcement that Cursor and SpaceX could be doing a potential deal.

你就立刻开始打电话跟进。

You start working the phones.

作为记者，你最想弄清楚哪些问题？

What are the questions that you want to get answers to as a reporter?

我们一直好奇的是，Cursor 能否维持这种爆炸式增长，还是已经到顶了？

Some of the questions that we've been wondering for a while are, can Cursor sustain its explosive growth or is it peaking?

我们也很想知道它的利润率和成本结构到底是怎样的。

I mean, we're also kind of curious about what its margins and cost structure looked like.

就在本周早些时候，SpaceX 宣布有权以六千亿美元收购 Cursor，或者支付一百亿美元的分手费，我们特别想深入了解一下，到底是什么促使 Cursor 选择被 SpaceX 收购，而不是继续独立发展。

And after SpaceX earlier this week said that it has the right to acquire Cursor for $60,000,000,000 or pay a $10,000,000,000 breakup fee, we really wanted to dig into what pushed Cursor toward the SpaceX acquisition instead of staying the course.

我们还纳闷，为什么在 SpaceX 的收购消息公布前一周，投资者还愿意以五千亿美金的估值投资 Cursor，因为这笔融资几乎会让它的估值比 2025 年 11 月的上一轮融资翻倍。

We were also wondering why investors up until late last week before the SpaceX deal was announced, were still willing to fund Cursor at a $50,000,000,000 valuation because that fundraise would have nearly doubled its valuation from the previous round in November 2025.

我们也在思考，来自 Anthropic 的 Claude Code、OpenAI 的 Codex，还有像 Lovable 这样的其他编程创业公司的竞争压力正在加大。

And we were also thinking about increasing competition from not only Anthropic's Claude Code and OpenAI's Codex, but also other vibe coding startups like Lovable.

所以我们想搞清楚，投资者到底对 Cursor 有多担忧，他们是否认为这家公司在如此激烈的竞争下还能长久生存下去。

So we really sought to figure out why investors were wary of Cursor, whether they thought the startup had durability and emote in light of this increased competition.

明白了。

Okay.

这些问题都问得非常好。

So there's a lot of great questions there.

那我们一个一个来梳理一下。

So let's go through them part by part here.

我们先从 Cursor 的财务状况说起。

So let's start with Cursor's financial profile.

这方面你们发现了什么？

What did you find there?

是的。

Yeah.

我们发现，虽然 Cursor 的收入还在增长，上个月它的年化收入已经达到了27亿美元。

So I mean, we found that while Cursor's revenue is still on the up, I mean, the company hit $2,700,000,000 in annualized revenue last month.

上一财年它创造了约7.7亿美元的收入，比上一财年增长了24倍。

It generated about $770,000,000 last fiscal year, which represented a 24 times increase from the previous fiscal year.

但公司的成本依然很高，截至今年1月的季度，毛利率为负23%。

The company still had really significant costs and its gross margins were negative, negative 23% as of the quarter ended in January.

我们听说他们之后已经转正了，但那些负的毛利率可能让一些潜在投资者感到担忧，他们原本打算参与最近这次融资，但最终取消了。

We've heard that they've since turned positive, but those negative gross margins may have been a concern for potential investors who were floated the most recent now called off fundraise.

我们还发现，那些原本拒绝投资的顶级投资方，对竞争和资金限制一直心存顾虑。

And we also found that top investors who had passed on the deal were wary of competition and capital constraints.

这些问题是SpaceX收购案可能解决的。

These are things that the SpaceX acquisition could solve.

收购能解决资金短缺的问题。

They would solve the funding issues.

同时也能帮Cursor突破算力瓶颈，因为他们能用上SpaceX庞大的服务器集群，但毛利率低始终是个关键障碍。

They'd also solve Cursor's compute hurdles since they get access to SpaceX's vast servers, but those gross margins were really a sticking point.

我不想轻描淡写地带过这一点。

I don't want to gloss over this point.

这家公司当时正试图完成一轮融资，但进展不太顺利，对吧？

The company was trying to complete a funding round, and what, was not having much luck there?

那次融资到底遇到了哪些问题，才让公司最终转向了SpaceX的收购方案？

What were the issues with that funding round that maybe prompted it towards this SpaceX deal?

所以这个融资轮，说白了，直到SpaceX的交易出现之前，一直都在进行中。

So the funding round, I mean, it was still on for all intents and purposes until the SpaceX deal came in.

我觉得SpaceX这笔交易给Cursor提供了一个获取海量算力的机会，就像我同事今天在另一篇报道里提到的，这对初创公司来说一直是大问题。

I think that the SpaceX deal just gave Cursor an opportunity to access a ton of compute, which is, as my colleagues reported in another story today, has been a huge issue for startups recently.

我觉得这正是推动Cursor接受收购的关键原因。

I think that that was a really big driver for Cursor towards the acquisition.

你刚才提到的另一个问题是，Cursor在竞争中表现如何，尤其是Claude Code和Anthropic，它们是这个领域的主要对手。

So now the other question that you said you were interested in is how Cursor is stacking up against competition, namely, I mean, Claude Code and Anthropic is the big competitor in the space.

客户是怎么看待它在竞争中的地位的？

How do customers see it as a competitor there?

投资者看到的，跟你我看到的一模一样，Claude Code的增长速度真的很快。

I mean, investors are seeing exactly what you and I are seeing, and Claude Code is growing really fast.

它只用了六个月就实现了十亿美元的年化收入，而Cursor花了差不多一年才达到这个数字。

It took it only six months to reach $1,000,000,000 in annualized revenue, and it took Cursor roughly a year to reach that same number.

规模也很重要。

And scale matters too.

Anthropic 在算力和研发上的投入能力远超 Cursor，而且 Cursor 的编程工具底层用的还是 Anthropic 的模型。

Anthropic can outspend Cursor on things like compute and R and D and Cursor's coding tools are powered by Anthropic's models.

不过这家公司最近也开发出了自己的编程模型。

And the company's recently developed its own coding model.

但不管怎样，这种局面可能会让投资者感到棘手，更不用说像 Lovable 这样的其他编程初创公司也在这个赛道里竞争了。

But regardless, this dynamic could create real hang ups for investors and not to mention all the other coding startups like Lovable that are in the mix here as well.

确实。

Right.

这点说得挺有道理。既然 Cursor 一开始就依赖 Anthropic 的模型，那 Anthropic 完全有可能在某些方面超越 Cursor，这确实是个潜在的风险。

And I mean, make a good point, which is that if Cursor is relying on Anthropic's models to begin with, I mean, here's a scenario where Anthropic, I'm sure, could very easily just out innovate Cursor in some way.

我这是推测啊，对吧？

I'm guessing here, right?

如果 Anthropic 想这么做，他们完全可以在竞争上更激进一些。

If Anthropic wanted to, they could go a lot harder on the competition in some ways.

其实我们已经在某种程度上看到这种情况了。

I mean, well, we've kind of been seeing that already.

自从去年年底人们开始热议Claude Code以来，它的增长简直势如破竹。

Just Claude Code has had such an explosive growth since people really started talking about it late last year.

所以我认为，投资者在评估这家公司时，心里肯定都惦记着这一点；而公司自己在规划未来时，也在权衡是继续独立发展、再融资，还是接受被SpaceX收购的可能性。

So I think that that is something that's definitely in the back of investors' minds as they're thinking about the company and the company was thinking about it as well when they were weighing their future prospects, whether to stay the course and raise more money and still operate independently or to have this right to be acquired by SpaceX sometime in the future.

对。

Right.

咱们再聊聊那个被收购的权利吧。

Now let's talk once again about that right to be acquired.

如果Cursor真的以六百亿美元的价格被收购，那在Cursor的股东名单里，谁是最大的赢家？

So if Cursor does get bought for that $60,000,000,000 price tag, who are the big winners here in the Cursor cap table?

安德森·霍洛维茨和Thrive Capital有望获得非常可观的回报。

Andreessen Horowitz and Thrive Capital stand to make pretty substantial returns.

这两家机构从A轮就投资了，当时公司的估值大约是四亿美元。

Both have been investors since the Series A, which was priced at a roughly $400,000,000 valuation.

他们的收益可能高达数十亿美元。

Those windfalls could be in the billions.

还有其他投资方，比如Excel和Benchmark，虽然进场晚一些，但如果收购价达到六百亿美元，他们也能获得可观回报。

Then there's others as well, Excel and Benchmark that got in a little bit later, but stand to benefit too from a $60,000,000,000 acquisition price.

最后一个问题，朱莉娅。

And last question for you, Julia.

你去做了这则报道，我想知道，你觉得这笔交易反映出AI编程领域整体和并购市场的大趋势是什么？

I mean, you went out reporting the story, I wonder what you think this deal says about the AI coding story at large, the M and A landscape at large.

能不能跟我分享一下，你在采访过程中对这件事有哪些更深层的观察？

I mean, just walk me through some of the broader insights that you had from reporting this story.

我们最大的一个发现是，算力获取对初创公司来说是个关键瓶颈，极其重要。

I One of our big takeaways is compute access is a huge bottleneck for startups and is super important.

正如我提到的，你也可以从我同事今天发布的另一篇报道中看到，很多初创公司正面临算力短缺的问题。

As I mentioned, you can see that in another story that my colleagues published today about startups facing a compute crunch.

我觉得投资者也在思考应用层和垂直整合之间的区别。

I think that investors are also thinking about app layer versus vertical integration.

拥有模型、基础设施，以及那些可能越来越重要的应用程序，对投资者来说会变得愈发关键，尤其是那些担心像Anthropic这样的对手会带来竞争的人。

So owning models, infrastructure, owning the apps that may become increasingly important to investors, especially those wary of competition from players like Anthropic.

我觉得随着局势继续发展，AI领域，尤其是应用层，还会出现更多整合。

And I think that we can expect to see even more consolidation across the AI market, especially at the app layer, as things continue playing out.

明白了。

Great.

朱莉娅，非常感谢你今天来做客。

Well, Julia, I want to thank you for coming on.

这位是朱莉娅·霍恩斯坦，我们《信息》杂志的风投记者。

That is Julia Hornstein, our venture capital reporter here at The Information.

今天的节目就到这里。

That does it for today's show.

提醒一下，我们每周一到周五上午10点（太平洋时间），下午1点（东部时间）直播。

A reminder, we are on this stream Monday through Friday at 10AM Pacific, 1PM Eastern.

如果错过直播，节目内容也可以在theinformation.com、我们的YouTube频道，或者你常用的播客平台收听。

If you can't make it then, episodes are available on theinformation.com, our YouTube channel, or wherever you get your podcasts.

别忘了在X、Instagram和TikTok上关注我们。

Make sure to follow us on social media on X, Instagram, and TikTok.