人工智能正助力构建其迫切需要的能源来源（布兰登·索博姆与共同财富聚变系统公司）

那么，你们如何在合理的时间内制造这些磁体呢？

Like, how do you make these in a reasonable amount of time?

因此，在Spark上安装第一个磁体，不仅仅是为了展示这项技术的可行性，更是为了证明我们已经消除了生产风险——既然已经证明了这项技术可行，你们能否建立一条像特斯拉那样的自动化生产线，配备机器人等设备，快速生产这些磁体的零部件，从而在几年内而非几十年内制造出18个磁体？

And so putting the first magnet in on Spark wasn't just it wasn't really as much of like a technology risk demonstration, but it was a demonstration that we had retired the production risk of, okay, now, now that you've shown that this technology works, can you build like an assembly line that looks like, you know, looks like a Tesla, you know, assembly line with robots and things like that, that you can actually crank out, you know, the subcomponents of these magnets fast enough so that you could actually build 18 of them in a couple of years as opposed to a couple of decades.

因为整个关键就在于速度。

Because the whole thing is speed.

对吧？

Right?

如果我们已经发现了这项技术，但发现要花很长时间才能建造出来，那就不妙了。

It it'd be great, you know, it it it or it would not be great if we had figured out this technology and then it was like, okay, it's gonna take a really long time to build it.

这完全不符合整个计划的预期。

That would that that would not work for this whole plan.

因此，证明这项技术可行的最后一步，就是展示你们也能实现规模化制造。

So that was sort of like the last step of showing that the technology works was showing that you can also manufacture it at scale.

是的。

Yeah.

这就是为什么那对我们来说是一个非常激动人心的时刻。

And that's why that was a really exciting moment for us.

第二点是我们还没有正式宣布，但第二个磁体其实已经顺利推进了，只是没有像第一个那样大张旗鼓，因为一旦你完成了第一个，大家就明白了。

Second one is actually we haven't, like, formally announced that yet, but the, know, the second one is is, is already, you know, been been carried over with not nearly as much fanfare, because the first you know, once you do the first one, it's like, okay.

现在它们将开始源源不断地进入组装区域。

Now they're gonna start, you know, just, like, rolling into the, into the assembly area.

很多人一听到核能，立刻就会想到历史上发生的各种灾难，比如切尔诺贝利、三哩岛等等，还有核废料储存问题。

So a lot of people hear nuclear and and immediately think of all of the various disasters that have happened over the years with Chernobyl, 3 Mile Island, etcetera, and the waste storage issues.

不过，据我理解，聚变从根本上是不同的。

Fusion is fundamentally different, though, as I understand.

你能给我们稍微解释一下吗？

Would you mind kind of explaining that to us a little bit?

是的。

Yeah.

对。

Yeah.

是的。

Yeah.

这是个非常好的问题。

That's a a really great question.

有趣的是，这让我们又回到了核磁共振成像的话题。

And it's actually funny because it ties us back to MRIs.

非常好。

Excellent.

所以，是的。

So, yeah.

因此，聚变从根本上说是一种核过程，因为它确实涉及原子核，但它与裂变非常不同，正如你所说，当人们现在提到核能时，他们实际上指的是裂变。

So, so fusion is a, fundamentally, it is a nuclear process in that it, it does, it deals with the nuclei of atoms, But it is very, very different from fission, which like you said, like when when people say nuclear now, they mean they actually mean fission.

是的。

Yeah.

但‘核能’这个词用得有点太宽泛了。

But nuclear is kind of used, it's used a little too broadly.

因此，当我们提到聚变时，即使它在技术上属于核过程，我们也不会使用‘核’这个词。

And so, we we don't use the word nuclear when we say fusion, even though it is technically a nuclear process.

我们在描述聚变时不用‘核’这个词，因为它与裂变的过程截然不同。

We we don't use the word nuclear when we're describing it because it is such a different process than fission.

所以，裂变是将非常大的原子分裂开来。

So, fission is breaking apart very large atoms.

比如你有一个铀原子，用中子撞击它，就能把铀原子打碎，分裂成几块，同时还会释放出几个中子。

So if you have like a uranium atom, and you smack it with a neutron, you can break that uranium atom apart, and it will it'll break apart into a couple of chunks, and it'll also release a couple neutrons of its of its own.

如果附近还有其他铀原子，这些中子就会引发它们发生裂变。

And that, if there's other uraniums that are close enough, it will then cause those to fission.

如果你把足够多的材料紧密地放在一起，就能形成链式反应。

And if you, if you have enough of it close together, you can have a chain reaction basically.

因此，在裂变中，或者说，如果你的目标是建造核电站而不是原子弹，关键就在于控制这个链式反应，防止它失控。

And so, the whole name of the game in fission, or at least for fission, if you're trying to make a power plant as opposed to like a bomb, the whole name of the game is you want to control that chain reaction so it doesn't it doesn't run amok.

所以，你需要控制棒，基本上是在平衡这个过程，让它刚好维持在临界点，因为你也不希望它逐渐熄灭。

And so, you have these control rods and you're basically balancing out a process where you get it just at the point, because you you don't want it to peter out as well.

所以，你总是让它刚好处于临界边缘，是的。

So, you're sort of like act you're you're always making it so that it is just at the edge Yeah.

保持反应持续进行。

Of being critical and it keeps going.

聚变则是一个完全不同的过程，因为在聚变中，你不是通过链式反应让反应自我维持，而是真正依赖等离子体的温度来驱动反应。

Fusion is is a much different process because in fusion, instead of having a chain reaction where you're actually your reaction is causing itself to go, the process that powers it is really the temperature of the plasma.

如果你无法约束住等离子体，温度会迅速下降，低到甚至无法在腔室中存在空气。

And if you don't confine that plasma, the temperature goes down very, very fast to the point where you can't even have air in the chamber.

因此，在这些系统中，无论是托卡马克还是其他任何聚变装置，你都会把它放在真空腔室里。

So, in one of these systems, whether it's a tokamak actually or any other fusion system, you have it in a vacuum chamber.

因为你不希望有任何其他物质会冷却这种高温等离子体。

Because you don't want any other stuff that's gonna be able to cool down this very hot plasma.

实际上，这还挺有趣的，因为这些等离子体的温度高达上亿摄氏度，对吧？

And it's actually, it's kind of funny because the temperatures of these plasmas are in the, you know, 100,000,000 degree range, right?

你可能会以为，上亿度就像熔岩一样，对吧？

Which you would think, like, you think 100,000,000 degrees, that's like lava, right?

它会碰到墙壁，烧穿墙壁，再烧穿下一个东西，再烧穿再下一个东西。

It's gonna like touch the wall and burn through the wall and burn through the next thing and burn through the next thing.

但等离子体比周围甚至空气都要稀薄得多。

But the plasma is so diffuse compared to the even the air around it.

聚变级别的等离子体密度大约只有空气的百万分之一。

Fusion grade plasma is about a factor of a million times less dense than air.

因此，如果等离子体真的接触到墙壁，尽管它的温度远高于墙壁，但它的密度却远低于墙壁，墙壁会瞬间冷却等离子体并使其熄灭。

And so what actually happens if the plasma were to touch the wall is even though it's so much hotter than the wall, it's so much less dense than the wall, that the wall will cool the plasma down basically instantly and kill it.

所以，磁场的作用其实并不是为了保护墙壁不被等离子体损坏——即使等离子体确实会导致墙壁表面剥蚀，比如几纳米的厚度。

And so, the magnetic fields are actually not to protect the wall from the plasma, even the plasma will cause the wall to ablate, you know, like the nanometers.

你必须担心这个问题，因为如果你不断把等离子体挤压到墙壁上，你不希望墙壁持续被剥蚀。

And you have to worry about that because there's stuff that, you you don't wanna continually ablate your wall if you're smooshing the plasma into it.

但事实上，磁场的存在是为了保护等离子体不被墙壁影响，而不是反过来保护墙壁。

But really, the magnetic fields are there to protect the plasma from the wall, not the other way around.

因为你不想让等离子体靠近墙壁任何一点。

Because you don't want that plasma getting anywhere close to the wall.

而且实际上，如果你不小心打开阀门，让空气进入正在运行的等离子体腔室，也会使等离子体熄灭并终止反应。

And actually, you were to, you know, if you were to accidentally open a valve and let some air into the plasma chamber while it's going, that would also extinguish and kill the plasma.

因此，聚变与裂变在本质上是完全不同的，因为聚变非常容易关闭。

And so, it's sort of fundamentally different from fission, fusion is, because it's so easy to turn off.

你可以把裂变想象成默认开启的状态。

You can sort of think of like fission is kind of like default on.

只要你有足够的材料聚集在一起，它就会自动启动，而你需要付出额外努力才能让它停止。

If you have enough of the stuff together, it's like default on and you have to like do work to get it to stop.

而聚变则恰恰相反。

Whereas fusion is the opposite.

所以聚变是默认关闭的，你必须做一系列特殊操作才能让它运行起来。

And so fusion is kind of default off and you have to do all these special things to make it work.

这就是为什么我们早在几十年前就建成了裂变电站，而聚变电站却迟迟未能实现——这其实是一把双刃剑。

And that's why, I mean, that's fundamentally why we've had, why we have fission power plants far before fusion power plants because it's it's, you know, it's a double edged sword.

它本质上具有被动安全性，但正因如此，它也极其难以控制，过去五十年来，我们一直在努力创造并维持这个地球上脆弱的‘小太阳’，让它持续足够长的时间来产生能量。

It's it's very, very inherently passively safe, but that's also why it's been finicky enough that it's taken all this work over the last fifty years to get this like fragile little star that we're trying to create on earth and keep it around long enough to make energy.

因为它总是想自己关闭。

It because it keeps wanting to turn itself off.

对。

Right.

哇。

Wow.

是的。

Yeah.

就像你无法访问这个一样。

It's like, you can't access this.

这是个作弊码。

This is a cheat code.

对。

Yeah.

你做不到这个。

You can't do this.

没错。

Exactly.

是的。

Yeah.

回到核磁共振成像，这挺有趣的。

And so going back to MRIs, it's funny.

我们实际上并不把聚变称为核聚变，因为我们不想把它和裂变混淆。

We, so we actually don't, you know, we don't refer to fusion as nuclear fusion because we don't want to confuse it with fission.

而“核”这个词已经和裂变紧密联系在一起了。

And the word nuclear has been so tied to fission.

核磁共振成像与之相关的地方在于，核磁共振成像最早被称为核磁共振（NMR），这和核磁共振成像其实是同一种东西，只是NMR代表的是核磁共振。

And the tie in to MRIs is that MRIs were first, and actually still in a research setting, you'll see these things called NMRs, which is exactly the same thing as an MRI, but it stands for nuclear magnetic resonance.

这是因为核磁共振成像的过程确实利用磁场使原子核产生共振，但这与裂变完全无关。

And it's because the process of an MRI that you are also affecting the nucleus with magnetic fields with a resonance, but it has also nothing to do with fission.

于是人们发明了核磁共振（NMR），我们惊叹道：哇，这是成像的绝佳方式，能获得非常高质量的扫描图像。

And so, people invented NMRs and we're like, wow, this is a great way to image things so we can get really high quality scans.

然后他们把这东西推向市场，结果没人愿意钻进一个核反应罐里。

And then they put it out in the market and like nobody wants to get in a nuclear tank.

所以他们迅速将其重新命名为MRI，即磁共振成像，这样人们就会觉得安全，不会觉得自己像是被送进了反应堆池中。

So, they quickly rebranded it to MRI, magnetic resonance imaging, so that people would feel safe in this thing and not feel like they were like launching themselves and, you know, inside of a reactor pool.

这很合理。

Makes a

说得通。

lot of sense.

我能理解人们为什么会因为本能反应而产生顾虑，因为这不是大多数人首先能完全理解的主题。

Can see why people would by just the knee jerk reaction because this isn't a subject most people one: have a

远非如此，我想说。

full grasp Far from it, I would say.

我有个关于数据中心的问题。

So I have a data center question.

我们不断听到，到2030年及以后，对数据中心的需求将大幅增长。

We keep hearing about that there's this big projected need that's going to continue to increase by 2030 and so on.

核聚变有可能及时到来并提供帮助吗？

Is there any chance fusion arrives in time to be of assistance there?

就像我们在人工智能领域听到的那样，还是这真的更像2035年及以后的事？

Like we've heard about from people in the AI space or is this really more of a 2035 and forward play?

我的意思是，我们正在尽最大努力，争取在2030年代初就把首批塔式反应堆接入电网。

I mean, are pushing as fast as possible to get the first tower plants on the grid, like early 2030s.

哇哦。

Wow.

这是一个非常激进的目标。

That's an aggressive goal.

但我觉得，你知道，每一天都很重要，这是其中一件事。

But I think, you know, every day, it's one of these things where like every day counts.

你面临着双重压力：一边是气候变化，另一边是能源需求。

You have dual pressures of, you know, you climate change on one side and you have demand on the other.

这两方面都促使我们必须加快步伐。

And both of those are pressures to go fast.

对吧？

Right?

这两方面，解决方案不是放慢速度。

Both of those things, you know, the solution is not to go slower.

而是要加快速度。

It's to go faster.

但你同时还得确保做对。

But you've to get it right at the same time.

对吧？

Right?

必须做对。

Gotta get it right.

是的。

Yeah.

是的。

Yeah.

我看过你的一场演讲，你一开始就说：‘我需要速度。’

I saw a talk that you did where you started really like, I've got a need a need for speed.

我喜欢这个家伙了。

I like this guy already.

我最喜欢的，是的，我最喜欢的名言之一。

My favorite, yeah, one of my favorite quotes.

天哪。

Oh my gosh.

那么，AI能帮助你完成这个过程吗？

Well, can AI help you with that process?

所有这些合作，你知道的，是否让你更有信心能按时完成目标？

Is all of this, like, these collaborations that you have, you know, does this make you more bullish that you're able to hit those timelines?

那么，在你看来，AI是如何影响这一点的？

Like, how how does AI impact that, in your opinion?

是的， definitely。

Yeah, definitely.

我的意思是，AI将让我们能够基本上做到，说到底，即使在Spark上也是如此。

I mean, think the AI will allow us to basically, I mean, at the end of the day, you, even on Spark.

所以Spark是我们目前正在建造并刚刚交付的第一个设备，我们将磁体送入托卡马克大厅，而ARC将是其后的发电厂。

So Spark is the first device that we're building right now that we just shipped, you know, a magnet into the Tokamak Hall for, and then ARC would be the power plant after that.

即使在Spark上，我们也能学到大量经验，这些经验将影响ARC的最终设计。

There's there's going to be a lot of learning that we're able to do even on Spark that will be able to affect the final design of ARC.

因此，我们越快地利用Spark进行学习——你知道，Spark，我们非常有信心Spark能达到q大于1，但在此之前，我们仍有许多学习空间，以优化ARC在此之后的运行。

And so, the faster we can utilize Spark as a learn you know, Spark, we're very confident that Spark will get q greater than one, but there's still a bunch of learning that we'll be able to do to optimize how ARC runs past that point.

对吧？

Right?

因此，AI技术有望帮助我们更快地在Spark上学习。

And so the the AI techniques will, hopefully allow us to learn much faster on Spark.

但你仍然需要在实际设备上进行实验，以获得真正的数据——你知道，自然究竟会告诉你什么？

So you still have to run the experiments on the actual device and get like, you know, truth of like what actually what is what is nature, you know, what is nature gonna tell you?

但我们希望利用这些技术来加速我们的探索，眼前有一整片广阔的空间等待我们去探索。

But we wanna use these techniques to basically speed our, you know, there's this whole canvas in front of us, this whole space to explore.

我们希望在绘制这片领域时，尽可能高效地探索它。

And we wanna be able to explore that space as efficiently as possible as we map it out.

这正是我们真正乐观认为这些人工智能技术能够帮助我们实现的目标。

And that's what we really think that's what we're really optimistic that these AI techniques will allow us to do.

你这是在实际行动啊，我的意思是，我本来想说别的，但你确实言出必行——你们和谷歌签了直接购电协议，这意味着你们必须真正交付成果，对吧？

You're putting your your well, I mean, I was gonna I was gonna say something else, but, I mean, you're you are literally putting your money where your mouth is because you have this deal with Google where you just signed a direct power purchase agreement with them, which means that you actually have to deliver, right?

是的，当然。

Yep, definitely.

所以

So

你必须让它实现。

you got to make it happen.

这有点像循环，对吧？

It's sort of circular, right?

我们正在使用人工智能工具，构建可能最终为数据中心供电的装置。

We're using AI tools build something that may eventually provide power to a data center.

是的。

Yes.

是的。

Yes.

我喜欢这个。

I like it.

在这方面非常有趣。

It's very interesting in that regard.

因为安全特性完全不同，我认为这在监管方面实际上是个好事，对吧？

Well, because the safety profile is so different, I understand it actually as a good thing in terms of the regulations, Right?

因为你的监管方式更类似于粒子加速器，而不是核反应堆。

Because you're regulated more like a particle accelerator than a nuclear reactor.

没错。

Exactly.

那就是

That's

太棒了。

awesome.

是的。

Yeah.

那真是太重大了。

That's that's huge.

是的。

Yeah.

因为裂变的大量成本源于必须为安全进行极其严格的监管。

Because a lot of the cost of fission is driven by the fact that you need to be very tightly regulated for safety.

对吧？

Right?

比如，你需要确保所有材料的等级远高于普通材料，这就给建造这些设施带来了巨大的成本负担。

Like, you need to make sure that, like, all of your materials are, way higher grade than standard materials that you would get, which leads to a lot of cost burden on building these things.

而正如你提到的，实际上在美国，我们被当作粒子加速器来监管，还有一些其他国家也正在跟进，因为对于这种本质上会自动关闭的装置来说，这种监管框架更相关、更合适。

And we're actually, like you mentioned, we're regulated as particle accelerators effectively in in The United States, and actually a couple other countries are following suit because that's just a much more relevant and appropriate regulatory regime for something like this, which default wants to shut itself off.

那么，如何将这一点与人工智能联系起来呢？人工智能如何帮助你建造聚变装置？

So how, and and to bring this around to AI a little bit here, how does AI help you build a fusion machine?

比如，Torax 正在帮助你解决哪些具体问题，而这些问题单靠人类自身是无法解决的？

Like, there specific problems that Torax, for example, is helping you solve that humans alone weren't on their own?

所以我们与 DeepMind 有两项合作，分别是 Torax 和 JAX，以及与 NVIDIA 的合作。

So, so we have these two collaborations with with DeepMind, which is with Torax, you know, Torax and JAX, then with NVIDIA.

它们属于不同类型的合作。

And they're both they're kind of different flavors.

我先说说与 DeepMind 的合作。

So, I'll start with the DeepMind one.

与 DeepMind 的合作主要是控制等离子体。

The DeepMind one is really controlling plasmas.

这些等离子体，你可以想象，是非常高温的物质。

And so, these plasmas, as you can imagine, they're very hot things.

对吧？

Right?

如果你放大到原子层面，看看原子级别上发生的事情，就会看到这些粒子在四处弹跳。

And if you think of if you zoom in to, like, the atomic level of, like, what's happening at the atomic level, you have these particles that are bouncing around.

从本质上讲，热量就是速度。

Heat is effectively just velocity.

从物理学的角度来看，粒子运动的快慢与温度是等价的。

Like, from a physics standpoint, like, there's an equivalence between, like, how fast the particle is bouncing around and temperature.

正如你所想象的，温度越高，粒子运动得就越快。

And as you can imagine, like higher temperature, the faster particles are moving around.

因此，等离子体的控制需要非常快的时间尺度，因为你面对的是一个极其不稳定的、不断波动的系统。

And so, the control of a plasma is on pretty fast time scales, because you have this very wiggly, wiggly thing.

对吧？

Right?

所以我们用来约束等离子体的一种方法是大幅增强磁场，这就像用一把大锤子把所有东西都挤压在一起，从而获得很好的约束效果。

And so, one of the ways that you can, that we've sort of clamped down on the plasma is you can really increase the magnetic field, and that's just sort of like a big hammer to like, sort of just like squeeze everything in, and you can get a lot of confinement that way.

但另一种获得良好约束的方法是使用快速控制系统，它可以实时监测等离子体状态；一旦发现等离子体即将失控、撞上器壁并自行熄灭，你就可以立即调整某个位置的磁场，从另一个方向把它推回去。

But another way that you can get good confinement is by having fast control systems that can look at the plasma, and if you see the plasma starting to look like it's gonna, you know, go out of confinement and hit the wall and turn itself off, you can say, be like, okay, adjust this magnetic field here so you push it in from this direction instead.

我有点简化了，但简单来说，这本质上是一个主动控制问题：当等离子体出现我们称之为‘破裂’的现象时——即等离子体移动到你不希望它出现的区域，或变成你不想看到的形状，最终导致自身熄灭——你可以主动调节塑造它的磁场，把它重新推回它该在的范围内。

I'm sort of oversimplifying, but like, you know, you can it's basically like an active control problem where if you have a plasma that wants to the term of art that we use is a disruption, where there's a plasma does something that, you know, moves it in an area that you don't want it to be or makes it, you know, turn into shapes that you don't want it to, which will eventually extinguish itself, you can actively control the magnetic fields that are shaping it to basically, you know, stuff it back into the box that it's supposed to be in.

因此，Torax 希望能帮助我们实现更好的等离子体控制系统，从而实现这种快速控制。

And so what Torax will hopefully allow us to do is have better control systems on plasmas to really do that fast control.

我想强调的是，我们并不依赖这一点；Spark 的一个高层次设计原则就是，我们不能把实现 Spark 的成功归功于快速控制。

And I wanna emphasize that's not something that we're banking on in order one of the sort of, you know, high level design philosophy of Spark is that we said we are not allowed to, like, take credit for fast control in order to make Spark work.

所以，无论有没有快速控制，Spark 都将实现 q 大于一。

So Spark is going to get q greater than one with or without fast control.

然而，如果在强磁场基础上再加上快速控制，那就是额外的增益。

However, with fast control on top of the high magnetic fields, that's that's all additive.

因此，如果能拥有这样一个快速控制系统，能够检测到‘等离子体即将破裂’，那将非常非常棒。

And so it would be really, really great to have a fast control system like this, and to be able to sort of detect, okay, like, it looks like the plasma's gonna disrupt.

然后立即调整等离子体的形状，使其在那一刻避免破裂。

Now, change change around the shape of the plasma so that it won't disrupt in the moment.

要实现这种足够快的响应速度，你不可能依赖人工操作。

And to do that on fast enough timescale, you know, you can't have a human.

这可不是以秒或分钟为单位的节奏，让人去手动调旋钮。

It's not on, like, second timescales or minute timescale where you can have a human adjusting a knob.

你真的需要一个快速控制系统，它能感知发生了什么，并且从等离子体中学习，因为这些现象都非常复杂——等离子体本质上是由大量粒子组成的，多到我都说不上来具体名称，但大概是10的20次方个粒子。

You really need fast control system that knows what's happening and is kind of learning from the plasma because, you know, a lot of these things are very complex phenomenon that, you know, plasma is sort of a collection of, you know, I don't even know the name for the number, but it's, you know, it's collection of, you know, 10 to the 20 particles.

我不清楚这相当于多少亿亿。

I don't know how many billion billion that is.

这大约是十万亿。

That's like 100,000,000,000.

所以有无数个粒子在电磁作用下相互影响。

So there's many, many particles that are all interacting with each other electromagnetically.

这是一个非常复杂的系统。

It's a very complex system.

因此，像AI强化学习系统这样的东西可以介入，说：我们并不完全清楚为什么这样调整旋钮，但我们注意到，当我们测量某些其他效应时，就会出现某种行为。

And so having like an AI, like reinforcement learning system that can go in and say, we don't know why exactly we're tweaking the knob like we are, but we notice that there's this behavior that happens when we measure certain other effects.

这对控制等离子体会非常有帮助。

That's something that would be really beneficial to control plasma.

是的

Yeah.

它也能实时运行吗？

Would it be able to work real time as well?

所以，一旦你有了模拟，就可以让它持续运行。

So, like once you have the simulation, like, you you could just have it going.

那么，你为每个聚变装置都需要一台超级计算机吗？

So you almost would be would you need a supercomputer with every fusion machine that you create then?

从某种意义上说，为了快速运行它？

In in in a sense to run this quickly?

实际上，我们的目标是不需要完整的超级计算机。

Well, the goal would be actually that you don't need a full supercomputer.

如果你有一个足够好的模型，它就能反应得足够快，你就无需再依赖超级计算机了。

Is that you actually if you have a model that's good enough, you could have it react fast enough that you don't need Yeah.

你知道，不需要那么庞大的超级计算机来帮你运行装置。

You know, this giant supercomputer, helping you run the machine.

而且，AI 正在承担主要的工作量。

And, like, the AI is really taking on the the heavy lifting here.

是的。

Yeah.

你基本上会先训练它，然后将其简化，以便能在任何方便的设备上运行。

You would train it essentially, and then, you know, you can distill it down so that it can run on, you know, whatever is convenient.

好的。

Okay.

但这实际上引出了另一个合作，就是与英伟达合作开发代理模型。

But that actually brings us to the other collaboration, which is with NVIDIA and doing surrogate models.

所以，回到超级计算机，人们会运行一些模拟，比如在特定条件下，等离子体的整体输出会是什么样的预测？

And so, going back to supercomputers, there are simulations that people run on, okay, like if you run a certain set of conditions, what is the prediction for the output of the plasma on a global sense?

在做这些模拟时，你会考虑到等离子体内部所有细节。

And in doing that, you're going to like all of the details of like what's going on, you know, inside of the plasma.

而这需要一台超级计算机。

And that requires a superconductor.

但人们已经开始行动了，我们麻省理工学院的一些同事正在引领这一方向。人们构建了所谓的代理模型，虽然仍然需要运行这些超级计算机模拟，但通过使用人工智能技术来确定最优的搜索路径，可以在空间遍历上将所需模拟次数减少一到两个数量级。

But people have started, and these are some of our colleagues back at MIT are leading the charge here, people have built these things called surrogate models where you still have to run these supercomputer simulations, but the traversing of the space of options, can greatly optimize by, like, order of magnitude or two orders of magnitude by using AI techniques to say, okay, what is, like, the best trajectory through this space to go through so that we can minimize the number of supercomputer simulations?

我们与英伟达合作的一个方向，就是构建这些代理模型，来实现所谓的——虽然有点陈词滥调——等离子体的数字孪生体。

And one of the things that we're working with NVIDIA on is making these surrogate models to make sort of this is, you know, buzzword, but like a digital twin of the plasma.

是的。

Yeah.

所以我们现在可以说，借助Spark，我们已经构建了这个系统。

So that we can say, we now you know, we've built this thing with Spark.

你可以把与DeepMind的合作看作是围绕等离子体实时控制系统的一部分。

You can sort of think of, you know, the DeepMind collaboration piece is sort of around the real time control system of the plasma.

但英伟达这一块的合作，我们还与西门子合作。

But then the NVIDIA piece, and we're also working with Siemens here.

西门子方面，我们大量使用西门子的产品，来记录机器的构建方式、数据采集方式等所有细节。

So Siemens, we use a lot of Siemens products that, you know, we're cataloging everything of like how the machine is built, how the data is collected.

通过与英伟达的合作，我们旨在创建这样一个数字孪生体，以便能够预测机器的整体性能。

And with the with the collaboration with NVIDIA, we're basically wanting to create this digital twin so that we can make predictions for, you know, the global performance of the machine.

如果我们想进行实验或调整参数，我们将能学得更快，因为我们相当于拥有一个数字实验室，可以开始预测：如果我们进行这个实验，我们预计会得到什么结果？

If we wanna run experiments or tweak things, we'll be able to learn a lot faster because we sort of have like a digital lab, if you will, of we can start to make more predictions of, okay, if we run this experiment, what do we think we're going to get?

并对我们将得到的结果有更准确的预测。

And have a much better prediction of what we're gonna get.

你们会怎么说呢？因为你们现在是从Spark运行后获取数据，会把这些数据反馈回模型吗？

What would you say, because okay, so you're taking the data from the Spark once it's up and running, will you be feeding that back into a model?

如果会的话，那么对你来说，哪种具体的AI突破会最有帮助？

And if you are, I guess what would be the thing that that like, what breakthrough in AI specifically would be most helpful to you?

因为现在我们听到很多关于科学被应用于AI，或者AI被应用于科学的讨论。

Like, because we're hearing a lot about science being applied to or AI being applied to science right now.

他们谈论的是新的物理问题，但目前有哪些AI能力是缺失的，而这些能力实际上能帮助你更快实现目标？

You know, they're talking about novel physics problems, but what kind of AI capability isn't available right now that would actually help you achieve this faster?

这是个好问题。

That's a good question.

目前还不存在？

Isn't available right now?

是的。

Yeah.

或者也许你其实已经拥有一切所需，只是需要安装更多磁铁。

Or or perhaps you haven't you have everything you need and it's just, you need to just install more magnets.

我不知道。

I don't know.

但你告诉我吧。

But you tell me.

我认为这其实只是在使用人们已经开发出来的工具。

I think it's really, I I think it's really just using the the tools that people have developed.

没有人真正应用过这些工具，因为缺乏可以应用这些工具的硬件。

Like, nobody has really applied these tools because there is a lack of hardware to apply the tools to.

我们现在正处在一个非常令人兴奋的领域，可以使用那些为其他目的开发的工具，现在我们可以这么说：好吧，现在把这些技术——比如英伟达和谷歌DeepMind开发的模型——真正应用到聚变等离子体物理这一具体问题上。

There's this is we're sort of in this, like, really exciting space where we can use these tools that have been developed for other things, but we can now say, okay, now take these, you know, take these techniques and models that, you know, that NVIDIA has and that Google DeepMind has developed and actually apply them towards the specific problem of, you know, add the physics of a fusion plasma.

然后结合我们即将拥有的实验条件，真正地在现实中运行这些模型并用数据来训练它们。

And then add the experiments that we're going to have available to us to actually run those things in real life and train the models with.

所以，我的意思是，我不确定自己是否能真正推测。

So, I mean, I don't know if I can really speculate.

我的意思是，正如你们所知，人们每天都在开发新的、更好的方法来运行AI模型。

Mean, people are, as you guys well know, you know, people are developing new and and better ways to to run, AI models every day.

但对我们来说，现在真正的问题只是把现有的技术应用到聚变问题上。

But I I think for us, it's really right now just a a question of taking what, you know, what exists and then putting that onto the fusion problem.

是的。

Yeah.

然后在真实数据上进行训练

It then training it That on a real

有道理。

makes sense.

是的。

Yeah.

我怀疑，一旦你有了这些数据，就能创造出一个全新的学习模型，形成一个数据不断反馈的良性循环。

I I I suspect that would be a really great opportunity to create a new, you know, learning model once you have that data and you can get that flywheel where you're just feeding the data back into it.

是的

Yeah.

完全正确

Totally.

是的

Yeah.

是的

Yeah.

这太令人兴奋了

That's so exciting.

我们有Spark什么时候能运行起来的时间表吗？

Do we have a timeline of when Spark is gonna be up and running?

如果你不知道也没关系

No worries if you don't.

尽快

As soon as possible.

我们的目标是今年年底前把机器组装好，然后在2027年启动它，实现首次等离子体运行，并尽快达到Q值大于一。

We're, you know, the the goal is to get the machine together by the end of this year, and then be turning it on, like getting first plasmas in in 2027, and then getting to q greater than one as fast as we can after that.

我得说，这是一类典型的情况，你知道，Spark里面大约有一万个独特的零部件。

I I do have to say, like, every it's one of these things where, you know, there's about, you know, some there's somewhere on the order of, 10,000 unique parts that go into Spark.

实际的零件数量要高得多，因为有很多东西。

They're much much higher actual part count because there's a lot of things.

比如我们有18个磁体，每个磁体都有不同的子组件。

You know, we have 18 magnets, and those all have sub components that are different.

但还有很多东西需要制造。

But there's a lot of things that have to be built.

所以现在，当人们问我经常被问到的问题时，比如：什么让你夜不能寐？

And so, now, I would say, you know, when people ask, you know, a question that I get asked often is like, what's the thing that keeps you up at night?

也就是说，最大的挑战是什么？

Like, what's the big challenge?

当我说现在让我睡不着的并不是科学和技术时，人们总是很惊讶。

And people are always surprised when I say at this point, it's not really the science and the technology that keeps me up at night.

就是这上万个零部件。

It is like, it is those 10,000 parts.

就是如何让一个组织能够把所有这些部件以正确的顺序和流程整合在一起，确保每一件事都能顺利配合。

It is just the execution challenge of having an organization that can bring all these things together and put them together in the right order and the right sequence, you know, and make sure that everything, you know, everything comes together properly.

这纯粹是一个巨大的执行难题。

It's just a huge execution problem.

我们有一个叫综合主进度表的东西，那简直是一个巨大的甘特图，列出了所有需要完成的任务，成千上万项任务。

Like, have a thing called the integrated master schedule, which is like enormous Gantt chart, right, of like all the tasks that have to be accomplished, these thousands and thousands of tasks.

我们有一支团队专门维护这个进度表，因为进度表一直在变化——比如某个供应商交货延迟，你就得立刻重新调整整个计划。

And we have like a team of people that maintain this schedule because the schedule is changing all the time because like if a supplier slips a delivery on something, you have to like refactor the schedule all of a sudden.

想象一下，你得把这种状况放大到成千上万项任务上。

And imagine, you know, you like multiply that by out to like thousands and thousands of these things that you're doing.

所以，这本质上就是一个项目管理的挑战：把所有部件整合起来，确保我们外发制造、运输的那些东西都能按时到位。

And so, really is just like a project management challenge of putting all the pieces together, making sure the things that we're going out and having fabricated and shipped and stuff like that.

你实际上还是在自己搭建整个供应链，对吧？

You're still basically building your own supply chain, right, effectively?

是的

Yeah.

哦，当然，我们的供应链团队至少有100人。

Oh yeah, we're definitely, our supply chain team is probably like at least 100 people.

我们公司目前大约有1200人，其中大约有100人属于供应链部门。

And we have about 1,200 people right now in the company, and there's about, like, roughly 100 of them are in supply chain.

而且，这还是一个非常技术性的供应链。

Every, you know, and it's a very technical supply chain too.

你可能会听说，供应链里有好几个人都拥有技术类博士学位，他们不只是简单地处理采购订单——虽然把采购订单做对也非常重要。

You hear like, there's actually several people on the supply chain who have like technical PhDs who are it's not just like, you know, like make POs for things, which that's really important too to get that right.

但还有些人会去供应商现场，指出：‘你们生产这个东西的速度比我们期望的要慢。’

But there's also like people who go out to the sites of vendors and are like, okay, Like, this thing is, you know, you're you're making this thing slower than we would like.

我们能怎么帮你们提高速度？

How can we help you make this faster?

是的

Yeah.

这种类型的事情。

Type of thing.

是的。

Yeah.

所以我认为重要的是要指出，融合并不是一个单一的突破，而更像是一个技术栈，比如磁体、机器整体。

So so I think it's important to call out that, like, fusion isn't so much a single breakthrough, but, like, a stack essentially, you know, magnets, machine Total.

工厂，正如你们团队之前提到的，还有电网经济。

Factory, as your team has mentioned before, then grid economics.

我想知道，这里的‘工厂’指的是什么？

And I was wondering, what does factory mean here?

你能为我们详细解释一下吗？

Could you walk us through that

稍微讲一下吗？

a little bit?

是的。

Yeah.

是的。

Yeah.

所以我们用来制造磁体的工厂是这样的：我们为所有部件都做了设计，但比如真空室，这曾是一个大工程，我们最近刚刚收到了真空室的一半部件运抵。

So the, the factory that we have making magnets is, so the, there's many pieces that were, you know, we make the designs for everything, but then, you know, like the vacuum vessel, for example, that was a big thing that we recently got one of the halves of the vacuum vessel shipped over.

这部分是由我们在欧洲的一家优秀合作伙伴制造的。

So that was fabricated out, by by a great partner that we have in Europe.

然后我们用安-124运输机把这一半运了过来。

And then we flew that half over on an Antonov.

另一半现在正在船上运输。

The other one is actually on a boat right now.

第二半我们不需要那么急，所以我们可以选择更便宜但更慢的海运方式，但代价是时间更长。

That one, the second one we didn't need as fast, so we were able to ship that one much cheaper going on a boat, but much slower, for that one.

但对于磁体这类部件，实际上我们根本找不到任何地方说：‘这是新型超导磁体的设计图，你们能帮我们制造出来吗？’

But, for things like the magnets, that's really like, there's nowhere that we could go to say, okay, here's some here's a design for a novel superconducting magnet.

你能为我们制造这个吗？

Can you please build this for us?

所以我们实际上建了一个工厂来生产这些磁体，我们设计磁体和技术时采用了模块化结构。

And so we actually built a factory to build the so we we we designed the magnets and the technology to be modular.

尽管有18个这样的大型环向场磁体——这些是呈D形的大磁体，大致构成托卡马克中的环形结构。

So even though there's 18 of these big these big toroidal field magnets, these are the big d shaped magnets that kinda make the ring in the tokamak.

尽管只有这18个磁体，但每个磁体都由16个几乎相同的子模块组成，我们称之为‘煎饼’。

Even though there's only only 18 of those, each of those 18 coils is made up of 16, like, nearly identical sub modules that we call pancakes.

因此，为全部18个磁体，你需要制造288个这样的煎饼。

And so, there's 288 pancakes you have to make for all of the 18 magnets.

另外，你还需要一些备用件。

Oh, Plus, you know, a few extra for spares.

我们索性就按300个来算吧。

Let's call it round up to like 300.

所以，你需要制造300个这样的煎饼。

So, you have to build 300 of these pancakes.

这个数量已经足够大了，足以让我们说：好吧，现在可以建一条生产线来制造它们了。

And that is a big enough number that you can say, okay, let's make an assembly line to make these now.

这可不是什么手工定制、你亲手绕线的那种活儿。

This isn't just like a craft bespoke, you know, you're like hand winding these things.

现在你可以开始算一笔账了：如果你得做300个这样的部件，假设每个饼花你一年时间，那你就完蛋了。

Now you can start to say, okay, if you have to make 300 of these, and you start doing the math, you're like, okay, if it takes takes me a year to make every pancake, I'm posed.

因为你根本不可能按时完成。

Because you're never gonna make them in time.

三百年。

Three hundred years.

是的。

Yeah.

对吧？

Right?

那可不行。

That would not be good.

所以我们基本上不断改进，现在工厂里的工具已经迭代到第五代或第六代了；一开始我们做的第一个饼，确实花了好几个月才完成。

So we we basically came up we're now, like, in the factory, we're sort of on, like, gen five or gen six of the factory tooling, where we've, you know, sort of continuous improvement, where the first pancake that we made, sure, that took, like, several months to make the first pancake.

我们现在已经为TF线圈生产了所有的饼状部件。

And we've actually made all the pancakes now for the TF coils.

但在我们产能高峰期，我们从第一个饼状部件耗时几个月，提升到了每天能生产两个。

But at our peak, at our peak rate, we were we went from, you know, the first pancake taking a couple of months to two pancakes per day coming out of that factory.

是的。

Yeah.

因为我们终于找到了如何实现自动化和并行化的方法。

Because Let's we were just able to figure out how to, like, how to get things automated, parallelized.

你有非常聪明的人，那些在生产线上工作的工程师和技术人员，他们一直在思考：我怎样才能让这个流程更快？

You have, like, really smart people, you know, the engineers and technicians who work on that line are constantly like, okay, how can I do how can I make this process faster?

我怎样才能消除这个系统中的浪费，就像汽车和飞机装配线上人们使用的那些方法一样，来思考：我们怎样才能更快地完成？

How can I eliminate waste from this system and just make it like, you know, using like, literally using the same techniques that people use in automobile and plane assembly lines to say, okay, how do we get these fast?

太棒了。

That's awesome.

AI能帮助你们吗？听起来，既然你们在自动化建设自己的工厂，AI应该能在这个过程中发挥作用。

And is AI able it sounds like AI would be able to help you in that process, obviously, if you're automating, like, building your own factory, basically.

是的。

Yeah.

我们当时还处于早期阶段，所以到目前为止只有少数几个领域，但这并不是因为缺乏应用领域。

We we were early days enough that there's only a couple areas so far, but that's not because there's a lack of areas.

只是因为我们做了大量的学习，但毫无疑问，我不知道我是否能具体说出我们应用了这些技术的哪些部分，或者这些是否属于受知识产权保护的内容。

It's just because there's there's so much learning that we did, but there's definitely, like, you know, I don't know if I'm I'm trying to think if I'm allowed to say exactly which pieces we applied this to, or if that's if that's IP protected.

但我们确实使用了图像识别技术来处理流程中的某些环节，比如，你可以让人坐在那里，仔细查看某种无损检测扫描的显微图像。

But we use, like, image recognition for some pieces of the process, where it was like, you know, you could either have a person like sit and look at micrographs of a, you know, of a scan, of a nondestructive scan of something.

或者你可以让一个经过训练的AI图像识别工具，来识别某种特定缺陷的特征。

Or you could say, okay, have a, you know, have an AI like image recognition tool that's been trained on what a particular type of defect looks like.

你训练它，让它自动扫描这些图像。

And you train it so you have this thing that's going over it.

然后有一个机器人会过去，真正地修复这些缺陷，几乎不需要人工干预。

And then there's a robot that goes over and actually is able to fix the defect, like without, you know, with very limited human intervention.

我现在可以说，仍然有很多方面有待改进。

And I would say right now, there's still a lot of things.

我们现在工厂里有很多机器人。

We actually have a lot of robots in our factory now.

工厂里大概有六七个工作站配备了机器人，这些机器人仍需人工协助，但我们正在研究如何实现自动化，让这些员工能去处理下一项任务，继续自动化下一个环节。

There's probably like six or seven stations in the factory that have robots that are still assisted by people, but that we're figuring out how to actually like automate that so that those people can go on to the next, you know, they can go on and automate like the next piece of it.

是的。

Yeah.

我的意思是，如果你让博士们去处理像采购订单这类工作，你就得充分利用他们的智力技能。

I mean, if you got PhDs working on like POs or whatever, you need to, you need to apply their brain skill.

你需要让他们去学习、去探索，然后你才能实现自动化，再让他们去解决下一个难题。

You need to have them learn something, you know, figure it out, and then you can automate it and then have them work on the next hard problem.

我们团队的情况真的很棒，这既是聚变行业的一个问题，也是它的优势——因为目前根本不存在成熟的聚变行业，所以我们必须自己从零搭建。

It's, it's, it's really cool that the team that we have, this is, you know, it's a, it's a bug and a feature of the fusion industry that there is no existing fusion industry, so we have to build it ourselves.

我们不可能直接去招聘有相关背景的人才。

Like, it's not like we could just go and like hire people who have a background in this.

所以这实际上非常棒。

And so it's actually really cool.

我工作中最令人欣慰的部分之一，就是能遇到并和这么多来自不同背景的优秀人才共事——比如那些制造高性能赛车的人，还有来自航天行业、制造过火箭等设备的人，正如你所想象的那样。

One of the most, I I say rewarding pieces of my job, like the physics and the technology is really cool, but just being able to meet and work with so many talented people from all different walks of life who've like people who've built like high performance race cars, like a lot of people, as you can imagine from, you know, from like the from the space industry who've like built rockets and things like that.

还有那些建造过石油钻井平台的人，他们所使用的科技简直令人难以置信。

And, you know, people who've built oil rigs and people who have built, which have crazy technology.

我以前根本没想到，石油钻井平台和深海钻探技术竟然如此先进。

I didn't realize like how crazy the technology is on like oil rigs and deep sea drilling.

那些技术简直离谱。

Like that is like ridiculous technology.

还有那些参与过其他可再生能源项目的人。

You know, people who, you know, who've worked on other renewable projects.

来自各个不同技术领域和背景的人才正汇聚在一起，将他们的专业知识和技能应用到如何实现核聚变上。核聚变仍是一个非常早期的行业，公司里的每个人都在共同塑造这个行业，这实在太令人兴奋了。

There's like people from all these different all these different areas of technology and backgrounds that are all sort of coming together to apply their expertise and skill sets to like, how do we make, you know, fusion is a very early industry and it's like, there everybody in the company is shaping that industry right now, which is super exciting.

对于正在观看的观众来说，如果想密切关注这里的发展，你们应该关注哪个关键里程碑，才能判断我们已经成功了？

So for people watching who wanna like, follow this place closely, what's, what's the milestone they should be watching out for that says we're there?

是的。

Yeah.

我们的首席执行官兼联合创始人鲍勃发布了一封非常棒的公开信，我不知道你们在节目笔记里会不会加超链接。

So, our CEO and my co founder Bob put out, there's a really great, I don't know if you guys put like hyperlinks your notes for the show.

我可以发给你，这封信主要是关于建立信任的。

I can send you he put out a really great open letter basically to the it was called building building trust.

让我看看，是《在聚变能源领域建立信任》。

Let me see if I can building trust in fusion energy.

这正是为了准确回答你的问题：如果你关注这个领域，而大多数人又没有接受过正规训练，那么我们应该关注哪些里程碑呢？

And this was the, just to exactly answer your question, like what are the milestones that we want to see, if you're following along this field that most people don't like have a formal training in?

所以，第一步是需要一个稳定的等离子体。

And so it's things like, you know, the first step is like you need a stable plasma.

然后，你需要把等离子体加热到足够高的温度。

And then you need to get the plasma hot enough.

接着，你需要达到我们之前提到的三重积的足够高数值。

And then you need to get a high enough value of this thing that we talked about earlier called the triple product.

再接下来的里程碑，就是让Q值大于1。

And then, the next milestone after that is getting q greater than one.

所以，这得益于高三乘积，也就是输出的能量比输入的更多。

So, that is enabled by having a high triple product, but that's like more power out or more energy out than than in.

然后，下一步是你必须产生足够的实际电力。

And then, the next step after that is you have to you have to make enough actual electricity.

所以，产生能量是一回事。

So, it's one thing to make.

当我们说 q 大于一时，那只是纯粹的能量。

When we say q greater than one, that's like, that's just like pure energy.

但显然，当你把能量，比如热能转化为电能时，并不是一对一的。

But obviously, you, you know, when you convert energy, like, as like heat to electricity, it's not a one to one.

你总是会，你知道的，

You just always, you know

是的。

Yeah.

熵总会找上你。

Entropy always gets you.

是的

Yep.

所以你总会损失一点点。

So you you always lose a little bit.

那么下一步就是，你能产生净电力吗？

So then the next step is is can you make net electricity?

而最后一步是，你能以具有成本竞争力的价格生产这种电力吗？

And then the the last step is can you make that electricity at a cost competitive rate?

你知道，归根结底，如果我们完成了整个过程，却说：太好了，

You know, at the end of the day, it does nobody any good if, you know, we get to the end of this whole exercise and we say, great.

这种强大的能源每千瓦时要一百万美元，

You know, it's a million dollars per kilowatt hour for this great source of power.

根本没人会买。

Nobody's ever gonna buy it.

对吧？

Right?

是的。

Yeah.

所以，没错。

And so Right.

是的。

Yeah.

从一开始，我们就必须考虑到，即使Spark不是一个商业产品，也永远不会成为商业产品，但我们设计Spark时所做的选择是：假设将这个技术放大到发电厂级别，它必须具有经济性。

From the beginning, we've had to, you know, you know, even Spark, even though Spark is not a commercial product and will never be a commercial product, Spark was designed we made design choices in Spark that was like, okay, if you project this forward to the power plant, has to be economic.

甚至在Spark的设计上，我们就设定了一个约束：你不能通过一个成本高出百倍的系统来欺骗自己，让自己看起来进步了。

And that was a constraint even on Spark of like, you can't you can't, like, cheat yourself forward on a system that is is going to cost a 100 times more than you could.

作为首个原型，当然可以接受，毕竟这是第一次建造这样的东西。

As a first of a kind, sure, you can get away with like, okay, it's the first time you build something.

任何首创的东西成本都会略高一些，但绝不能比最终产品的成本高出百倍。

There's always, you know, first of a kind things always cost a little bit more, but they can't be, like, a 100 times more than your final product is going to be.

十亿美元的产品对普通人来说是不可行的。

A billion dollars is not a feasible product for real humans.

是的。

Yeah.

对。

Yeah.

有道理。

Makes sense.

我有个小问题想接着你刚才说的问一下。

So I had one small follow-up to what you just said.

比如，你可能因为竞争原因不想透露，我完全理解，那我换一个问题。

Like, do and maybe you don't wanna say this for competitive reasons, which I totally understand, I'll ask a different question.

但你们在设计ARC——也就是核电站版本时，有没有一个目标电价在你们的路线图上？

But do you have a target like, when you're designing, like, ARC, right, the power plant version, do you have a target costs for the electricity that you have on your roadmap?

还是说，你们已经锁定了一个目标价格？

Or you're like, this is the price we're going for?

或者你们只是单纯想尽可能降低成本？

Or are you just like, we're gonna try and make it as cheap as possible?

我的意思是，我们努力让它尽可能便宜，但我认为我们相信，在不依赖补贴的情况下，我们的成本可以与天然气和可再生能源竞争。

Well, I mean, it's trying to make it as cheap as possible, but I think that we think that we can, you know, sort of be competitive in like target costs of like natural gas to renewables.

比如，如果我们从第一性原理出发，看看成本模型之类的东西，我们实际上认为这是可行的。

Like, that's, we actually think there is a, like, if you like, you know, first principles, you know, look at should cost models and things like that.

我们真的认为，即使不依赖补贴，也能做到这一点。

We, we actually think that you could, and like on a non subsidy basis, right?

你实际上可以在这类能源上实现成本竞争力。

Like, you could actually be cost competitive with those things.

而且这指的是某种类型的反应堆。

And that's for, you know, that's for an anth of a kind.

你必须沿着成本曲线逐步下降。

You have to climb down a cost curve.

这并不是第一座电厂。

That's not the very first plant.

对吧？

Right?

是的

Yeah.

对

Yeah.

但说到底，我认为这背后的首要原则是，聚变的物理特性本身就支持这一点，因为一旦建好了反应堆，燃料实际上几乎是免费的。

But I mean, at the end of the day, I guess the the first principles behind that is at the end of the day, fusion is really like, the physics of fusion lends itself to that because once you've built the plant, the fuel is effectively free.

它并不是真的免费。

It's not actually free.

比如，燃料确实会花一点钱。

Like, you do actually fuel does cost a little bit.

没错。

But Right.

燃料之所以成本低，是因为你消耗得极少，因为聚变的效率高出大约一百万倍。

The fuel, because you burn so little of it, because fusion is about a million times more efficient.

因为你是在打破原子核键，而不是像大多数化石燃料，甚至像太阳能这类传统能源那样打破电子键。

Because you are breaking nuclear bonds as opposed to electronic bonds, like most, you know, like fossil fuels and, you know, even conventional, you know, like solar and things like that.

你每单位燃料的效率大约高出一百万倍。

You're about, you're about a million times more efficient per volume of fuel that you use.

因此，燃料本身并不昂贵，比如氘，你可以从大约每6000个水分子中过滤出来。

And so, and the fuel itself is not, I mean, deuterium is something that is, can, you can basically filter out of every, I think it's every 6,000 water molecules.

实际上是D2O分子。

It's actually D2O molecules.

所以，你可以过滤出来，而且质量差异只有两倍，你知道，氘的重量是氢的两倍。

So, you can, you can filter And, it the mass difference is two, you know, a deuterium weighs twice as much as a hydrogen.

因此，建造系统来分离它相当容易。

So, that's pretty easy to build systems to, you know, to separate that out.

至于氚，虽然它在地球上不会自然存在，但你可以从丰富的锂中培育出来。

And then tritium, even though it doesn't occur naturally on earth, you can breed it from lithium, which is extremely abundant.

因为你知道，人们可能会问：你们会不会和电池争夺锂资源？

Because, you know, and people then then people are all lithium, are you gonna be competing with batteries?

但因为你用来培育燃料的量极少，实际上你使用的锂相当于一座ARC核电站整个生命周期内，仅需消耗约100块特斯拉电池所含的锂量。

But because you're using so little of it to breed your fuel, it actually, you know, it's you you use the equivalent, I think we did, like, the napkin math once and we said for, you know, the entire lifetime of an ARC power plant, it was something like, you would use the equivalent of, like, the lithium in a 100 Tesla batteries over the lifetime of the plant.

哇。

Wow.

你甚至可以回收特斯拉电池。

You actually can even recycle Tesla batteries.

你知道吗？

You know?

是的。

Yeah.

也许吧。

Maybe.

是的。

Yeah.

对吧？

Right?

是的。

Yeah.

那将是一个

That'd be a

很好的应用场景。

good use case.

所以，实际上，你知道，从基础物理学的第一性原理来看。

So so that's actually, you know, sort of from, like, a fundamental physics first principles.

比如，当你开始深入思考时，就会觉得，好吧。

Like, you you start going through and it's like, okay.

要制造这些东西，需要大量的钢材。

To, like, build these things, it's a lot of, like, steel.

就是钢材、混凝土。

It's like steel, concrete.

超导体是个奇怪的东西。

The superconductor is a weird thing.

但就连超导体本身，如果你真的去计算一下，就会发现。

But the actual superconductor itself even, like you you go and actually run the numbers on okay.

你拿一条这种超导带材。

You take one of these tapes that's a superconductor.

这种超导体被称为稀土钡铜氧化物。

The superconductor is called rare earth barium copper oxide.

所以，当人们听到‘稀土’这个词时，都会惊慌失措，心想：天啊，你们哪来这么多稀土？

So, of course, everybody freaks out when they hear rare earth because they're like, oh my god, are you gonna have enough rare earth?

但当你仔细算一下，看看这种超导带材的结构，就会发现大部分其实是基底材料。

But then when you do the math, you look at it, one of these superconducting tapes, and most of the tape is substrate.

它就像钢材一样。

It's like steel.

然后，是的。

Then Yeah.

而实际的超导带材中，只有大约1%是超导部分，而这一部分本身厚度还不到一毫米。

About 1% of the actual superconducting tape, which itself is is like less than a millimeter thick.

这层超导带材厚度的1%才是REBCO材料。

1% of the thickness of that is the REBCO material.

而REBCO里面其实还有很多其他成分，不只是稀土。

And the REBCO, like, is a lot of other stuff besides rare earths.

所以当你回过头来看，一个完整的弧形核电站，所有超导体加起来大约只需要100公斤的稀土材料。

So when you you you run it back and you say for an entire arc power plant, you need roughly a 100 kilograms of rare earth material in all the superconductor.

你实际上是把这些材料分散在大量磁体绕组所用的带材表面上，覆盖面积非常大。

You just have spread it out over a very large a large surface area effectively of all this tape that you're wrapping the magnets with.

所以你就像在撒灰尘一样，但这是非常有策略地进行的。

And so you're like sprinkling this like dust in very You're doing it very strategically.

但确实如此。

But yeah.

因此，你真正需要的原材料其实并没有多少稀奇古怪的东西。

And so, the actual like raw materials of stuff that you need is not like, is not a lot of crazy stuff.

这就是为什么我们认为，说你实际上能够建造这样的设施并非空想——你看看这些电站的规模，它们大致和同等规模的天然气电厂差不多。

Which is why we think that it's not, you know, fantasy to say that you could actually have And you look at the size of these plants, and they're roughly the size of like a comparable, like natural gas plant.

也就是说，如果你要建造一个形状和大小都与天然气电厂相当的设施，主要材料是钢、混凝土，没有太多稀有材料，当然，你只是把这些材料以与天然气电厂完全不同的方式组合起来。

It's like, okay, if you're building something that's the shape and the size of a natural gas plant and uses steel and mostly steel and concrete and not too many weird materials, sure, you're arranging those materials in a very different configuration than you would a natural gas plant.

但归根结底，材料就是材料。

But at the end of the day, like the materials are the materials.

这些都不是无法获得的东西，也不会造成巨大的供应链灾难，对吧。

It's not anything that's unattainable or is going to be a giant supply chain nightmare or yeah.

没错。

Exactly.

是的。

Yeah.

哇。

Wow.

那么在一个科幻世界里，这是否可能最终导致出现自动化工厂，这些工厂制造出聚变装置和发电厂，然后为自动化工厂本身供电，从而形成一个完全循环、充满丰裕的世界？

So in a sci fi world, would, would this potentially lead to a point where you have these automated factories that are making these, you know, fusion machines and power plants that then feed power to the power of the automated factories, and you basically have a fully cyclical, fully cyclical like world of abundance potentially?

我们走着瞧吧。

Well, we'll see.

是的。

Yeah.

我的意思是，这其实是一个正向反馈循环。

I mean, this is, you know, it's a reinforcing, you know, reinforcing loop.

我相信总会有聪明的人类，希望以某种方式、某种形式存在。

I think there's always gonna be clever humans in Hopefully in some hopefully in in in some in some way, shape or form.

是的。

Yeah.

祝好运。

Fingers crossed.

祝好运。

Fingers crossed.

但没错，正是这一点让我对核聚变感到非常兴奋，因为这不仅仅是一项清洁技术，你知道的，一开始我之所以对它感兴趣，是因为我觉得这是一项应对气候变化的绝佳技术。

But yeah, I do think that this is one of the things that got me really excited about fusion, is that it's not, you know, I I got excited about it at first because I said, okay, this is a clean technology, like, you know, for climate change, obviously, this would be a great thing.

但它带来的意义远不止于此。

But it's so much more than just that.

比如应对气候变化，它确实对缓解气候变化非常有帮助。

Like climate change, it's like really great for climate change.

而这仅仅是冰山一角，你不仅可以将所有东西替换成不排放碳的产品，还能开始产生更多的能源。

And then that's just like scratching the surface of not only can you replace everything with something that doesn't emit carbon, But you could also start producing more energy.

正如历史所显示的，当人类拥有更多可用能源时，他们会用这些能源做出非常聪明的事情。

And like history has shown that when humans have more energy available to them, they do really clever stuff with that energy.

我能马上想到一些例子，但我相信还有我们根本还没想到的用途。

Like I can think of a few things, you know, off the top of my head, but I'm sure there's things we haven't even thought of yet.

如果你把现在的全球能源使用量作为一个边界条件，然后提升到一百倍。

That if if you were to say, like, boundary condition now goes from current energy usage of the world to a 100 times that.

世界上有很多聪明人，他们会想办法解决各种问题。

There's very clever people out there that are gonna figure out how to solve problems.

是的。

Yeah.

利用这些额外的能源。

Using that extra energy.

这才是真正让我对核聚变感到兴奋的地方。

That that's really the thing that that gets me excited about fusion.

是的

Yeah.

要达到能够开始做这些创新事情的阶段，需要多大的规模呢？

How much scale would be required to to get to that point where you can start doing this novel stuff?

比如，你需要多少座发电厂？

Like, how many power plants would you need?

为了替代所有的电力生产，你可能需要大约……让我想想，达到20%到30%的水平。

So in order to replace sort of like all of the electricity generation, you'd need probably about, let's see, to get to like 20 to 30%.

我们之前算过这些数字，但你需要大约五到一万个发电厂，这其实……你知道，这些发电厂的装机容量大约是400兆瓦电，这差不多是普通发电厂的平均规模。

We did we ran these numbers a little while ago, but you'd need roughly like five to 10,000 power plants, which isn't I mean, that's like that's just like, you know, these these power plants are the are 400 megawatts electric, which is kind of like the average size of of a a power plant.

而世界上目前大约有两万到三万个这样的发电厂。

And that, like, that is the number of, you know, there's somewhere around, I think, like, 20 or 30,000 of these power plants on this curve in the world.

所以，如果你替换掉其中一半，那就是说，你需要五到一万个这样的发电厂，是的。

And so, if you replace, like, half of that, that's, you know, you're talking, like, five, ten thousand of these Yeah.

这些发电厂。

Of these power plants.

所以这是一个很大的数字，但与此同时，你知道，像飞机这样的东西是非常复杂的。

So this is a big number, but then at the same time, like, you know, there's like peop, you know, airplanes, which are very complicated things.

我们已经建造了数以万计的飞机。

Like we've built like tens of thousands of airplanes.

是的。

Yeah.

世界上，

The world,

人们本来就已经建造了所有这些发电厂。

people have built all those power plants to begin with.

对吧？

Right?

是的。

Yeah.

这些现有的资产，我们确实已经建造了它们。

Of those assets that exist today, like we have built those.

没错。

That's right.

是的。

Yeah.

你知道吗，困难的问题就是为了被解决的，对吧？

You know, hard problems were built to be solved, you know?

正是如此。

Exactly.

是的。

Yeah.

好吧，布兰登，非常感谢你抽出时间参加，我们真的很感激。

Well, Brandon, thank you so much for taking the time out to join I us really appreciate it, man.

人们在哪里可以了解更多关于你和共同财富正在做的事情？

Where can people go to learn more about what you and Commonwealth are doing?

是的。

Yeah.

所以我们在网站上发布了很多博客文章，并且会持续更新。

So we have on our website, we've been, we have a bunch of blog posts that we have and we update there.

我知道我们的市场传播团队会发布更新。

I know our MarComms team has updates.

他们会通过各种社交媒体渠道分享最新动态。

They share updates on all the various social channels.

如果你搜索‘Commonwealth Fusion Systems’加上你常用的社交媒体平台，我们应该都有账号。

If you search Commonwealth Fusion Systems, social media platform of choice, we should hopefully have an account there.

我知道他们在领英上也有账号。

I know they're on LinkedIn as well.

他们会发布相关消息。

They announce things.

所以我们尽量保持透明；如果你真的想深入了解，我们也会在科学文献中发表研究成果。

So, We we try to be pretty transparent to, also, if you if you really wanna geek out about stuff, we actually do like to publish in the scientific literature as well.

我们非常重视这一点。

We feel very strongly about that.

但作为这个话题，我们有责任向世界公开，因为这个领域很多人并不了解，我们必须在科学文献中发表，表明我们所做的一切并非空谈。

But as a topic, like, that we we sort of owe it to the world for this, like, topic that not a lot of people know a lot about to actually publish in the scientific literature and say, all this stuff that we're doing isn't just, like, full of hot air.

实际上，这些都是经过同行评审的。

There's actually peer reviewed.

还有其他与我们无关的科学家可以审查并确认：这看起来确实是可行的。

And there's other scientists who are not affiliated with us who could look at it and say, yes, that looks like it's feasible.

所以你也可以在科学文献中搜索。

So you can also search in the scientific literature.

我们什么都做。

We we do everything.

我们还特别坚持将所有成果以开放获取的方式发表。

We've we've also made it a point to publish all of our stuff open access.

所以，你可以去查阅。

So, know, so you can you can look.

我认为，第一篇ARC论文是在公司成立之前发表的，当时我们还没有资金支持开放获取出版。

I think actually the very first ARC paper that was before the company started and didn't have the money to publish it open access.

但档案中有一份副本。

But there's a copy of it out on archive.

太好了。

Excellent.

太棒了。

That's awesome.

很好。

Excellent.

各位观众，非常感谢大家今天参与我们的活动。

Well, to everyone watching, thank you so much for joining us today.

这非常有趣，希望你们对核聚变有了很多了解。

It's been a lot of fun, and I hope you learned a lot about fusion.

如果你们还没有，麻烦花一分钟时间点赞并订阅频道，然后前往 neuron.ai 注册我们的通讯。

If you haven't yet, please take just a minute to like and subscribe to the channel and pop on over to the neuron.ai and sign up for the newsletter today.

说到这里，今天就到这里了。

And on that note, that's it for today.