20VC：英伟达 vs Groq：训练与推理的未来 | Meta、谷歌和微软的数据中心投资：谁将胜出 | 数据、算力、模型：AI的核心瓶颈及价值分布，专访Groq创始人乔纳森·罗斯

所以占用空间更大。

So the footprint is higher.

把它想象成工厂和后院车库之间的区别。

Think of it as the difference between a factory or a backyard sort of garage.

后院的车库效率不会那么高。

The backyard garage is not gonna be as efficient.

然而，它的能耗更低。

However, it has a lower energy footprint.

或者另一个例子：如果你想把一吨煤从城市这一端运到另一端，是用摩托车运，还是用货运列车运，哪种更高效？

Or another example would be if you're trying to transport a ton of coal from one side of the city to the other, and you did it on mopeds, or you did it with freight trains, which one would be more efficient?

摩托车每次运输耗能更少，但需要更多次运输，因此总体耗能更多。

The moped would use less energy per trip, but it would need more trips and therefore would use more energy overall.

事实上，这是大多数人误解的地方。

In fact, this is one of the things most people misunderstand.

他们以为边缘计算的能耗更低。

They think that edge computing is lower energy.

实际上，边缘计算比在数据中心计算能效更低。

Actually, edge computing is less energy efficient than computing in the data center.

当你在数据中心进行计算时，这有点像货运列车。

When you're computing in the data center, it's a little bit like that freight train.

你可以同时处理大量任务。

You're actually getting to do a whole bunch of jobs simultaneously.

因此，我们无需从外部内存读取数据，也就省下了这部分能耗。

So the fact that we don't have to read from that external memory means that we don't have to spend the energy doing that.

即使使用GPU，你也能进行批量处理。

Even with GPUs you get to batch.

但回到为什么它能效低的问题上，在芯片中消耗的能量主要来自这些物理导线。

But going back to why it's so energy efficient, the amount of energy used in a chip, there's these physical wires.

这些物理导线有宽度，当你观察其宽度和长度时，需要给导线充电以设置为1，然后再放电以设置为0。

And the physical wires have a width, and when you look at the width and you look at the length, you charge that wire up to set it to a one, and then you discharge it to set it to a zero.

这就像给电容器充电和放电，都需要消耗能量。

It's sort of like charging a capacitor and discharging a capacitor using energy.

导线越长，需要的电荷就越多。

The longer that wire, the more charge.

当这里有HBM，另一个芯片在那边时，每次传输一位数据，你都必须在芯片之间充电并放电。

When you have HBM here and another chip here, you're actually having to charge a wire between the chips and then discharge it every time you send a bit.

这段距离很长，而且导线比芯片内部的导线更粗。

That's a long distance to travel, but also the wires are wider than the wires that are inside the chips.

因此你消耗了多得多的能量。

So you just use a lot more energy.

当我们把内存保留在芯片内部时，数据只传输很短的距离，使用细得多的导线，因此能耗低得多。

When we keep that memory in the chip, it's only traveling a little distance using much thinner wires, and therefore it uses a lot less energy.

那么，我们是否能看到LPUs和GPUs混合使用的场景？这两种技术的使用比例会是怎样的？

So do we see a world of LPU and GPU usage in combin like how how does that distribution look between LPU usage and GPU usage?

这有

There's a

几个方面。

couple of things.

首先是，训练应该在GPU上进行。

The first is training should be done on GPUs.

NVIDIA会卖出他们生产的每一颗用于训练的GPU。

NVIDIA will sell every single GPU they make for training.

目前，他们约40%的市场份额来自推理任务。

Right now, about 40% of their market is inference.

如果我们部署大量成本低得多的推理芯片，你会看到同样数量的GPU会被售出，但对训练的需求会增加，因为推理越多，所需的训练也越多，反之亦然。

If we were to deploy a lot of much lower cost inference chips, what you would see is that same number of GPUs would be sold, but the demand for training would increase because the more inference you have, the more training you need and vice versa.

另一个应用场景是，我们的速度实际上比GPU快得多。

The other use case is we're actually so crazy fast compared to GPUs.

我们已经做了一些实验，将模型的一部分放在我们的LPU上运行，其余部分仍在GPU上运行。

We've actually experimented a little bit with taking some portions of the model and running it on our LPUs and letting the rest run on GPU.

这实际上加快了速度，并让GPU的使用更经济。

And it actually speeds up and makes the GPU more economical.

由于人们已经部署了大量GPU，我们考虑的一个应用场景是销售一些我们的LPU来实现加速提升。

So since people already have a bunch of GPUs they've deployed, one use case we've contemplated is selling some of our LPUs to sort of nitro boost

那些GPU。

those GPUs.

好吧，这正是我想问的，

Well, this is my question, is that,

你知道，人们早就买好了GPU，是的。

you know, people have bought GPUs so far ahead of time Yeah.

等你拿到的时候，它们已经被部署安装了，几乎已经过时了。

That by the time you get them, they're deployed and installed, they're almost out of date.

实际上，我们已经和一些客户沟通过，他们提前一年多就下了订单。

Actually, we've we've spoken to some customers that put orders in over a year in advance.

他们提前一年付款，但至今还没收到货。

They paid a year in advance and still haven't gotten them.

我们在沙特阿拉伯最近的一次部署，从签约到在本国生产环境中首次生成token，只用了51天。

The recent deployment we did in in Saudi Arabia, fifty one days from contract to the first tokens being served in production in country.

你们是怎么做到的

How are you able to do

这么快？

it so quickly?

五十一天令人震惊。

Fifty one days is astonishing.

是的。

Yeah.

部分原因是我们在架构上更加简洁。

Part of it is architecturally, things are much simpler for us.

我们不需要一大堆其他硬件组件。

We don't have a bunch of other hardware components.

我们实际上不使用交换机来连接我们的芯片。

We actually don't use switches to communicate between our chips.

我们直接将芯片插到芯片上。

We just plug our chips into our chips.

我们的芯片就是交换机。

Our chips are the switch.

我们没有

We don't have

所有这些网络调优工作。

all of this network tuning.

考虑到能效和可预测性，为什么英伟达在LPU上不更积极一些？

Given the energy efficiency, given the predictability, why is NVIDIA not being more proactive on LPUs?

你凭什么认为他们不想在这件事上更积极呢？

What makes you think that they don't want to be more proactive on it?

他们从不谈论它。

They don't talk about it.

那他们为什么要谈论它呢？

Well, why would they talk about it?

那就像在试图展现实力而非脆弱时，故意谈论自己根本没有的东西。

That would be like talking about something you don't have when you're trying to project strength rather than vulnerability.

想想看，如果你想要保护股东价值，想要维持华尔街对你占据主导地位、遥遥领先的形象

Well, think if you wanted to protect shareholder value and you wanted to protect a Wall Street image of dominance and being ahead of

如果你在打这场游戏，至少你会说，当然，我们也在研发LPU。

the game, you'd at least say, oh, we are of course working on LPUs as well.

但在他们拥有LPU之前，这样做实际上会暴露他们有所缺失。

But then, until they had LPUs, they would effectively be exposing that there's something missing.

比如，看看最近的GTC大会，他们宣布最新一代GPU比上一代快了30倍。

Like, if you look at the last GTC, there was an announcement that the latest GPUs were 30 x faster than the previous generation.

当你看它是如何实现的时，会看到一条曲线大致是这样的，然后就在这里戛然而止。

When you look at how it was done, there was this curve that looked kinda like this, and then it basically ended here.

接着又出现了另一条曲线，大致是这样的。

And then there was another curve that was kinda like this.

而这30倍的提升，是从这条曲线的末端到那条曲线的差距。

Now that 30 x was from the end of this curve to this curve.

如果你把比较点移到这里，提升就不到30倍了。

If you moved it here, it would have been less than 30 x.

如果你把比较点移到这里，提升就会变成无限大。

If you moved it here, it would have been infinite.

所以他们的芯片比前一代快得无限，但这样说听起来不太合理。

So their chip is infinitely faster than the previous one, but that wouldn't have sounded reasonable.

对吧？

Right?

这个市场一直有夸大宣传的历史，因为很难真正接触到芯片。

There's a history in this market of specsmanship because it's so hard to, like, get access to chip.

这是一个关于企业销售的教训。

And this is a lesson on enterprise sales.

在企业销售中，人们依赖夸大宣传，因为夸大宣传是什么？

In enterprise sales, people rely on specsmanship because Specmanship is what?

就是我的规格比你的规格更好。

It's well, my specs are better than your specs.

我的芯片比你的芯片更快。

My chip is faster than your chip.

我每秒获得的teraFLOPS比你多。

I get more teraflops per second than you do.

但谁在乎呢？

But who cares?

告诉我每美元能获得多少令牌，告诉我每瓦能获得多少令牌。

Tell me what the the tokens per dollar is, and tell me what the tokens per watt is.

其他一切都无关紧要。

Nothing else really matters.

但人们会找出所有这些其他奇怪的指标来证明自己更优秀。

But people will find all of these other weird things to measure that they might be better on.

就像我会卖给你一辆转速更高的车。

Sort of like, I'll sell you a car with better RPMs.

转速并不重要。

RPMs don't matter.

重要的是每加仑能跑多少英里，也许还有你能达到的速度，不过限速让这些也变得无关紧要。

What matters is miles per gallon and maybe the speed that you can drive at, although speed limits kinda render them, you know, moot.

对吧？

Right?

在企业销售中，曾经有一段时间，你会推销肥皂。

In the case of enterprise sales, there there was a time when you would market soap.

广告牌上会写着：我们的肥皂比其他品牌的肥皂泡沫更多。

The billboards would say, our soap has more bubbles than this other brand's soap.

谁在乎呢？

Who cares?

他们后来意识到，不如在广告牌上放一些使用肥皂后非常开心的人，这样人们可能会将这种快乐与产品联系起来。

And what they figured out was, let's put really happy people up on a billboard after they use the soap, and then maybe people associate that happiness.

对吧？

Right?

生活方式营销。

Lifestyle marketing.

当然。

Sure.

不知为何，企业界至今仍未吸取这个教训。

For some reason, enterprise still hasn't learned this lesson.

现在还是这样，我们的气泡更多。

It's still, we have more bubbles.

我们的TeraOps更多。

We have more TeraOps.

我们有更多各种各样的东西。

We have more whatever.

这些都是人们根本不在乎的东西。

Things that people just literally don't care about.

所以你觉得……

So you think

NVIDIA说‘我们快了30倍’，这并不是好的营销。

NVIDIA is, hey, we're 30 times faster is not good marketing.

它之所以有效，是因为人们已经习惯了，但我们的回应是发布了一篇新闻稿，上面写着：Grock，仍然更快。

It worked because it's what people are used to, but our counter was we we did a press release to that that said, Grock, still faster.

就这样。

That was it.

人们为此疯狂了。

And people went gaga over it.

对吧？

Right?

因为我们就只是。

Because it was just we we are.

我们仍然更快。

We're still faster.

所以谁在乎呢？

So who cares?

你觉得华尔街理解这种方式吗？

Do you think Wall Street understands that way?

我觉得他们开始明白了。

Think they're starting to.

是的。

Yeah.

但再说一遍，我不认为这里存在真正的竞争。

But again, I don't think there's real competition here.

我认为，如果你在竞争，那说明你已经犯了严重的错误。

I think if you are competing, you have done something seriously wrong.

如果你在竞争，那就意味着你还没有找到一个未被解决的客户问题。

If you're competing, it means that you haven't found an unsolved customer problem.

因为如果你在竞争，说明别人已经解决了这个问题。

Because if you're competing, someone else has already solved the problem.

那你为什么还要花时间在这上面？

So why are you spending time on it?

你不把英伟达视为竞争对手。

You don't view Nvidia as a competitor.

对。

No.

他们不提供快速令牌，也不提供低成本令牌。

They don't offer fast tokens and they don't offer low cost tokens.

这是一个非常不同的产品。

It's a very different product.

但他们非常擅长训练。

But what they do very, very well is training.

他们在这一方面做得比任何人都好，而且优势如此之大，这已经是一个被解决的问题。

Do They it better than anyone else, and by such a wide degree, it's a solved problem.

我们为什么要费心去解决一个已经被解决的问题呢？

Why would we bother trying to solve a problem that's already been solved?

所以你的想法是，让出训练市场给他们，我们来掌控推理市场。

So you're like, seed the training market to them, we'll own the inference market.

是的。

Yeah.

而他们说，我们也想要推理市场。

And they're saying, we also want the inference market.

当然。

Of course.

事情一向都是这样发展的。

The way it always works.

那我们现在该怎么办？

So what do we do now?

所以我们现在是在推理市场中竞争。

So now we are competing in the inference market.

但我们真的在竞争吗？

But are we?

是的。

Yeah.

我们并没有听到有人说要改买GPU而不是用你们的产品。

We don't really have people saying we're gonna buy GPUs instead of you.

我们确实听到有人说要同时购买两者。

We do have people saying we're gonna buy both.

这种情况确实会发生。

That happens.

但我们并不在意。

But we don't care.

我给一个人展示了演示，他说：我们是不是干脆别再买GPU了？

I showed a demo to someone, and he's like, should we just not buy any more GPUs?

我说：不，别这样。

I'm like, no.

你能搞到多少GPU就买多少。

You should buy every single GPU you can get your hands on.

他困惑地看着我，我说：你打算怎么训练？

And he's looking at me very perplexed, and I'm like, how are you gonna do training?

我们不做训练。

We don't do training.

去买GPU吧。

Buy the GPUs.

尽可能多买一些，因为我希望你们的模型能运行在我们的平台上

Get every single one you can because I want your models running on us to be

确实如此，但对于推理来说，他们

really Totally, but for inference, they

不再需要购买NVIDIA了。

don't need to buy NVIDIA anymore.

他们不需要为推理购买GPU，但如果你能搞到，我的意思是，它们有点贵，但如果你已经习惯了，为什么不呢？

They don't need to buy GPUs for inference, but if you can get them, I mean, they're a little expensive, but if you're used to it, why not?

仍然有很多人出售大型机。

Plenty people still sell mainframes.

如果你想要更低的成本和更快的速度，那就选择LPU。

If you want lower cost and faster, then you want an LPU.

成本低多少？

How much lower cost is this?

超过五倍更低。

More than five x lower.

超过五倍更低？

More than five x lower?

仅最新显卡的内存成本就超过了我们每块芯片部署的全部资本支出。

Just the memory alone in the latest GPUs costs more than our fully loaded CapEx per chip deployed.

此外，我们之前谈到了能效问题。

And on top of that, so we talked about the energy efficiency.

因此，我们每个词元消耗的能量大约只有三分之一。

So we use about a third of the energy per token.

在三年周期内，我们的运营支出占总成本的三分之一，主要是能源和数据中心租金，而资本支出占三分之二。

Over a three year period, one third of our cost is the OpEx, which is mostly energy and data center rent, and two thirds is the CapEx.

这意味着，由于我们的能耗只有三分之一，运行显卡生成相同数量词元的费用，就等于我们总的总成本。

Which means that since we're one third of the energy, the cost to run that GPU to produce the same number of tokens for inference is the same as our total cost.

仅显卡的运营支出，就等于我们的资本支出加上运营支出的总和。

Just the OpEx for the GPU is the same as our CapEx plus our OpEx.

真的很抱歉。

I'm really sorry.

我问的都是最蠢的问题，但我还是想继续问下去。

I'm I'm asking the most stupid questions, but I'm just rolling with it.

我们现在在巴黎，快到一天的尾声了。

We're in Paris and it's nearly the end of the day.

那为什么他们40%的收入来自推理呢？

Why is 40% of their revenue inference then?

那你为什么没有拿下更多这部分业务呢？

Why have you not taken so much more of that?

2024年初，我们只有640颗芯片。

At the beginning of 2024, we only had 640 chips.

到年底，我们已经有40,000颗了。

At the end, we had 40,000.

我们还没有达到那个规模。

We're not at that scale yet.

你必须提供高质量。

You have to provide quality.

你必须提供低成本。

You have to provide low cost.

你必须提供速度，但也必须提供容量。

You have to provide speed, but you also have to provide capacity.

因此，这就是不使用HBM最为关键的部分所在。

And so this is where that that most important part of not using HBM came in.

这意味着我们实际上没有规模限制。

It means that we effectively have no scale limits.

因此，GPU本身是使用与你的手机相同的工艺制造的。

So the GPU itself is actually manufactured using the same process that you use for your mobile phone.

所以你手机里的硅芯片，和GPU用的是同样的硅芯片。

So the same silicon that's in your mobile phone is the same silicon for the GPU.

事实上，他们先制造手机芯片，因为手机芯片更小，良率更高。

In fact, they build the mobile phones first, the mobile phone chips first, because they're smaller, so they yield better.

英伟达实际上是在苹果之后才拿到的。

NVIDIA actually gets it after Apple.

区别就在于内存。

The difference is that memory.

这就是唯一的区别，但这种内存制造起来非常困难，且产能受限。

That's the only difference, but that memory is the hard part to manufacture that's that's limited in scale.

因此，通过避免使用它，我们实际上在扩展规模上几乎没有限制。

So by us avoiding that, we effectively have almost no limit on how much we can scale up.

这对推理非常重要。

And that's important for inference.

英伟达的利润率是多少？

What is NVIDIA's margin?

70%到80%。

70 to 80%.

70%到80%。

70 to 80%.

所以他们可以砍掉70%到80%的成本，从而变得极具竞争力。

So they can take 70 to 80% off and be radically more Yeah.

相比你们，成本效益要高得多。

Comparatively cost effective compared to you.

你可以摧毁他们的利润率。

You could destroy their margin.

但同样地，你几乎可以说，我们是给英伟达带来过最好事情之一，因为他们可以制造出原本计划生产的每一款GPU，并以高利润率出售用于训练，而这些成本会分摊到整个部署中。

But in that same vein, you could almost say we're one of the best things that's ever happened to NVIDIA because they can make every single GPU that they were gonna make, and they can sell it for training, high margin, right, gets amortized across the deployment.

我们接手低利润率、高容量的推理业务，让他们不必玷污自己的利润率。

You know, we'll take the low margin, high volume inference business off their hands, and they won't have to sully their margin.

什么是低利润率？

What's low margin?

取决于具体交易。

Depending on the deal.

我们在后端确实能拿到一些，但前期大约是20%。

We do get some on the the backside, but upfront it's about 20%.

大约20%？

About 20%?

是的。

Yeah.

好的。

Okay.

所以他们的占80%，你们占20%，但你们的20%还要再看。

So theirs is 80, yours is 20, but then you're looking at a 20

但我们之后还能从它那里获得更多收益，所以我们承担了一些风险。

But then we get more later off of it, so we take some of the risk.

你什么意思？

What do you mean

你说之后还能获得更多？抱歉，没听清。

you get more later, sorry?

我们做的这些交易中，合作伙伴会承担费用，因为我们自己不会投入资本支出。

So the deals that we do, the partner will off because we don't deploy our we don't spend money for our own capex.

合作伙伴会为我们垫付部署所需的资金。

The partner will put up the money for us to deploy.

我们会以不错的内部收益率偿还，但收益会分成，大部分归合作伙伴。

We pay back with a decent IRR, but we split, and most of it goes to the partner.

一旦我们达到内部收益率，情况就会反过来。

And then once we hit the IRR, it flips the other way.

所以是别人为我们垫付资本支出。

So others are putting the CapEx up for us.

那么最后看起来是什么样子呢？

What does it look like at the end then?

嗯，最后的情况和其他商业模式不一样。

Well, at the end, it's not like other business models.

所以我们不仅仅在芯片上进行了创新。

So we we didn't just innovate on the chip.

我们还在商业模式上进行了创新。

We also innovated on the business model.

我们的盈利上限取决于我们能部署多少，而不是我们拥有多少资金，因为资金是由合作伙伴提供的。

And we're limited in how much money we can make based on how much we can deploy, not how much money we have because the partners are putting that money up.

所以当我考虑我们能做什么时，一切都关乎我们能扩张到多大规模。

So when I'm looking at what we can do, it's all about how much we can scale.

什么是

What are

你们部署的限制是什么？

the limits to your deployment?

仅仅是芯片的限制吗？

Is it purely chip constraints?

主要是。

Mostly.

所以你在问关于AI的误解。

So you're asking about misconceptions in AI.

我认为其中一个关于能源。

I think one of them is about power.

确实，市场上在芯片拥有者和电力拥有者之间存在不匹配，但部分原因是因为你需要一个中间的数据中心，而目前数据中心数量不足。

It is true that there is a mismatch in the market between people with chips and people with power, but that's partially because you need a data center in the middle and there aren't enough data centers.

建造这些并不是世界上最难的事情。

Those aren't the hardest thing in the world to build.

它们并不容易，但也不是最难的。

They're not easy, but they're not the hardest thing.

建立电力供应更困难。

It's harder to build up the power.

然而，由于这种不匹配，大型超大规模企业四处宣称：我需要一吉瓦的电力。

However, because of that mismatch, you have big hyperscalers going around and saying, I need a gigawatt of power.

他们会向多达60家潜在的数据中心建设方提出这个需求。

And they'll say this to to 60 different potential data center builders.

然后突然间，你听到了回声。

Then all of sudden, hear this echo.

嗯，我听说这里有一个吉瓦，那里也有一个吉瓦，另一个地方还有一个吉瓦。

Well, I heard that, you know, there's a there's a gigawatt here and a gigawatt here and a gigawatt here.

突然之间，需求变成了60吉瓦，这正是从最初那一吉瓦引发的回响。

And all of sudden, there's, like, 60 gigawatts of demand, and it's this echo from that first gigawatt.

事实上，我知道目前有大约20吉瓦的电力正被计划用于数据中心。

The thing is I am aware of about 20 gigawatts of power that people want to make available for data centers now.

目前，全球大约有15吉瓦的数据中心，是当前容量的两倍多。

Right now, there's about 15 gigawatts of data centers worldwide, so more than double the current capacity.

我担心的是，人们现在正在增加电力供应。

Concern that I have is that people are now building up more power.

在未来三到四年里，人们会发现：我建了这么多电力，却没人使用。

And what's going to happen in the next three to four years is people are going to be like, I built up all this power and no one's using it.

这简直就是完全的浪费，我们再也不会这么做了。

This was like a complete waste and we're never going do this again.

那么，接下来会发生什么？还记得芯片每十八到二十四个月翻一番的情况吗？

Then what's going to happen, remember that doubling of chips every eighteen to twenty four months?

三到四年的时间里，你把这15吉瓦翻两番，那就会变成多少？120吉瓦？

Well, three to four years, you double that 15 gigawatts twice, and now you're talking about, what, 120 gigawatts?

根本没那么多电力可用。

There isn't that much power available.

再翻一番，那就变成240吉瓦了。

And then another one after that, now you're at two forty.

因此，由于这种不匹配和当前存在的沟通不畅，我们目前会稍微过度建设。

And so what's gonna happen is we're gonna overbuild slightly right now just because of that mismatch and the miscommunication that's going on right now.

然后我们会放缓建设步伐，并逐步停止这种行为。

And then we're gonna dampen our building and we're gonna close down on that.

接着，我们才会真正需要这些电力。

And then we're gonna have the real need for the power.

这正是我目前最大的担忧，因为这种电力将在三到四年内成为难以突破的瓶颈。

That's my big concern right now because that power will become a hard bottleneck in three to four years.

当我们正迈向一个推理需求是训练20倍的世界时，为什么还会出现数据或数据中心供过于求的情况呢？

Why will we have that data over or data center oversupply when we are moving into a

因为推理的需求将是训练的20倍。

world of inference which will be 20 x larger than training?

数据中心的问题在于，每个人都认为数据中心是房地产。

So the problem with data centers is everyone thinks that data centers are real estate.

而且很多人确实从事房地产。

And a lot of people do real estate.

数据中心不是房地产。

Data centers are not real estate.

现在行业里一个常见的笑话是，有人会说：‘我给你提供100兆瓦的容量，三个月内搞定。’

The common joke in the industry now is someone says, I'm going to have, you know, 100 megawatts of capacity for you, and I'm going to have it in three months.

你愿意签吗？

You willing to sign?

然后你问一个问题：‘你的正常运行时间是多少？’

And then you ask a question like, what's your uptime?

他们却说：‘不知道啊，就看电网怎么样吧。’

And they're like, I don't know, whatever the the power grid is.

你会说：‘等等，什么？’

You're like, wait, what?

你的发电机在哪？

Where are your generators?

哦，还没下单呢。

Oh, haven't ordered those.

我现在就下单。

I'll order them now.

你知道现在发电机的交货周期是90个月吗？

You know that there's a ninety month lead time on generators right now?

真的吗？

Oh, really?

然后下一个‘90’。

And then the next '90.

09/00。

09/00.

接下来的问题是，你的水从哪里来？

And then the next question is, where are you getting the water from?

等等。

Wait.

数据中心需要水？

Data centers need water?

我以为那只是些芯片。

I thought it was a bunch of chips.

所以有很多人根本不知道自己在做什么，因为他们以为这是房地产。

So there's a bunch of people who have no idea what they're doing going into it because they think it's real estate.

这些人现在正在建造过多的数据中心，但他们其实并没有真正建造出来。

Those people are now building an oversupply of data centers, but they're not really building them.

所以他们建造的是虚假的数据中心，人们以为那是真的。

So they're they're fake data centers that people think are real.

这些数据中心会怎么样？

What happens to those data centers?

因为它们不会被利用，对吧？

Because they're not gonna be utilized, are they?

亚马逊不会为一个根本不存在的数据中心付费，嗯，亚马逊不会上这种当。

Amazon is not gonna pay for a data center that doesn't Well, Amazon doesn't fall for this.

亚马逊有非常优秀的人才。

Amazon has really good people.

无论买家是谁，都不会为一个

Whoever the the buyer is is not gonna pay for a

没有水源或没有电力的数据中心付款。

data center that's got no water or got no power.

是的。

Yeah.

所以，在你看来，这些项目大多数都永远不会建成，只是浪费了吗？

And so is it just wasted from your Most of these projects will never be developed.

我们能建得足够快吗？

Will we build them fast enough?

你提到过数据方面的供过于求。

You said about the data like the oversupply.

建造一个数据中心确实需要时间。

It take it does take time to build a data center.

是的。

Yeah.

所以这几乎是可接受的。

So it's almost okay.

如果你训练一个模型，你真的希望在六个月内摊销成本。

If you train a model, you really wanna amortize it for about six months.

如果你部署芯片，你真的希望在三到五年内摊销成本。

If you deploy chips, you really wanna amortize it for three to five years.

在三年这一端更多一些，其他人则更偏向五年。

One more on the three year side, others are more on the five year side.

如果你建造一个数据中心，你可能要考虑十年到十五年。

If you build a data center, you're probably talking ten to fifteen years.

在发电厂，你可能要考虑十五年、二十年。

In a power plant, you're talking like fifteen, twenty years.

我们这个行业面临的问题在于，融资方式与实际需求之间存在这种不匹配。

The the problem we have in the industry is not on this and there's this mismatch between the sort of financing and the needs here.

所以，有人想训练一个模型，他们只会做六个月。

So you have someone who wants to train a model, and they're gonna be doing this for six months.

他们不明白为什么人们希望芯片有三到五年的承诺。

They don't understand why people want three to five year commitments on the chips.

而部署芯片的人也不明白，为什么有人希望数据中心有十五到二十年的承诺。

And then the people deploying the chips don't understand why someone wants a fifteen to twenty year commitment on the data center.

对吧？

Right?

现在数据中心的期限已经是七年了。

It's at seven years now on the data centers.

然后建造数据中心的人就需要长期承诺，这是一份七十八年的承诺。

And then the people building the data centers then need a long It's a seventy eighth commitment.

是的。

Yeah.

他们要的就是这种东西。

That's the kind of thing they're asking for.

所以整个生态系统中都存在这种完全的不匹配。

So you've got this complete mismatch throughout the ecosystem.

但有趣的是，尽管他们都希望零风险，并且希望另一方拥有主权财富基金级别的信用评级和长期承诺，但回报周期越长，基础设施就越通用。

But the funny part about it is while they all want to take zero risk and have a committed sovereign wealth level sort of credit rating on the other side of it with long commits, the longer the payoff time, the more generic the infrastructure is.

一个模型有非常特定的用途，但像LP和GPU这样的加速器除了用于生成式AI或大语言模型外，还能用于其他用途。

A model has a pretty specific use, but accelerators like LPs and GPUs can be used for other things besides generative AI or LLMs.

数据中心除了用于加速器之外，还能用于其他用途。

The data center can be used for other things besides accelerators.

电力可以用于任何事情。

The power can be used for anything.

当他们在这里寻求最低风险时，这恰恰是风险最低的地方，因为如果我们不把它用于AI，我们就可以用它来为所有电动汽车供电。

While they're looking for the least risk over here, it's the place where there is the least risk because if we don't use it for for AI, we'll use it to power all of the electric cars.

这是不是一种

Is this a

由于只有 incumbents 能够匹配数据中心提供商所需的期限，所以他们能胜出的情况吗？

case where incumbents win because they're one of the only ones who are able to match the durations required by data center providers?

没错，这正是我们与沙特阿美以及沙特阿拉伯的新实体合作的原因，因为他们拥有长期资助这种项目的能力。

Well, and and this is why we partnered with Aramco and this new entity in Saudi Arabia because they have an enormous ability to fund this over the long term.

他们有着非常长远的视角。

They have a very long term perspective.

他们拥有极佳的信用评级。

They have an amazing credit rating.

当你说到他们

When you say they

有能力为它提供资金时，这里存在一个误解，人们以为这是15亿美元的融资轮。

have an ability to fund it, and this is where the misconception was, people think it's a funding round of a billion and a half.

这并不是10亿美元的融资轮。

It's not a funding round of a billion.

不，他们没有筹集15亿美元，那是收入。

No, did not raise 1,500,000,000.0, that's revenue.

这实际上大约是OpenAI收入的30%。

That's actually about 30% of the revenue of OpenAI.

你能给我详细讲讲这个交易的结构吗？

Can you just walk me through how that deal is structured?

是的

Yeah.

我们去年刚开始的时候，部署了19000片我们的芯片。

We started off last year, right, and we got to 19,000 of our chips deployed.

我们大约在51天内完成了这一点。

Did We that in about fifty one days.

问题是，今年我们能做些什么？

The question was, what can we do this year?

他们已经去各地收集了大量电力，这笔交易的结构是：他们会为我们提供资本支出，以便我们在这些数据中心部署我们的芯片，而我们则根据所获得的收入进行偿还。

So they've gone off, they've collected up a bunch of power in the country, and the deal is structured so that they will put up the capex for us to deploy our our chips in that data center or those data centers, and we pay back based on the money that we make.

这与债务略有不同，因为他们参与了收益的上行部分。

It's a little bit different than debt in that they participate in the upside.

其性质是相似的。

It's similar in nature.

这是收入，因为我们实际上提前实现了盈利。

It is revenue because we actually make profit upfront.

这如何改变了你能做的事情？

How does that change what you can do?

好吧，我们现在不再受资本限制了。

Well, we are not limited by capital anymore.

关于Grok，有一个误解。

There is one misconception around grok.

有一篇论文声称，我们不可能在定价最低的同时实现盈利。

There was a paper that was written that said that we couldn't be profitable while being the lowest price.

我们可以收取更高的费用。

We could charge more.

但实际上，我们目前的贡献利润率非常积极。

But actually, we have a very positive contribution margin right now.

据我们所知，我们是唯一真正通过运行这些开源模型赚钱的公司。

As far as we know, we're the only ones that are actually making money running these open source models.

因为对于开源模型来说，大家都在用风投资金竞争，试图像优步那样抢占市场份额。

Because with the open source models, everyone's sort of competing with VC dollars trying to take market share Uber style.

对吧？

Right?

与此同时，我们坐在这里想，我们可以一直这样做下去，因为我们正在赚钱。

And meanwhile, we're sitting here going, we could do this all day long because we're making money.

我们甚至能够偿还内部收益率，并让我们的合作伙伴赚钱。

We're able to even pay off an IRR and and and make our partners money.

模型的另一部分是。

There's another part of the model.

我们还在与一些专有模型提供商合作。

We're also working with some proprietary model providers.

我们实际上在周日的Leap活动上展示了第一个，那是我们与Play.ai合作的语音模型。

We actually showed off the first one at Leap on Sunday where we did a voice model with Play dot ai.

这个也是收入分成。

That one is also a rev share.

但关键是，他们能从这个模式中赚钱，而行业中的大多数其他人都因为模型的同质化而亏钱。

But the thing is, they get to make money off of that, whereas most others in the industry are losing money because of the commoditization of the models.

那么，随着你们垄断力量的减弱，价格会变得更便宜吗？还是随着垄断增强，价格反而会上涨？

So do you have cheaper pricing over time as you bluntly have less monopoly power, or do you have higher prices as your monopoly increases?

我们希望利润率保持稳定，但希望价格下降，因为这样就能触发杰文斯悖论，生活就会变得美好，因为我们将实现规模化。

We want the margin to stay about the same, but we want the prices to go down because then we get into Jevan's paradox, and and life gets great because we're gonna scale.

我们的重点是实现规模化。

Our focus is on getting to scale.

为了在人工智能时代保留人类代理，我们需要成为全球最重要的计算提供商之一。

To preserve human agents in the age of AI, we need to be one of the most important compute providers in the world.

我们的目标是到2027年，提供至少占全球一半的AI推理计算能力。

Our goal by the 2027 is to be providing at least half of the world's AI inference compute.

我们认为，如果我们没有受到所有限制，甚至可能超过两倍。

We think we could be further than two x, given that we don't have all the constraints.

但为了实现这一目标，我们必须大力扩张，并通过额外收费，让人们没有任何理由不使用我们的平台运行他们的模型，也不使用我们平台上的模型。

But in order to get there, we do need to be very aggressively building out, and we need to give people no excuse for not running their models on us and and using the models that are on us by charging extra.

我不断提醒团队，因为有时候你必须反复强调：我们的增长速度超越了指数级。

And and what I keep telling the team over and over again, because you have to remind them sometimes, is we're growing faster than exponential.

当你以超越指数的速度增长时，任何利润数额都变得无关紧要。

And when you are growing faster than exponential, there's no amount of profit that you can make that matters.

重要的是在市场中站稳脚跟并变得重要。

What matters is getting a toehold in the market and becoming relevant.

什么可能阻止这一点？

What could prevent that?

我们曾经担心有人会定价低于我们，但后来我们意识到这并不是问题，因为投入这个领域的资金太多了，人们更愿意通过使用我们的服务来减少亏损。

We used to be worried that someone would try and price below us, and then we realized that wasn't a concern because there's so much money going into this that people are going to want to lose less money by running on us.

这不再是问题了，但在我们意识到这一点之前，这曾是最大的担忧。

That isn't a concern, but that was the big one early on until we realized that.

当我们看到扎克伯格投资650亿美元建设数据中心时，这究竟意味着什么？

When we see Zuck investing $65,000,000,000 in data centers, what does that actually mean?

这意味着他正在内部化原本必须支付给之前提到的供应商的数据中心利润，而Facebook正在做全栈布局。

That means he's internalizing all of the margins that he would have had to spend on data centers with the providers that we mentioned earlier, and Facebook is doing full stack.

Meta每年投入650亿美元。

Meta's doing 65,000,000,000 a year.

我认为谷歌说70或75，萨提亚说微软在做80，然后你还有Stargate。

I think Google said 70 or 75, and Satya said Microsoft's doing 80, and then you've also got Stargate.

是的。

Yeah.

这些是惊人的巨额资金。

These are crazy sums of money.

这些全部都是用于数据中心建设吗？

And this is all for data center build out?

不是。

No.

它还包括了里面配套的东西。

It also includes the stuff that goes in.

包括芯片、系统，所有东西。

Includes the chips as well, the systems, everything.

我们从未见过这么多钱。

We've never seen money like this.

不是吗？

No?

不是。

No.

以前从未有过这样的情况，但以前也从未有过如此明确的情况，即最终会带来巨大的价值。

There's never been anything like this, but there's never been a case where it was so clear that there was gonna be value at the end.

如果你早知道搜索会如此成功的话。

If you knew how successful search was gonna be.

对吧？

Right?

记住，谷歌之所以尽可能长时间保持私有，是因为他们担心微软会发现搜索能赚多少钱，然后试图复制。

Remember, Google stayed private as long as they did because they were afraid that Microsoft would figure out how much money search was making and then would try and replicate.

他们一上市，Bing就出现了。

And the moment that they went public, bing.

他们对这一点的判断非常准确。

They they called that perfectly.

每个人都清楚人工智能能赚多少钱，所以大家都在争相进入这个领域。

Everyone knows how much money there is in AI, so everyone's going after it.

你认为这些价值是分散在众多参与者之间，还是集中在一两个玩家手中？

Do you think that value is distributed amongst many players or concentrated towards one or two?

我完全同意

I completely agree with

你在明确价值分配方面的观点，但这种价值是相对均匀地分布，还是高度集中？

you in terms of the clear value when assigned, but is it distributed to some levels evenly or concentrated?

这符合幂律分布。

It's a power law.

经济中的价值越大，就越有可能出现某个实体远远领先，从而彻底主导市场的风险。

The more value there is in the economy, the more risk there is of a single entity being so far on one end that they just dominate.

你在七巨头身上就能看到这种现象。

And you see this with the mag seven.

人们预测，经济规模越大，经济结果的波动就会越剧烈。

And it's predicted just the bigger the economy gets, the more you will have big swings in in the economic outcomes.

目前，超大规模云服务商的市值都差不多。

Right now, the hyperscalers are all sort of even in their market caps.

这很奇怪。

It's strange.

你本以为其中一家会表现得特别突出，远远领先。

You would expect one of them to just be killing it and taking it much further.

所以我搞不懂为什么它们如此接近。

And so I don't understand why they're so closely grouped.

那么当我们思考这种分布时，我们该如何改变它呢？

So when we think about that distribution, how do we think about changing that then?

比如，显然要成为

Like, obviously, with

Grok，你希望成为镁光灯下的七巨头之一。

the Grok, you want to be one of the MAG seven.

你希望成为世界上最重要的公司之一。

You want to be one of the most important companies in the world.

你如何看待这一点？

How do you see that?

你达成目标的方式和保持地位的方式是两件截然不同的事。

So the way that you get there and the way that you stay there are two very different things.

初创公司中存在着一种生命周期循环。

There's sort of a circle of life that happens in startups.

第一阶段是解决一个未被解决的问题。

The first circle the first stage is solve an unsolved problem.

这就是你走红的原因。

That's how you go viral.

这就是你取得成功的途径。

That's how you do well.

第二阶段是营销阶段，此时其他人开始模仿你的做法，因为他们自己想不出新点子。

The second stage is the marketing stage, which is now other people are trying to copy what you've done because they can't think of something themselves.

现在你必须在广告和营销等方面与他们展开竞争。

And now you have to fight it out in in advertising and marketing and whatnot.

你看到消费品公司常常卡在这一阶段，它们的关注点更多变成了产品在货架上的位置，而不是其他任何东西。

You see CPG companies often get stuck there, It and becomes more about where on the shelf they are than anything else.

然后是最后阶段：七大力量。

Then the final stage is the seven powers.

一旦你找到了其中一些优势，并真正开始提升它们，你就拥有了系统性的优势。

It's once you've found some of those and you've really started improving it and you have sort of systemic advantages.

然后，当有人解决了某个未被解决的客户问题时，整个生命周期就继续循环下去。

And then what happens is someone solves an unsolved customer problem, and the whole cycle of life continues.

对吧？

Right?

现在谷歌必须重新开始，因为大语言模型比搜索更优秀。

Now Google has to redo this because LLMs are better than search.

所以，你要成为七巨头之一，首先就要解决那个未被解决的问题。

So the way that you start off to become a mag seven is you solve that unsolved problem.

而保持这一地位的方式，首先是找到这七大力量中的一项或多项，但你还要随时准备应对被颠覆，并继续反击、解决客户问题。

The way that you stay there is first you find one of those seven powers or or multiple, but then you have to be ready for when you get disrupted to continue fighting back and and solving customer problems.

我们提到了这里投入的巨额资金。

We mentioned the different huge amounts of money that's being spent here.

这是一个良好的泡沫吗？它为未来十年到二十年打下基础，最终资本真正变得有生产力，尽管在账面上看起来并非如此？

Is this a good bubble that lays the foundations for an incredible next ten to twenty years, where, finally, the capital actually turns out to be productive but not seemingly so on paper?

还是说，实际上大量资金被浪费在了贬值的资产上？

Or is it where actually just a huge amount of money is incinerated on depreciating assets?

我可以保证，会有巨额资金被浪费。

I can guarantee you that a huge amount of money will be incinerated.

我也打赌，总体而言，赚到的钱会比投入的更多。

I also bet that in total, more money will be made than will be put in.

这就是问题所在。

This is the problem.

你必须从整体上或单个投资的角度来看待它。

You you have to look at it either in aggregate or individual bets.

当所有人都在市场中进行投资时，有些人会亏钱，因为并非每家公司都能成功。

When everyone is making investments in the market, some people are gonna lose money because not every company is gonna be successful.

你总是会看到，当有一些真正的技术进步或新事物出现时。

What you always see is when there is some real tech improvements or things coming.

那些早期的、人们大力投资的、极其成功的东西，然后所有人都想加入进来。

You've got the things that were early, that people are investing in heavily, that are super successful, and then everyone else wants to get in on it.

于是，从AI芯片和AI模型，变成了现在满大街的AIT恤，接下来你就会看到AI导热膏。

And, you know, it goes from you have AI chips and AI models to now you've got AI t shirts, and next thing you know, you've got AI thermal grease.

人们开始把AI应用到一切事物上。

People just start applying AI to everything.

转眼间，你就会看到AI公寓。

Next thing you know, you'll have an AI condo.

所以关键在于分辨什么是真实的，什么是虚假的。

So the the trick is discerning what is real and what isn't.

每当有真正的东西出现时，总会有这些令人讨厌的骗子涌入，这很不幸。

You're always gonna have all of these really obnoxious charlatans coming in whenever there's something real, and that's unfortunate.

但一旦人们开始理解这项技术，明白什么是真实的、什么是虚假的，这些骗子最终就会被淘汰。

But eventually they get cleared away once people start to understand the technology and what's real and what isn't.

工作就是开始进行教育。

The job is to start educating.

人们越有知识，就越不会投资于AI保暖膏。

And the more educated people are, the less they'll invest in AI thermal grease.

什么是最有可能导致巨额资金打水漂的最大单笔投资？

What is the largest individual bet that will lead to the largest incineration of cash?

我不会点名任何人，但我认为这会在每一个学科中都发生。

I'm not gonna call anyone out in particular, but I actually think it will happen across every single discipline.

你知道凯恩斯的选美比赛吗？

The are you aware of the Keynesian beauty contest?

不知道。

No.

约翰·梅纳德·凯恩斯，那位经济学家？

John Maynard Keynes, The Economist?

这能解释你关于风险投资需要知道的一切。

This will explain everything you need to know about VC.

所以我有点紧张，但你继续说。

So I'm nervous, but keep going.

想象一本满是模特的杂志，就是那种长得很好看的真人模特。

So take a magazine full of models, human models, like, you know, good looking models.

让一群风投人士坐在房间里。

Have a whole bunch of VCs in the room.

他们可以下注，猜谁是最美的模特。

They're allowed to make bets on who the the most beautiful model is.

最后，押注金额最多的人获胜。

In the end, whoever has the most money on them is the winner.

你押在某个模特身上的比例，决定了你分得所有资金的比例。

And based on the proportion that you put on that particular model's face, you get the share of all of the money.

如果你押的不是最终获得最多资金支持的那一位，那你就会输掉你的钱，归那些押对的人所有。

If you put money on one that isn't the most beautiful by dollars, then you lose your money to the the people who bet on that one.

这正是软银所下的赌注，他们相信自己能赢得这场凯恩斯式选美比赛。

That was sort of the the bet that SoftBank was making, which was they could win the Keynesian beauty contest.

我就要再多投点钱，然后赢。

I'm just gonna put more money in and I'm gonna win.

当你拥有真正的技术优势而非营销时，这就成问题了。

That is problematic when you have true technological advantages as opposed to marketing.

当你在解决客户问题时，这就像一台秤。

When you're solving customer problems, it's a weighing machine.

一旦客户问题被解决，你就进入了这种营销的流行度竞赛。

Once the customer problem has been solved, you then get into this sort of popularity contest of marketing.

这次出现了一件不同寻常的事，我认为这在风险投资史上从未发生过：你看到一些人筹集了数十亿美元，而他们的竞争对手也筹集了数十亿美元。

Now something unusual has happened this time around, which I don't think has ever happened in in VC before, which is you see people raising billions of dollars who have competitors who've raised billions of dollars.

通常情况下，凯恩斯式选美比赛会有明确的赢家。

It usually, there is a clear winner in the Keynesian beauty contest.

你不会遇到这种争斗，比如，我得再投点钱。

You don't have this, like, fight where, you know, it's it's sort of like, well, I gotta put a little more money in.

我得再投点，你知道的，投一百亿美元。

I gotta put a little more I gotta, you know, put 10,000,000,000 in.

不。

No.

我要投200亿美元。

I'm gonna put 20,000,000,000.

我要投5000亿美元，因为凯恩斯式选美比赛已经完全失控了。

I'm gonna put 500,000,000,000 in because the Keynesian beauty contest has gone completely amok.

这以前从未发生过，所以现在人们甚至不知道该如何应对，因为以前如果有人融了十亿美元，你会觉得，哦，他们是赢家。

This has never happened before, and so now people don't even understand how to react because it used to be if someone had raised a billion dollars, you're like, oh, they're the winner.

现在的情况是，有三四个竞争对手都融了十亿美元。

Now it's like there's three or four competitors who have a billion dollars.

那谁赢谁输呢？

So who wins and who loses?

比如，马西会烧掉有史以来最多的现金吗？

Like, is Massey gonna incinerate the largest amount of cash ever?

我认为

I think

凯恩斯选美竞赛在这里不再适用，因为现在有大量资金被分散出去，我认为真正有最好产品的人才会成为赢家，因为每个人都能获得资本。

the Keynesian beauty contest no longer applies here because there's so much money available being spread out, and I think you're gonna see that the people who have the best products are actually gonna be the winners because everyone can be capitalized.

但赢家也会因为这种情况而面临问题。

But there will be problems for the winners because of this.

问题在于，你原本打算雇佣的员工，却被别人提供了荒谬的高薪。

The problems are gonna be of the sort you had this employee that you were gonna hire, and someone offered them a ridiculous amount of money.

是的。

Yeah.

你现在经常能看到这种情况。

You see this all the time now.

他们本可以去支持赢家，但现在却去支持了一个本不该存在、或同样可能获胜的竞争对手，导致人才被分散了。

And they could have gone and contributed to the winner, but now they're contributing to a competitor that shouldn't even exist or is equally likely to win, and now you're splitting the talent.

当你面对如此高的薪资时，你还会怎么做？

What do you also do when you have such high salaries?

我们已经看到一些公司给初级到中级员工开出一百万、两百万的薪资，而且他们

We've seen 1,000,000, 2,000,000 for kind of junior to mid level in some of these companies, And they

他们实际上生活在很棒的地方，过着精彩的生活。

are living an amazing life actually in great places.

你以为他们在广东为深度求索或其他中国竞品公司工作时，过着这样的精彩生活吗？

You think they're living that amazing life in Guangdong when they're working for DeepSeek or any other Chinese alternative?

我认为他们实际上拿的钱少得多，每天拼命工作二十小时，连康普茶都喝不上，却只拿两百万年薪。

I think they're actually getting paid much less working their fucking ass off twenty hours a day and not getting kombucha and being paid 2,000,000 a year.

公平吗？

Fair?

不仅公平，我们有一项政策：从不提供最高薪资，因为我们希望人们选择我们，而不是选择薪水。

Not only fair, we have a policy that we never offer the highest because we want people to choose us, not choose the salary.

如果我们卷入了薪资竞价战，那下次再有人开出更高的薪水，他们就走了。

If we win in a bidding war, then that means the next time someone comes along with a higher salary, that's it.

他们只会去拿那份其他的工作。

They're just gonna go take that other job.

根本没有什么忠诚度。

There's no loyalty.

他们不相信这个使命。

They don't believe in the mission.

相反，我们关注的是：我们要打造这个产品。

Instead, we focus on, look, we're gonna build this.

这是你的机会。

This is your opportunity.

你将有机会与杰出的人共事。

You're gonna get to work with amazing people.

花点时间与团队相处。

Spend some time with the team.

这些是你想共事的人吗？

Are these the people you wanna be working with?

因为坦白说，你会赚到很多钱。

Because frankly, you're gonna make so much cash.

这并不重要。

It doesn't matter.

但要押注于股权和结果。

But bet on the equity, the outcome.

帮助我们让这件事变得有价值。

Help us make this thing valuable.

那些认同这一点的人更容易管理，因为他们以使命为导向。

And people who buy into that, they're so much easier to manage because they're mission oriented.

他们都想做同样的事。

They all wanna do the same thing.

他们不是因为想喝康普茶才来，也不会因为拿铁机坏了就抱怨。

They're not there because they want the kombucha, they're not going to complain because the cappuccino machine is broken.

他们会直接去隔壁买咖啡。

They'll just go and buy their coffee next door.

你会吗

Will you

NVIDIA 会进入模型空气领域吗？

and NVIDIA move into the model air?

每个人都说模型提供商正在成为应用提供商，或者基础设施提供商正在成为模型提供商。

Everyone talks about model providers becoming application providers, or infrastructure providers become model providers.

我们决定不自己训练模型。

We have decided that we're not going to train our own models.

我们只会为特定情况做一点微调，但我们不想参与竞争。

We'll do a little fine tuning for specific cases or whatnot, but we don't want to compete.

这一点非常重要，因为人们把他们的模型和权重都放在我们这里。

That's really important because people are putting their models with their weights on us.

他们不希望我们从中学习并把它们用于自己的利益。

And they don't want us to learn from and take that stuff for our own benefit.

当你与超大规模云服务商合作时，就会遇到这个问题，因为你知道他们也在做你正在做的一切。

This is the problem you have when you work with a hyperscaler because you know they're also doing everything that you are doing.

所以我们决定：模型提供商，你们负责做模型，我们不做这个。

So we've decided model providers, you make the model, we don't do that.

还有用户和查询的数据层面。

There's also the data side of the users and the queries.

所以我们能做但没有做的另一件事是记录查询，这样如果我们想训练的话就有数据了。

So the other thing that we could do that we do not do is log the queries and then we've got data if we wanted to train.

我们不进行训练。

We don't train.

我们没有理由保留这些数据。

We have no reason to hold the data.

我们只在DRAM中临时存储数据。

We only temporarily store things in the DRAM.

没有持久化存储。

There's no persistent storage.

如果断电了，所有数据都会消失。

If the power went out, everything's gone.

而且DRAM容量有限，所以我们无法长时间保存数据。

And DRAM is limited, so we can't hold things for a long time.

所以你知道我们不会拥有你的数据。

So you know that we don't have your data.