OpenAI 提出向美国提供5%股权，Anthropic 和三星计划开发芯片

OpenAI 正在向美国提供一个主权人工智能财富基金的5%股份，而伯尼·桑德斯则主张征收50%的税。

OpenAI is offering The United States a 5% stake in a sovereign AI wealth fund, and Bernie Sanders is pushing for a 50% tax.

Cursor 与 SpaceX 达成的600亿美元交易，让它们与 OpenAI 和 Anthropic 的关系变得微妙了，毕竟现在它们被其中一个主要竞争对手收购了。

Cursor's $60,000,000,000 SpaceX deal is putting their OpenAI and Anthropic relationship into question as that's getting a little bit more tricky now that they've been purchased by one of their biggest competitors.

Anthropic 目前正与三星洽谈，打算开发一款定制AI芯片，而Meta则把其智能眼镜放在了Meta One高级订阅服务后面。

Anthropic is currently in talks with Samsung to create a custom AI chip, and Meta is putting their smart glasses behind a Meta One premium subscription.

人们愿意为此买单吗？

Will people pay?

这才是关键问题。

That is the question.

微软刚刚向Frontier投资了25亿美元。

Microsoft has just committed 2,500,000,000 to Frontier.

这是一家新的AI部署公司。

This is a new AI deployment company.

这其实并不新鲜。

This is not new.

OpenAI 和 Anthropic 都干过同样的事。

We've seen the same thing from OpenAI and from Anthropic.

如果你想用 AI 来发展和扩大你的生意，我有个免费的社群叫 AI Hustle，现在有四百多个成员，大家都在分享怎么用 AI 工具把生意做起来。

If you wanna use AI to grow and scale your business, I have a free school community called AI Hustle where we have over 400 members, and we're all sharing how we're growing and scaling our business using AI tools.

我和朋友 Jamie McCulley 每周都会发视频，拆解我们自己是怎么在生意里用这些工具的。

My friend Jamie McCulley and I, we post a video every single week on there where we break down how we're using it inside of our own businesses.

我们会公开收入、数据、具体数字，还有全程用的所有软件。

We share revenue, we share metrics, we share numbers, and all of the software we use along the way.

如果你想看一百五十多个视频，我们有个付费专区，每月二十块；但你先免费看五段视频，还能加入社群和每周的直播，去 school.com/aihustle 就行。

If you wanna get access to over a 150 videos, we have a premium section for $20 a month, but you can get five videos for free and access to the school community and our weekly live streams by going to school.com/aihustle.

我会在描述里放个链接，帮你用 AI 工具把生意做大做强。

I'll leave a link in the description to help you grow and scale your business using AI tools.

OpenAI 打算给美国政府——其实理论上是给整个美国——5%的股权。

OpenAI is offering to give the United States government, and I think really kind of the entire United States theoretically, a 5% equity stake.

这跟阿拉斯加的做法有点像。

This is kind of the same as Alaska.

他们有个石油财富基金。

They have an oil wealth division fund.

这其实跟特朗普政府早前的一些讨论有关。

This is parts in of, you know, some earlier talks that the Trump administration was doing.

而现在，这个提议正好夹在我们正看到的几场大规模公众压力运动中间。

And right now, this is also sitting right between some big public pressure campaigns we're seeing.

参议员伯尼·桑德斯有个不同的方案。

We have senator Bernie Sanders that is you know, he has a competing plan.

他说他希望对所有领先的AI公司一次性征收50%的税，直接拿走一半的股份。

He says he would like 50%, a one time tax on all of the leading AI firms where they just give, 50%.

所以我觉得，OpenAI愿意拿出5%，其实是因为桑德斯那边提出的50%太吓人了。

And so I think OpenAI is, like, happy to offer 5% when the Bernie Sanders 50% is, the other offer being proposed.

桑德斯的方案如果实施，通过一次性征收50%的股票税，能筹集大约7万亿美元，这笔钱会直接用于医疗、住房和其他桑德斯认为合适的项目。

Bernie Sanders proposal would raise about $7,000,000,000,000 through a one time 50% stock tax, which is 50% of the company, and the proceeds and direct payments, would go to health care and housing and other, you know, programs that Bernie Sanders deems, fit.

特朗普政府也单独找过谷歌和Meta，谈过类似的股权分配方案。

The Trump administration has separately approached Google and Meta about a very similar equity stake.

我认为这两家公司都还没公开表示会签署这样的协议。

I think neither of those companies has publicly said that they are going to sign any sort of agreement like that.

现在有七成美国人反对在本地建设AI数据中心，还有五成更多的人对AI感到担忧，而不是兴奋。

And I think 70% of Americans right now oppose AI data centers in the area, 50% more have concern about excite rather than excitement about AI.

如果你看看国内的一些民调，就会发现公众对AI的负面情绪相当强烈，我觉得这正是OpenAI和特朗普政府想借机缓解的。

So if you look at some of the polling that's going on inside of the country, there is, you know, a lot of negative sentiment that goes towards AI, and I think that's what OpenAI and maybe the Trump administration is trying to help.

伯尼·桑德斯只想直接拿走一半。

Bernie Sanders, wants to just take 50% of it.

最终会是哪种方案胜出，我们拭目以待。

We'll see which of these two models plays out in the end.

Cursor被SpaceX以600亿美元收购了，但众所周知，它的两大主要合作伙伴——不管是供应商还是客户，你怎么叫都行——是OpenAI和Anthropic。

Cursor was acquired for $60,000,000,000 by SpaceX, but, you know, famously, its two biggest, I guess, suppliers or customers or however you wanna call it are OpenAI and Anthropic.

对吧？

Right?

Cursor依赖OpenAI和Anthropic来驱动他们的工具，或者允许开发者使用它们。

Cursor uses OpenAI and Anthropic to power their tools or they let developers do that.

所以现在有人开始质疑，这种合作关系还能不能继续下去。

And so people are calling into question if this relationship can continue.

对吧？

Right?

如果Cursor被SpaceX收购了，而XAI的代码又要主导Cursor的运作，那它还能跟OpenAI和Anthropic保持良好关系吗？

If Cursor was purchased by SpaceX, which is gonna have XAI and XAI code running things inside of Cursor, can they keep, you know, a good relationship with OpenAI and Anthropic?

我们之前也见过不少公司，成长起来后被收购，结果都经历了类似的棘手局面。

We've seen other companies that have grown and been bought out, go through, you know, very similar challenging relationships.

Scale AI被Meta收购了50%的股份，他们的CEO后来还去负责Meta AI了。

Scale AI was purchased or 50% of it was purchased by Meta, and their CEO was brought on to run Meta AI.

结果一出事，他们最大的客户Google和OpenAI立马就断了合作，因为成了竞争对手。

And when that happened, some of their biggest customers were Google and OpenAI, and both of those companies cut ties with them because of the competition.

Cursor已经在SpaceX的基础设施上训练自己的下一代模型了。

Cursor is already training their own next model on SpaceX infrastructure.

他们现在拥有的算力是以前的十到二十倍，全都是SpaceX给的。

They have 10 to 20 times more compute than they previously had because SpaceX gave it to them.

所以很明显，马斯克正在投资打造他们自己的原生能力。

So, obviously, Elon Musk is investing in building their own native capabilities.

但就编程工具而言，我觉得这个先例比Scale AI的情况更有说服力，因为去年OpenAI收购了另一个非常知名的编程工具Windsurf。

But the precedent for these coding tools specifically, I think, is stronger than just what happened to Scale AI because Windsurf, another very famous, coding tool, was purchased by OpenAI last year.

收购之后，Anthropic就切断了他们对Claude的访问权限。

And when that happened, Anthropic cut off their Claude access.

所以我觉得这样的先例已经很多了。

So I think there's a lot of precedent.

Cursor很可能会被切断对OpenAI或Claude的访问权限。

It's very possible that Cursor is gonna get their OpenAI or or Claude access cut off.

但这个情况稍微复杂一点，因为Cursor在业内几乎就是编程的代名词。

But this one seems a little trickier because Cursor really is just synonymous with coding in the industry.

我的意思是，除了Claude Code和可能的Codex之外，它就是目前最大的编程工具了。

I mean, outside of Claude code and maybe Codex, like, this is the biggest coding tool that exists out there.

如果他们真把访问权限切断了，我会很震惊，因为这会失去大量用户，而这些用户原本一直在为OpenAI和Anthropic免费输送收入，尤其是Cursor还签了不少企业合同。

I would be, you know, shocked if it you know, if they really did cut it off because that would be a ton of users that no longer would be would be, you know, essentially giving free money to OpenAI Anthropic, especially because, Cursor signed up with a lot of enterprise contracts.

其实挺有意思的是，OpenAI的初创基金当初是Cursor的早期投资人，现在他们打算把股份换成SpaceX的股票。

And actually, one really funny turn of events in all of this is that OpenAI startup fund, they were actually an early investor in Cursor, and they're gonna convert their stake into SpaceX stock.

所以在这整个收购潮里，马斯克和阿尔特曼之间关系微妙，OpenAI不仅会直接从Cursor的成功中获利，连XAI也沾光，尽管Codex明明是直接竞争对手。

So with all of this, acquisition that's going on, Elon Musk and Sam Altman, both have a lot of beef, OpenAI is going to get a direct financial stake into Cursor's success and also XAI even though Codex is a direct competitor.

这事儿真是有点讽刺。

So that's kinda funny.

现在Anthropic正在酝酿一个大动作。

So there is a big move happening with Anthropic right now.

他们正和三星洽谈，打算联手开发一款定制AI芯片。

They're in talks with Samsung to create a custom AI chip.

他们加入了OpenAI、谷歌、亚马逊的行列，都在试图减少对英伟达的依赖，毕竟英伟达的销量正创历史新高。

They're joining OpenAI, Google, Amazon, all of them that are trying to basically reduce their reliance on Nvidia as Nvidia is, you know, making record sales.

眼下这么做，既是为了控制成本，也是为了缓解供应链压力。

Right now, doing all of this is both kind of targeted at costs and supply constraints.

一旦这些公司自己设计芯片，就能更好地优化自己的模型架构。

When you make this custom silicone, it's in, you know, when these labs are creating it themselves, it's gonna let them optimize their own architectures.

而且他们还能借此对任何一家主导供应商施加谈判压力。

And also, they can negotiate leverage against any sort of dominant supplier.

对吧？

Right?

长期以来，英伟达一直占据主导地位。

So Nvidia has been very dominant for a long time.

事实上，当你想从英伟达采购显卡之类的产品时，他们会按一定配额分配给不同公司，因为他们得尽量把有限的产能分摊开——毕竟需求实在太大了。

And in fact, when you're trying to go and get GPUs or whatnot from Nvidia, they have a certain amount that they'll allocate to different companies, but they're trying to spread it out because the the demand is just so high.

这问题不小，像OpenAI、Anthropic、谷歌这些公司，能拿到的英伟达芯片数量都是有上限的。

So this is not great because if someone like OpenAI, Anthropic, Google, they only can get a maximum amount from Nvidia.

如果能自己造，不仅能缓解这种供应紧张，还能提升自己的议价能力。

So if they can build their own, not only does that kind of ease that constraint, that supply constraint, but also, you know, maybe they have some better negotiating power.

我们自己造，成本可以控制在某个价位，你们要是能降价，那大家都有得谈。

Look, we can build our own for x, you know, price if you would be able to bring the price of yours down.

所以我觉得，这本质上也是为了在和英伟达的谈判中争取更多主动权。

So I think it's kind of a long term play for negotiating power with Nvidia as well.

OpenAI 刚刚发布了他们自己的定制芯片。

OpenAI also just unveiled their own custom chip.

这款芯片叫‘Jalapeno’。

It's called a Jalapeno.

他们上周才公布的。

They did this last week.

他们和博通合作开发这款芯片，声称能比其他竞争对手提供更高的能效比。

They're building that with Broadcom, and they're claiming that it's gonna offer better performance per watt than any of the other competitors.

当你为自己的软件和硬件栈专门设计芯片时，能实现这样的效率其实很自然。

And I think when you're, you know, when you're building this custom for your software or hardware stack, it's, you know, it makes a lot of sense that they're able to achieve those kind of efficiencies.

我觉得 Anthropic 也注意到了这一点，正在跟进做同样的事。

So I think Anthropic has seen this and trying to do the same thing.

三星本来就在为 Nvidia 生产芯片，用于他们的 AI 训练。

Samsung already manufactures Nvidia's chips, and they do that for their AI training.

他们现在正和 Nvidia 在韩国合建一座专门的 AI 芯片工厂。

And they're also building a joint AI chip factory in South Korea with Nvidia right now.

所以，我觉得他们正好处在供应链的核心位置。

So, you know, I think they're kind of at the center of the supply chain.

三星正在为英伟达代工芯片。

Samsung's working Nvidia.

因此，Anthropic 选择三星作为合作伙伴来开发这款芯片，几乎是顺理成章的事。

So it's kind of a no brainer for Anthropic to go there as a partner to build out this chip.

不过，Anthropic 的这个项目还处于非常早期的阶段。

Anthropic's project though is definitely in the early stages.

这些芯片不是能快速造出来的。

These chips are not quick.

研发周期特别长，而且服务器的用途、性能目标这些关键点都还没定下来。

They take a very long time to build, and the purpose, the server role, the performance targets, none of that have actually been finalized.

所以你可以想象，这至少得是个好几年的开发过程。

So I think you can kind of imagine this is gonna be a multi year development timeline.

我早就听说 OpenAI 在搞这些东西好几年了，现在看来，Anthropic 才刚意识到，他们也必须走这条路。

I mean, I've been hearing, you know, rumors of OpenAI working on this stuff for many years, and it seems like Anthropic is just getting there and realizing this is something that they need to do.

另外，Meta把智能眼镜的高级功能都放到付费订阅里了。

In other news, Meta is putting their smart glass features behind a premium subscription.

这还挺有意思的，对吧？

It's kind of interesting, right?

Facebook一直给人的印象是免费 forever，靠广告赚钱，但现在他们开始推出新产品，还尝试搞付费订阅模式。

Facebook has kind of always been known as the free forever, they just monetize through ads, but they're starting to spin out these new products, and they're trying to do premium subscription models.

如果你想用智能眼镜的所有最新功能，就得付订阅费，起步功能是对话增强，主要提升在嘈杂环境里的语音清晰度。

So if you wanna get all of the latest features with their smart glasses, you're gonna have to pay a subscription that's starting with conversation focus, which is basically gonna enhance voice clarity and noisy settings.

我听说这个功能是这么宣传的：比如你在一个吵闹的火车站，周围人声鼎沸。

So I've heard this kind of pitched as like if you're in a noisy train station and there's a ton of people talking.

如果你在跟人说话，耳机就会把对方的声音单独提取出来，放大给你听。

If you're if you're talking to someone, their voice is gonna get isolated and amplified by the speakers in your headset.

这样你就能听得更清楚了。

So you can actually hear them better.

你只要朝对方的方向看一眼，他们的声音就会自动盖过周围所有的杂音。

All you have to do is look in their direction, and you'll be able to hear, you know, the sound of their voice louder than all of the noisy surroundings.

但有意思的是，他们还是提供了一些免费额度。

But what's interesting, they do give you some free.

每个月给你三小时免费，或者花十美元买 Meta One 高级版，能用十五小时。

They give you three months for or three hours a month for free, and they give you fifteen hours for $10 a month of Meta One Premium.

我不确定这个功能是手动开启的，还是系统会自动检测并提醒你用完了。

I'm not sure if you turn this on or if automatically detects and tells you when it runs out.

说白了，这跟其他平台的按使用量计费的 AI 服务没啥区别。

I mean, it's basically usage based AI like every other platform.

有意思的是，他们打算怎么让这体验顺滑？万一你正聊得开心，声音清晰得不得了，突然就断了呢？

It's just interesting, how they're gonna try to make this seamless, and if they're gonna just, you know, all of a sudden you're halfway through a conversation, everything's being perfectly amplified, and then it cuts out.

对吧？

Right?

我是说，你真能在一个嘈杂的火车站或者工地里，一个月聊超过十五个小时的天吗？

So, mean I and I don't know, like, how many conversations if you're having more than fifteen hours of conversations in noisy train stations or maybe on, like, a noisy job site.

挺值得看看这功能最后会在哪些地方推出来。

It'll be interesting to see where this gets rolled out.

这明显是个重要的盈利策略。

This is definitely a big monetization play.

可不是为了回收成本那么简单。

It's not kind of a cost recovery one.

这个功能完全在设备上运行，Meta根本不用付服务器费用。

The feature runs entirely on device, so there's no server cost to Meta.

他们就是想靠这个赚钱。

So they're just trying to generate revenue from this.

我觉得这说明，AI硬件在用户买下眼镜之后，还能持续挖掘价值——这意味着他们甚至可以亏本卖眼镜，只要能靠这种对话增强功能，再加上未来可能推出的其他AI功能来赚钱。

I think this shows that AI hardware is gonna get a lot of, value that they can extract after their users already have the glasses, which means they can basically sell their glasses at a loss, assuming they have, you know, this conversation focus and probably a bunch of other AI features they'll roll out in the future.

尤其是如果所有运算都能在设备上完成，他们就不用承担额外成本。

Especially if it can be computed on device, they're not incurring any extra cost for it.

你只要付费就行。

You just pay for it.

说实话，这让我想到特斯拉的自动驾驶，技术早就装在车里了，你只要付钱，功能立马解锁。

And, honestly, it kinda reminds me of Tesla and their self driving, how, like, the technology is already in the Tesla cars, and if you just pay for it, then all of a sudden it's enabled.

Meta新推出的299美元品牌眼镜是按盈亏平衡价销售的，他们真正想做的是扩大用户基础。

Meta's new $299 Meta branded glasses are sold at breakeven, so they're really trying to grow their adoption.

以前如果你想要Meta版的雷朋眼镜，还得再多花一百美元，但现在他们直接单独卖了。

You know, previously, it was about a $100 more if you wanted the Meta Ray Bans, but now they're just selling them by themselves.

这种产品的盈利模式非常薄，硬件本身几乎不赚钱，主要靠后续的软件订阅来驱动收入。

And there is a very razor thin, model of, you know, monetization on this, and then, basically, the hardware gets cost, and they're trying to get the reoccurring software subscriptions to drive all of the revenue.

他们不是唯一这么做的，今年谷歌也要推出竞争性的智能眼镜。

And they're not the only people doing this because Google is also launching a competing smart glasses this year.

谷歌是和三星、Warby Parker还有Gentle Monster合作推出这款产品。

They're partnering with Samsung to do that and Warby Parker and Gentle Monster.

我觉得谷歌在推理效率上的提升，能让它把更多功能打包进去，而且不太可能像Meta那样搞那么激进的分层收费。

Google's own inference efficiency gains, I think, are gonna let them bundle a lot more features than Meta is, and they're not gonna have as much of aggressive tiering, I don't think.

但也不是只有他们在这么做。

But they're not the only ones doing this.

苹果也在把不少AI功能锁在付费墙后面，谷歌那边也有类似的做法。

Apple is also having a bunch of their AI features behind a paywall, and Google has other ones.

谷歌的Pixel视频增强功能需要用户订阅他们的Pixel专属服务，而Gemini Live也得靠Gemini高级版才能用。

Google specifically with their Pixel video boost, it's gonna require a specific Google one tier on their Pixel devices, and Gemini Live is also behind the Gemini Premium.

就连苹果的iOS 27，也把不少照片编辑功能锁在了更高阶的iCloud+套餐里。

And if you even like with iOS 27 for Apple, they have a bunch of photo editing that is locked in higher iCloud plus plans.

所以你想用那些更厉害的功能，就得额外付费。

So if you want some of the better features, you're gonna have to also pay for that.

微软正在推出一家全新的AI部署公司，名叫Frontier。

Microsoft is launching a new AI deployment company called Frontier.

他们为此投入了25亿美元。

They're putting $2,500,000,000 into this.

这会是一个独立运营的部门，配备大约六千名工程师，直接派驻到各个企业，目标是端到端地部署AI系统。

It's gonna be its very own unit, and they're gonna have about 6,000 engineers that are gonna be gonna be directly embedded into different companies, and they're trying to basically deploy AI systems end to end.

这其实和OpenAI的做法类似，他们也拆分出独立的公司，吸引外部投资。

So this is something that we've seen OpenAI also spin out like separate companies and get investors in these separate deployment companies.

Anthropic也这么干了，还和不少大公司展开了合作。

Anthropic has also done this, and they're partnering with a bunch of big players.

这挺有意思的。

So it's interesting.

我觉得这些实验室下一步想增加企业客户和销量，就是直接派工程师过去，要么免费，要么成本价，甚至收点钱，然后进去帮企业把微软、OpenAI 或 Anthropic 的系统全都部署起来，尽可能提升使用量。

It feels like the next step for these labs is they're trying to get more sales and more users in companies is to just basically get these engineers to and they're, you know, sending them in there for at cost or free or charging them something, and they're gonna go and implement all of the Microsoft or all of the OpenAI or all of the Anthropic systems inside of that company and try to get their usage up as high as possible.

现在和 AWS 之间正展开一场激烈的部署竞赛。

There is a huge development deployment race right now with AWS.

就在两天前，他们刚宣布了 10 亿美元的竞品项目，我觉得这两家公司都在赌：企业真正需要的是有人手把手帮他们落地。

They announced a rival $1,000,000,000 venture just two days earlier, and I think both of these companies are betting that enterprises need hands on implementation help.

这可不是简单地说：‘你去注册 OpenAI，让团队用起来’就完事了。

So it's not just about, you know, hey, go sign up for OpenAI and let your team use it.

而是说：‘我们直接派工程师上门，帮你把系统真正搭起来。’

It's like, look, we're gonna give you these engineers that are gonna go and actually implement this for you.

不过说句公道话，Frontier 拥有六千名工程师，规模是 Palantir 上市时整个员工团队的六倍，也几乎是 AWS 公开承诺投入的两倍。

Just to put that into context though, with their 6,000 engineers, Frontier is about six times larger than Palantir's entire FTE organization when they first IPO'd, and it's about double AWS's announced commitment.

所以说，这对他们来说是个大动作。

So, I mean, this is a huge move for them.

他们已经有不少早期合作伙伴，包括伦敦证券交易所集团、联合利华、Land O'Lakes 和埃森哲。

They have a bunch of early partners, including the London Stock Exchange Group, Unilever, Land O'Lakes, and Accenture.

他们的客户中有一些是数据量大、供应链复杂的大型系统集成商。

They have a mix of some of the big data heavy supply chain system integrator customers.

微软规模这么大，这根本一点都不意外。

I mean, Microsoft's massive, so this is not a shocker in any way.

不过，他们现有的 Azure 和 Microsoft 365 客户关系，已经让不少财富 500 强企业里都嵌入了微软的工程师。

But their existing Azure and Microsoft three sixty five relationships, I think, give them a lot of embedded engineers already inside of most of the Fortune 500 companies.

这一点让他们在对抗 AWS 时拥有巨大的结构性优势，因为 AWS 得从零开始搭建这套体系。

This is a big structural advantage they have over AWS, which is gonna have to build this from the ground up.

所以，没错，微软凭借 Azure 一直以来的布局，确实占了先机。

So, yeah, Microsoft has kind of a leg in this from everything they've been doing with Azure.

如果你们想每月花 8 美元就能访问前 80 款主流 AI 模型，不妨去 aibox.ai 看看。

If you guys wanna get access to the top 80 different AI models for just $8 a month, go check out aibox.ai.

我可以让你访问超过 80 款不同的 AI 模型，涵盖文本、图像、音频、视频，甚至音乐生成。

I give you access to over 80 different AI models, everything from text, image, audio, video, even music generation models.

你得到的基本上是所有顶尖模型，OpenAI、谷歌、Anthropic、Eleven Labs、Claude，全都有，每个月才8.99美元。

You get I mean, it's basically all the top models, OpenAI, Google, Anthropic, eleven Labs, Claude, all of that, and it's only $8.99 a month.

如果你想在一个平台上一次性接入所有这些服务，还只付一份订阅费，那就去看看我自己的创业项目，aiBox.ai。

So if you wanna get access to all of that in one platform for one subscription, go check out my own startup, ai Box dot AI.

我会在描述里放个链接，祝大家接下来的一天都过得特别好。

I'll leave a link in the description, and I hope you all have a fantastic rest of your day.