Cerebras CEO 谈数据中心的未来、Token成本与内存｜美国公司是否应向中国销售

Andrew, dude, it is so lovely to have you on the show. I have to say, I was quite emotional last week when I saw the IPO because you are one of the kindest, greatest people. I love getting to know you, and I so appreciate our relationship. To see that culminate last week with the IPO was really special.

Congratulations for last week, dude.

Thank you so much. Those were really kind words, and it was a really exciting day for the company, the team, and the people who'd believed in us and backed us for a decade. It was great. Thank you for saying those nice things.

Noted, too. I was thinking in terms of this conversation about how I wanted to structure it. I always get back to Eleanor Roosevelt's kind of statement when I have amazing people like you on the show: not very intelligent people discuss other people, mediocre people discuss current events, and intelligent people discuss the future and ideas.

1. Is There an AI Infrastructure Bubble?

I thought I'd grapple with my own ideas and wrestle with your incredible brain to help me understand where we're at and where we're going. I want to start with this: on the one hand, we look at the AI landscape today and it's like, "Oh my gosh, an AI infrastructure bubble." On the other hand, we look at Jensen Huang, who comes out and says we're going to be spending $3 to $4 trillion on AI infrastructure by 2030. How should I balance the idea that there's an AI infrastructure bubble with this appreciation of $3 to $4 trillion being spent by 2030?

2. Memory Shortages Will Last Years

I've been thinking a lot about this. I think when you look at other bubbles and bubbles in the past, I was in one in the late '90s when we built out an enormous amount of fiber optics. You sometimes have economists who think it's relevant to look at the 1880s and the building out of rail. I'm not sure that's relevant, but what I see is that there was a penchant to believe that if we built it, they would come.

The infrastructure buildout was way ahead of demand. That was true in railroads, and that was true in fiber-optic cabling. In a strange way, that is the exact opposite of where we are with AI. The infrastructure buildout is behind demand. We can't build data centers fast enough to keep up with demand.

We have a $25 billion backlog. NVIDIA has a backlog. AMD has a backlog. Others have backlogs. They have backlogs because we can't get data centers built fast enough. It's not that we're building on the come. We're not building ahead of demand; we're building behind demand. That is a very different observation from those who say there's a bubble.

I don't think they've really gotten their head around the fact that we're trying to keep up with demand, not the other way around. I don't think that's a characteristic of a bubble, when you're trying with your infrastructure to keep up with what people want today—not in the future, today—and their demands are growing over time.

Is it ultimately a good thing that we've all been metered in our ability to build out data centers, because it almost tempers the demand? If we were able to have it all today—and Gavin Baker said that the delays, permitting, and challenges incurred today actually help—because if you were able to have it all today, all demand would be met with all supply, and that would actually be a challenge.

Look, I think sometimes the world is like when I was in my 20s, the first time I went to Vegas and went to the buffet. You eat so much you feel sick for days. It's all in front of you, and you just gorge yourself. I think the market can sometimes be that.

I think Gavin is an extraordinarily thoughtful sort of guy about this. I think we are being metered. We also know that the reason you put meters on a freeway is because it makes the freeway traffic smoother and avoids hiccups. That's exactly what metering is designed to do. He used that analogy extremely thoughtfully.

One of the advantages OpenAI had—and I think one of Sam's brilliances was that he saw exponential growth and saw what that would mean in a year or 2 for the demand for compute—was that he wasn't afraid of it. He went out and took action.

Perhaps others couldn't believe it, or they were looking at the same demand, sort of steep exponential growth, and were like, "Well, we can't need that much. You can't need tons of compute. My mind hurts if you do that."

Whereas what OpenAI did was go out and say, "We're going to contract for it here and here. We're going to get power. We're going to get data centers. We're going to sign up for hardware and an ability to believe your data in an exponential growth environment a year or 2 or 3 out." That is a superpower.

Do you get rewarded for that insight if you can just buy it from Elon now on demand?

I don't think they can buy the same thing from Elon on demand. They bought down-rev gear.

I'm sorry. I've learned from doing this show for a long time. I can ask you the questions. They bought down-rev gear.

Yeah, they bought it. They got H100s. They didn't get the B200s. They didn't get the most current ones. They are a generation and a half, maybe 2 generations, behind.

This was not a great deal. It was a good deal for Elon; he had them sitting around. But they were forced to take action in a deal that I think was not the ideal deal they wanted. It was a deal that was available.

Going back to what we said about the delay in data centers and data centers being a constraint, I just hear everyone say, "Well, memory is the shortage too, Harry, and that's why we're seeing its cost increase 4 or 5 times in certain cases." Is memory—is that true? How should we think about memory being a shortage as well?

Well, look, what's happening here is there is such extraordinary growth in demand that it is putting pressure on all parts of the supply chain. After TSMC and fab space, memory is the #2 item that's needed.

There are only 3 companies that make the memory GPUs use. We don't use that memory. That HBM is made by Samsung, Micron, and Hynix. They couldn't keep up, and so the prices shot through the roof. Micron is producing numbers where they have 80% to 85% gross margins. They're getting software gross margins on making memory.

Yeah, I think it's extraordinary. Again, going back to my idea of what the future looks like, what should one expect from that? Does it ease over time? What happens to the cost?

I think the challenge here is that these are extremely lumpy items. You can't just add a little bit of manufacturing capacity at a fab. You have to build a fab for $40 billion, and it takes 5 years to build.

If you see demand explode, you cannot respond quickly. All you can do is fill your factory. Once your factory is full, you've got to build another factory. It's a step function in your ability to meet that demand, and the step is huge and takes years.

If demand stays high, we're going to continue to see memory shortages for at least the next several years.

Do you think we'll see a peaking of demand?

3. 2025: The Year AI Became Actually Useful

Not if AI continues to improve in usefulness. What's happened here—and this is something that I haven't heard others talk about—is that somewhere in 2025, the models got smart enough to be really useful.

Before that, these were sort of a novelty. AI was cool, and then nobody used it. Remember, we make AI with training, and we use it with inference. Once the AI we made in the first half of 2025 got smart, we began using it.

This explosion in demand that Jensen described, and that we very much agree with, is happening because people are using it every day. They're using it on more and more problems. They're using it on harder problems, and it is sweeping through different demographic groups.

It's not just 28-year-olds in Silicon Valley. It's my 85-year-old father. It's my 11-year-old niece. It is sweeping through demographic groups, and they're using it all the time.

If we continue to find ways to make the frontier models smarter and more useful, we'll keep using them, and the demand will continue on this sort of exponential curve.

You've compared power cycles before in this conversation. Sara Fry said that cloud providers can be similar, in some respects, to what we're seeing in terms of frontier models, and you said that last night.

To what extent do you think you see the commoditization there, and they essentially become utilities versus differentiated providers with meaningful moats?

4. Will Frontier Models Commoditize Like Cloud Did?

I think it has been NVIDIA's strategy to try to create competitors for the traditional hyperscalers. I think that has been a strategy of theirs. I think they have funded, backstopped, and overallocated to the neoclouds. I think they have created a dependence, which is probably not healthy.

But I think the truth is that what AWS and Azure offer is extremely useful for most enterprises. They offer credibility and legitimacy, security, and layers of different software for different parts of your organization. If you'd like to enter the AWS world, you can enter with Bedrock, and you can use tools like SageMaker. You have a collection of different ways to enter.

You can store your data there; you have your S3 instance. You can have an entire offering, and I think that is really valuable to a segment of the market. There might be other segments of the market that are like, "Give me cheap compute. I don't care about anything else."

In that case, your strength as a hyperscaler becomes your weakness. You have the security, you have the other layers of software, and you have some of the costs that are associated with that. If people don't want that, if you don't care about leather seats—right?—and there are leather seats in the truck, there's extra cost in the truck. When you buy the truck, you find somebody who's got a truck with Naugahyde seats.

Our business, just because it's wrapped up in technology, is no different from any other business. It's segmented. There's value, and that value comes at a cost. You have to make that value. The hyperscalers make the value through software, through security, through having rules about their data centers, physical security, and the various security checks they put in. Those are enormously valuable to most parts of the market, but not all.

You said about the cost there. When we look forward, how do the costs of your business change significantly over time? We spoke about the cost of memory going up 5x. If we look at the COGS in 5 years' time, how do you think they will look most significantly different?

The increase in the cost of memory has been very good for us because we don't suffer it. This has given us an opportunity. We use SRAM, and there's no shortage of SRAM. The cost of SRAM hasn't changed. No SRAM maker—because TSMC etches it into your chip while it's making the logic—means there are no extra margins to pay the HBM maker. We have been advantaged in this environment.

We have been advantaged by the fact that there are constraints on CoWoS at TSMC. We don't use CoWoS. We are advantaged by the fact that we're at 5 nanometers, and the 3-nanometer node is the most oversubscribed. Our supply chain is advantaged on these dimensions. NVIDIA and others are paying the price. The price of GPUs has gone through the roof.

And so, to my question on COGS, do we see a plateauing of COGS in terms of it can't get cheaper and this is the stable state? Do we see a meaningful reduction?

What I think happens over time, Harry, is that all of us improve our designs. The designs deliver more tokens per unit time. They deliver faster tokens. Now, we are 15x faster because of architectural reasons. We will continue to improve over time. NVIDIA will continue to improve over time. I believe the gap will widen between our performance and theirs.

But all of us—the whole industry, us, NVIDIA, AMD, Qualcomm, ARM, everybody—will have better chips in 3 or 4 years than we have today. They will produce more per unit of power, and they will produce more per dollar of cost. Over time, the history of our industry is a massive reduction in the cost per unit of compute.

I was chatting to a friend who's a phenomenal mind, and he said that Google will become the lowest-cost producer of tokens because they own the full stack, from TPUs to data centers, networking, and power procurement. Do you think that's right—that full-stack ownership will give them the highest-margin, lowest-cost ability?

5. Can Google Win by Owning the Full Stack From TPUs to Tokens?

There are pros and cons to that strategy. The pro is that you have everything from the ground—land—all the way up to tokens. The downside is that you can only sell your TPU to yourself. Historically, volume mattered a lot, and so your market is constrained by your own demand.

Whereas if you were able to sell to the whole market, you might have more demand and be able to drive down the cost. It's an open question. Google is challenging that argument. I think your friend's argument is reasonable, but there has historically been a challenge if you only have one customer—yourself—for your hardware. That has historically limited the size of the opportunity landscape for you.

Do you think they should sell to external customers?

I think you are already seeing them step outside of their own data centers for this exact reason. What it says in your friend's construct is, "Our ability to sell hardware is constrained by our ability to build data centers."

One can imagine a world where you don't want that constraint. You would like to be able to sell hardware to anybody's data center. I think these arguments are extremely complicated and rarely unfold in a simple form.

But it is true that when Google puts equipment in its own data center, or when Cerebras puts our equipment in our own data center, we have a significant advantage over any cloud. Neoclouds are buying hardware with gross margins of 70–80% for NVIDIA, and then they have to make their margin. That's not what Google is doing. That's not what we're doing.

Does that mean that you're automatically overvalued when you look at Abacus, CoreWeave, or any of the others?

I think CoreWeave has been an extraordinarily innovative company. I think they've solved a series of financial challenges with really innovative financial engineering. They were the first to use that very innovative approach. I give them enormous credit for that.

They have been extremely good at rapid deployment, which itself is a really important skill in this environment. I don't know about the others, but I think all of us have challenges as our business grows.

I think they have produced really interesting things through creativity. It's a different kind of creativity from what I have, but they've gotten paid for real innovation in financial thinking.

Speaking of real innovation, I saw the post about you running Kimi K2.6 6.7x faster than the next-fastest GPU cloud.

We posted it while one bozo at an analyst firm was on TV saying we couldn't do it. If ever there was an example of being empirically proven dead wrong, it was having these numbers posted while you were on TV saying they could never do it. It was perfect. I enjoyed that. I'm a collector of examples of people being dead wrong.

My wife has a list of when I'm dead wrong. I've sort of embraced this.

You should be a venture investor, my friend, with a portfolio of 30. You'll be dead wrong a lot. If you're lucky, 80% of your portfolio is where you were dead wrong. You should have that if you're doing it right.

Yeah, that's right. I agree with that. How important was that for you? And is there a stage where it doesn't matter being that incrementally more important? Like, 6.7x—this is so much more important. It's not 20% more important.

For hard problems, there is no upper bound on how much faster you want to be, nor on the value of speed. If in 3 minutes we can solve problems that take others 20 minutes, then think of all the extra problems we get solved.

If I'm your competitor and I'm solving your hard problems in 3 minutes while you're taking 20, imagine over a day or a week—you get smoked. You will be smoked in this example. That is the way this is going.

Speed is of the essence. It's true in coding, it's true in agentic flows, and it's true in every part of the AI landscape.

Let me just ask you this question: How big is the market for slow search?

Really? How big is it? It's zero. How big is the market for dial-up, for slow internet?

All right, how much would I have to pay you? Let's turn it around and say there's a negative market here. If I gave you $1,000 a month to have slow internet in your home, you wouldn't take it. That's how impossible it is to engage with an important technology slowly.

Why do we believe that inference will be any different? There'll be zero market for slow inference.

I'm kind of pushing back on this: when you power Codex and you're able to be so much faster, if you're a likely Claude coder, you're not like, "Ah, bugger." They are. You have to be. Are you able to sell to them as well?

Again, please tell me to fuck off.

No, no. Look, right now we are digesting one of the largest deals in the history of Silicon Valley. You're like, "For fuck's sake, Harry, give me a break. I've just signed a $20 billion deal. You want more?"

While we were on the road, some investors would ask. They'd say, "Oh, you're heavily concentrated. You have a big portion of your business with OpenAI." And we say, "I talked to you a year ago when I had a billion-dollar deal with G42."

And you said, “You’re heavily concentrated. You have a billion-dollar deal.” I said, “I come back to you in a year with a 20-plus-billion-dollar deal, and you tell me the same thing, but with a different customer as well.”

With a different customer.

With a different customer. And I tell everybody that the way you get good at—and the way you have succeeded with—many customers of size is, first, you win one. The way to catch big customers is, first, catch one, and learn: build the muscle, change your supply chain, learn how to work with a large customer. Then you’re in a position, when the next one comes, to have a chance to win and, what’s more, a chance to keep them happy once you’ve won.

Once you have that muscle, you’re in a position to go out and win the next one. It is a huge deal. It’s a huge deal.

What are the biggest challenges in fulfilling it? And, with the greatest of respect, do you go to sleep at night going, “Whoa, that is quite a lot”?

Look, I think what has happened—and Sam said this—he said the first time people used GPT—I think it was 4 or something—they said, “Oh, this is amazing.” And the next day, they’re like, “How come it’s not faster?” The rate at which you get accustomed to something and then want better is amazing in our industry.

It used to be the case that 20 megawatts was a lot. Then 100 megawatts was a lot. Then 1 gigawatt was a lot. Now we’re running around looking for multigigawatt facilities. At any other time, 750 megawatts would have been a mind-boggling amount, and now we’re like, “Yeah, we got that.”

I mean, the change in mentality over the last 1 or 2 years for everybody in our industry has been extraordinary. 5 years ago, if you had said we were engaged in a multigigawatt build-out—if you think about what Crusoe is doing or some of the other cool companies, what SoftBank Power is doing, what some of these groups are doing—you’d say, “That’d be delusional 5 years ago.” Right now, it’s like, “Oh, another one? Yeah, that makes sense.” I mean, we should try and get our UAE Stargate to 5 gigawatts. “Oh, yeah, yeah, no problem. That seems reasonable.” I mean, that’s what’s happened.

It’s this extraordinary change in thinking. If we’re all nonchalant about multigigawatt build-outs today, what are we in 5 years’ time? It’s difficult to imagine. And, by the way, that is exactly where I think Sam is the best in the world—maybe Elon—where everybody else’s brain shuts down. When you’re trying to think about 100 gigawatts or 500 gigawatts, those guys have this ability not to be constrained by the way the world has always been. And that is such an extraordinary power.

When you have such scale as the multigigawatt—you mentioned the 500 gigawatts—does energy not just become the core crux and bottleneck that enables winners and losers?

I certainly think that people like Sam and Elon and others have said that’s what they believe: that, at the end, we’re in the business of turning electricity into intelligence. Therefore, the limiting factor is electricity. I don’t know if I agree with that, but that is certainly a very reasonable view from where we are.

Why? What’s the bad case against that? What’s the alternative argument? It doesn’t have to be yours, but—

No, no. That you bump into something else. What happens, in fact, is our models can’t keep getting smarter. You hit something that has an assumption built in: that the models keep getting smarter, and you keep feeding them more energy. At the end of the day, the models are either smart enough or keep getting smarter, so that it makes sense to keep feeding them energy. That might be true. I don’t know.

Do you think we’ll be able to build out data centers in the way that we need to and build our capacity in the way that we need to, when we see that 40 out of 100 data centers are now not being built out even post-approval because of local municipalities’ permitting disruption?

AI is not popular here. I think it’s hilarious. People say, “Oh, my God, the data centers are late. Oh, my God, there are delays.” Have you built a kitchen? Your contractor was late.

Pick a little tiny project in your home. Was it built on time and on budget? No. Now imagine building something the size of 50 football fields and requiring interaction with local municipalities, power companies, and regulated industries. These things aren’t going to be delivered on time, historically. People’s minds explode, and they never think about their own experience in their own homes.

Did your contractor show up every day? No. Does he do exactly what he says he’s going to do? Rarely. Do the materials—the tiles or whatever you selected for your home—sometimes get delayed? Yes. All that same thing happens when you build a data center.

The generators are sometimes late. Sometimes they fall off a truck, literally. They fall off a truck, and damage is done to them. Are the transformers late? Sometimes. I mean, everybody suddenly throws their arms up and says, “Oh, everything’s late.” Or they have to deal with localities.

Anybody who’s built anything big knows this is par for the course. This is what building is. I totally get that.

So, you’re not concerned, then, about—bluntly—local neighborhoods seeing data centers as a symbol of—

Yeah, I think our industry did a shitty job of engaging the community properly. I think Brad Smith put out a post a while ago that should have been the way we all worked from the get-go. It was: these can be clean. They can make jobs. They can be good for communities. We can do this thoughtfully.

These create thousands of local jobs. Thousands. And thousands of local jobs means restaurants and lunches and hotels. The way they were done—I don’t know if “sneaky” is the right word—was sort of shielded and wasn’t open. And they weren’t good neighbors.

Now, there is no reason why we can’t be good neighbors. There’s no reason why we can’t add these to communities and have the community benefit from it. We have to do some thinking. We have all the heavy equipment out there. Build a football field for the local school. Build a school. Add a church or a synagogue to the community.

We can be good neighbors at very, very low cost. We can pay our own way. Don’t try and use loopholes in the way companies have historically amortized the cost of new power lines over 30 years and push that on the community. That’s BS.

We ought to pay our own way. We ought to look after our neighbors. And when we do that, I think the neighborhoods that embrace this will benefit enormously. But we didn’t do a great job. I mean, we didn’t do a great job as an industry at all.

It feels like the cartels in Colombia: they built the churches and they built the schools, and they had such good businesses and such high margins that they could kind of get away with it because of that. Hey, great.

I don’t think that’s right. I think these localities have a resource that isn’t being used. They have power. Much of this land is cheap because nobody wants it. We’re not going to parts of metro New York. Everybody wants that chunk of land. You’re going to places where the price of land is depressed or near power resources.

My position is that we ought to be good neighbors. We ought to be transparent. We ought to pay our own way. Now, this seems not to be very controversial in my mind. I think the best and most concise description is what Microsoft has put forward, and we should have been doing that from the get-go.

Everybody—or most communities—is comfortable when their neighbors pay all their own way. It’s only when groups try to shift costs or not pay for the full resources they’re using. Our data centers don’t need to use a ton of water. They can recycle it. You can have a closed loop.

We can pay our own way. We can upgrade substations. We can upgrade grids and pay for it in its entirety. We shouldn’t be pawning that off on local communities.

6. Data Centers & Local Communities

Do you worry about AI as a brand? You see massive layoffs of a huge amount of people, and it’s challenging to see 4 a.m. emails from Zuck and jobs being lost.

7. AI Layoffs

I do worry about it. Those are people, and they have families. I think there are 2 views, Harry. I think, to date, most of the layoffs were AI-washed. They were because we did boneheaded hiring during COVID. It is actually because a great deal of productivity gains have occurred over the years that we’re just now harvesting.

The ability to gather information from across the organization, to synthesize it and put it in one place, is now changing what it means to be middle management. The role of information gatherers and presenters is being eliminated. We have the ability to automate roles, and none of this is AI yet. That is really 90%, 95% of what, in my view, the terminations have been about.

Now, AI is starting, just now, to have meaningful enterprise impact. But if you are an engineering organization that can’t see how to take advantage of vastly more productive engineers, I don’t think you’re long for this world. I mean, the list of things I want our engineers to do is 50 times as much as we have engineers. As we get more productive, we do more things.

We're going to hire more engineers. We're not going to hire fewer engineers. Can I ask you? We saw Benioff say that he spent $300 million a year on Anthropic, which equates to about 3.8% of developer salaries on Anthropic. To justify the valuations we're seeing for these companies, it needs to be 20%. Do you have any concern about that movement from 3.8% to 20%?

No. I think if you look at what we pay hardware engineers and at the tools they use—the EDA tools—I bet you're much closer to 15% or 20% than 2% or 3%. What's happened is, historically, software engineers used very low-cost tools, while hardware engineers used extremely expensive EDA tools.

That's interesting, isn't it? I think the cost of bugs in hardware is so high that we became accustomed to using many expensive tools. In software, we threw people at the problem rather than tools. As AI becomes more productive, I certainly don't see a problem with software engineers using $50,000 or $100,000 a year each in tokens. There are 47 million software engineers in the world. I mean, that's $5 trillion just in software engineering token use.

Well, we've mentioned hardware engineers and software engineers. What role doesn't exist today that you think will be incredibly commonplace in 3 to 5 years?

I've been a part of several technical transformations over the past 25 or 30 years that produced jobs in companies that didn't exist. Prior to the mid-90s, the role of CIO didn't exist. The CIO arose as a role with Cisco and its rise to dominance. Prior to the mid-90s, the amount of enterprise networking was de minimis.

There was a role that was often a VP of telco infrastructure. That job is gone. We don't have a phone system. In fact, we don't have phones on people's desks. They call me on my cell phone. That job disappeared completely.

Later, in the 2000s, with the rise of Palo Alto Networks and these other security companies, the role of CSO emerged. It never existed prior to that. What you're going to see is the rise of roles that reflect the governance of AI in companies. Some companies have chief AI officers. I don't know if that's what it is, but as these technologies become important in companies' lives, new jobs emerge—jobs that never existed before. New organizations exist where there were none, and previous ones disappear.

I think the role of HR changes fundamentally. The part of HR that just answered questions and provided information about benefits disappears. AI can answer all your questions. It can provide better answers, faster answers, more thoughtful answers. The management of people becomes something different. I think there are all sorts of other parts of organizations that change fundamentally because AI can answer the questions that they used to answer.

Do you agree that the biggest inhibitors of enterprise adoption of AI are data structure and data cleanliness?

8. The Real Blocker to Enterprise AI Adoption

No. The biggest are lawyers. No, really. I think the security apparatus and the lawyers who, when they don't understand a technology, say, “No, we can't do it.” They're in the saying-no business, and entrepreneurs are in the getting-it-done business.

The reason for that—and there's a reason for this—is that your security apparatus and your lawyers, I mean, everybody, are in jobs where everyone just blames them. No credit, failure, blame. No credit, no credit, never failure, blame. I mean, that's their life, and it's brutal.

No, it's brutal. You're selling it so well.

Being a CISO is brutally hard. We had a year where nothing happened. Well done. That is their dream. I mean, every day their phone doesn't ring, and they're like, “Made it through another day.”

But when confronted with new technology, because their payoff structure is such that they're in the business of trying to avoid risk, they are a drag on adoption of new things. You see this across the board. Lawyers don't know how to contract for it. There's no precedent. That's a business of backward-looking precedent.

You want to make a lawyer uncomfortable? I know your girlfriend's a lawyer. Ask her to work in an area with no precedent. They don't know what to do. Their whole training is about what everybody else has done before, how do we synthesize this, and how do we work within those rules. So I think the wide-scale adoption and use of AI in organizations is today limited by security and legal.

I think once they agree, “We need to do this. Here are the rules we will use,” there's a huge amount of productivity to be gained. Then you're immediately constrained by the way you chose to husband and marshal data—the way you chose to organize data over the years. Organizations like Mayo Clinic that have been on a 30-year quest to organize data are at a huge advantage. Same with companies like GlaxoSmithKline.

Yeah.

And other companies that haven't perhaps been as disciplined or thoughtful about organizing their data are at a disadvantage.

On the security and provisioning side, do you think we'll see industries tip like legal has done, where the biggest firms in the world are now going, “Oh, fuck. We need AI. Our clients are saying we need AI. Harvey or Allegra.” I'm not going to get into which one, but there are 2 options. Boom.

Do you think all industries will follow the tipping, or do you think most will follow the slow agreement that it's the new normal?

I think what's happening is that the leaders are tipping. I think even Jensen told the story that he was battling with his own internal lawyers around the use of, I think it was, Cursor. Finally, he just decreed, “We're going to do it.” I think I got that right, but somebody will correct me for sure if I got it wrong.

I think at some point leaders weigh the productivity gains against the unseen boogeyman of risk. The problem with unseen boogeymen is that sometimes they're actually real.

Yeah, not often, but sometimes.

And that's the problem. What does he call him in John Wick? Baba Yaga?

John Wick is the guy you send to kill Baba Yaga.

Uh-huh. I think, for me, yeah, I'm not that young anymore, but I'm definitely capable of exuberance.

I know you're in your 60s, but you look good.

Yeah, and it's the facial moisturizing routine.

We mentioned security and legal and everything in between. They get even more freaking nervous when it's open source. They shit the bed. How do you think about that? I see more and more companies, especially in the Valley, really pushing the boundaries with frontier models and then trying to get as close as possible with open source, given the cost advantages. Is that the future, and what does that mean?

Look, I think we as an ecosystem have made real progress in the legal gunk around open source, but the result has been a complexity that hurts your head. If you ever want to dive down a rat hole that has no bottom, begin a discussion with lawyers about open-source software. There is no end to the depth and the boredom which you will suffer as you head down this hole.

This is made doubly worse by some of the best open-source models being made by Chinese companies. Kimi K2, DeepSeek, likely Qwen, and GLM—these are extraordinarily good models. They're not quite as good as the closed-source models, but they're exceptionally good models.

I think that is a case of people trying to decide whether it makes sense to save money. They have been easy for us to adopt, to demonstrate extraordinary speed on. It's a hard problem. I don't envy the legal teams and security groups that are thinking about these things. But the truth is, the tidal wave is so big and the demand is so high that they often just get washed over.

9. Should the US Be Selling Chips to China?

Do you think we should be selling chips to China as a result?

No. Let's remove all of us who are self-interested. Even though I'm arguing against my self-interest, if you remove me, and you remove Jensen, you remove Lisa, and you remove everybody in the chip industry, and you ask somebody in the security business this question—if we sell leading-edge technology to China, will their military use it?—everybody says yes. There is no debate on that point. Their military will use it.

You ask a second question: If you sell our leading-edge technology, will their government use it through their industry to compete with us, all right, in an advantaged way? The answer is also yes. And so that's where I stop. There's complete agreement that those 2 things are true among everybody in the security business and outside of the chip business.

Now, you can say that keeping them in our ecosystem is the best way to manage that problem. That's 1 argument, and there's some merit to that. Keeping them from building their own ecosystem is something that's in our interest. There's some real merit in that. I don't agree with either of those arguments, but they're real arguments, and they have real merit.

They are, at least today, our industrial adversary. As you travel the world and see the results of some of their industrial policy—for example, the driving down of the cost of solar, the driving down of the cost of lithium batteries, the results it's had in their auto industry, and the fact that you travel the world and see Chinese cars and fewer and fewer American cars—they're an industrial adversary.

And I don't love that. For years, I did business with extraordinary entrepreneurs there at Baidu, Tencent, likely Didi, and all these companies, and they're every bit as good as anybody in Silicon Valley. I would love a world in which they weren't an industrial adversary and instead we were working together to solve real problems. But the state of the world is the state of the world. If it means American industry sold fewer chips and we didn't sell them to China, I'm just fine with that.

People would argue back and say, exactly as you said, that if we don't sell to them, they'll build their own capabilities and get very good at it, and then we won't control it. Why do you not think that's a credible argument?

10. Why Europe Can't Build Great Tech Companies

I think the chip industry requires you to go through TSMC, and TSMC requires you to go through ASML or Samsung. I think there are reasonable choke points to manage those challenges. I think the strategy, in any case, is that even those who disagree with me would suggest that you don't want to sell them your cutting-edge technology; you want to keep them down-rev. I'd like to keep my industrial adversaries more than down-rev.

With that, how important is it that we onshore TSMC-like capabilities and companies, given Taiwan's vulnerability to China?

We have problems in the US with long-range policy—policy that endures more than a single administration. Right? We have problems building infrastructure that is clearly needed and crosses municipality lines.

So let's look at things China has done extremely well. Their power infrastructure is extraordinary. And in the US, we are a patchwork of 1950s technology, if we're lucky. And that's really bad. What was the fundamental question? I lost my train of thought. I'm sorry.

It was, how important is it that we onshore TSMC capabilities, given the vulnerability?

So there are things we don't do well. And one of them is thinking about long-term consequences of decisions like not investing in fabs in the US. We didn't just lose the fabs; we lost the surrounding ecosystem. We lost the packaging expertise. We lost a whole set of surrounding strategic jobs and industry. And it is extraordinarily important that we get it back.

I've been saying for a decade and a half that it's not the CHIPS Act, not subsidizing Intel; it's important that we have cutting-edge fabs in the US and that we surround them with cutting-edge packaging technologies. And these are strategic assets.

If I said that you had 1 policy change that you could usher through with no resistance, what would it be?

I would allow TSMC and Samsung both a 20-year period free from all local ordinances—all of them—to build fabs in their desired location in the US. If that's Arizona, that's great. If that's Texas, that's great. For 20 years, no local rules: allow them to build fabs.

And I would say that we use the same rules we use in Taiwan. Don't build garbage. Use exactly the same construction techniques and rules, et cetera, that you use to build fabs successfully elsewhere in the world.

But local ordinances are disastrous and not intended to cover pyramids, right? Fabs are modern pyramids, Harry. I mean, they are the greatest things humans make in the manufacturing world by far. Nothing's close.

Can I ask you, Andrew? I sit here in London. Should I be worried? You have the best frontier labs in the US. You have amazing open source and amazing manufacturing capabilities in China. What does Europe really have?

We kind of failed on the model front. Mistral is the leader, but sadly nowhere near others. Should I be worried?

You should be worried at the pattern—the pattern of a sort of lack of success across a range of technologies. It's not just that most of the leading AI companies are in the US, but the leading chip companies and the leading software companies. Of course, there are some examples, SAP and some others. But there has emerged in Europe a “be afraid of it, then regulate it, tax it” sort of mentality that works against entrepreneurship.

And I think Europe—and this isn't true across the board, and clearly there are pockets outside of Cambridge and in London and in Stockholm, where the guys at Lovable are doing really interesting stuff, and there are all sorts of counterexamples—but on the whole, given its population, the opportunity to do vastly better on the innovation front across industries is sitting there, unexercised. And that, I think, is a worry.

How much of your business do you think will be in Europe in 5 years' time?

Along with this, they have been slow to adopt new technologies. Not only have they been slower to invent new technologies, but they've been slower to adopt new technologies. So I think the fastest adoption will not be in Europe. But in the 2.5-to-3-to-5-year range, it will be a meaningful portion.

Is that in line with your experience? I mean, my experience is from a long way away and from visiting regularly and talking to customers. Is that your experience?

On the application layer, no. I think we have some of the world's best companies, whether it's 11 Labs, Synthesia, or DeepMind. I think 100% on the infrastructure, on the chip side, on the model side—unwaveringly so.

So yes, in large part, with a little bit of nuance, which you, to be fair, added there with Lovable and some hot spots. So I think we're totally aligned there in that respect. I think you've done real work to argue against that, and hats off to you and the others in the venture community.

I think capital plays an important role. I think a culture in which it's okay to fail plays a role, and that is not traditional in Europe. Careers are at one company and are long, and that breeds a conservatism.

I think one of the most powerful parts about Silicon Valley is the absence of a stigma if you try to do something extraordinary, crash, and burn. VCs don't hold it against you. They ask you what you learned, and often it's great experience and a credit to you. I think that is something that I've not—I don't understand its history, but it's clearly present.

Can I ask you, before we do a quick five? You mentioned the IPO at the start. You timed the IPO, with the greatest of respect, in my mind to absolute perfection: before an xAI IPO, before Anthropic or OpenAI. Was that strategic and deliberate, with the greatest of respect, or was it relative luck?

11. Timing the Cerebras IPO: Luck or Strategy?

No, let me share: it was 100% deliberate. We tried to go public a year and a half earlier, and we couldn't get it done because we bumped into an unclear challenge. We're 10 years old. We tried to get public for years. It was 100% luck and grit and sort of a relentlessness and an unwillingness to fail.

Did you have in your mind the other IPOs and when liquidity would be best and excitement would be highest?

Did we know, when we set the date, that chips would be on a run and that it was impossible for xAI and OpenAI, et cetera, to get public before us? We didn't know any of that when we set the date.

But what we did know was that we had a chance to be the first and only AI pure play in the entire market. There's only one, and that's us. And we had a chance to bring an extraordinary growth story to public-market investors who'd been shut out.

We tried again and again, and that's how you get lucky, Harry. It is smart, hardworking people, relentless work. They get lucky, and occasionally they find the perfect time.

Should you be investing in companies building on top of you? You said that you kept trying and trying again. Jensen said before that he wishes there were companies he'd invested in, and he talked about investing money in the ecosystem around NVIDIA. Do you think Cerebras should be investing more aggressively in the application layer built on top of you?

I think that's an opportunity that's newly available to us. I think probably not with venture dollars—or traditional venture dollars, right? I think you have to think very carefully about your investors.

When you're using venture dollars, the question is, should we be investing in them, or should our venture partners be investing in them? With public dollars, the mandate is different and your investors have different access. And so, the opportunity for us to do really interesting things with our customers and our partners grows.

That includes acquiring companies, investing in companies, and different structures of partnerships. We have to explore them all.

You mentioned trying multiple times to go public and the persistence. What do you know now about going public that you wish you'd known when you were trying multiple times?

No, look, I think what happened was we bumped into a Sisyphus challenge that was sort of obstructionist. There were sort of unnamed concerns that never got articulated, that sort of lived in the ether about some of our large customers. And then we got a new government, those concerns disappeared, and we were able to move through it really quickly and thoughtfully with a really fair resolution. And, by the way, a resolution that we had proposed a year earlier.

Is the Trump administration unwaveringly better for business? Again, I sit here in the UK.

12. Is the Trump Administration Better for Business?

Unwaveringly better for business. There are things I agree with, and there are things I disagree with in this administration, but it is unwaveringly better for business. You've got to be at bat, taking swings. And you've got to be building every day.

When we got public, we were a much stronger company. We had larger sales, were further down our roadmap, and had better customers. And so, you sort of have to separate. We didn't get public because of CFIUS, but we kept building the business.

The business got better and better and better. That gave us the opportunity to try again. I think the message to your builders, to your audience who build companies, is that a lot of stuff will happen that is not in your control, right? There'll be bad times, there'll be good times.

I was raising money in the summer of 2008. Bear Stearns fell apart in March, and Lehman Brothers exploded in September. VCs didn't want to put money to work. The only thing we could do was keep trying and keep building.

I was 11. I was playing Pokémon, dude.

13. Quick-Fire Round

When you were out of nappies, we were out raising money. What you can do is run with the things you can control. You are always stronger if you keep building. If you keep adding customers, moving your technology forward, and adding space between you and your competitors, that's what you can control. Good times, bad times—that's what you're in charge of.

I have to move into a quick-fire round. Number 1, dude: What have you changed your mind on most in the last 12 months?

As you prepare to go public, the number of people who call you and try to sell you stuff is insane. Suddenly, a presentation that should cost $20,000 becomes a $200,000 project. Suddenly, you get 20 emails a week about wealth management.

It's like when you get married, Harry. It's the same. You want a photographer for a corporate event, and it's $3,000. You want a photographer to do the exact same thing, only they call it a wedding, and it's 3 times as much. Same for the caterer, same for everything, right?

Why? Because you can't put a price on love. I think I know you.

That's the same reason. No, because they can. That's something I didn't expect, and it's sort of uncomfortable. The number of people trying to take a little nibble of your IPO and get paid on it—that was a surprise to me. I didn't really think carefully about that prior to getting out the door.

I mean, as you touched on, from an American's perspective—and if I touch on America from a European's perspective—there's always a take. It's always about the money in America. The transaction, the money, the money, the money. It's like, in America, we have a problem with that in our society. How did money change you as an entrepreneur?

It changed me as an investor. I go for way bigger upside. I'm not so fearful of losing money. I grew up on the Stanford campus, and the only currency was intellectual horsepower.

My dad played doubles tennis every Saturday and Sunday. There were 6 or 8 guys in rotation, and I look back and 4 ended up with Nobel Prizes and 1 had a Fields Medal. William Shockley lived next door to us. The dude invented the transistor. What we knew about him growing up was that, on Halloween, he gave full-size candy bars. That was what we thought about as kids.

After I sold my last company, nothing changed. Nothing. Nothing's changing now. What made me proud in my last company is that we made 100 millionaires. What has made me proud in this company so far is that we've made 800 millionaires. If you don't like doing that, you have no business being CEO. If you don't like delivering for your team, you're not a real leader. That feels good every day here.

800 millionaires?

800 millionaires.

Yeah, that must feel pretty great.

It feels pretty great. These are people who bet—many of them bet long periods of their career with us, right? Maybe you get 35 years as a career as a top working engineer. Many of these guys have been with me for 3 or 4 companies. Some of them have been here 8, 9, 9 and a half years.

We've spoken before and off the record about personal lives. I'm intrigued. When you are a public company CEO and you're going public, the world wants a piece of you. You're public now. Any advice on how to sustain an amazing marriage and an amazing relationship while also being a public company CEO and going through that process?

I would say, pick a wife with patience. Pick a partner—a husband or a wife—who understands what it is to be an entrepreneur.

I look at my co-founders and our leaders. Every day, when you're the leader of a startup, is a pressure test on your soul. Every single day. If you're a real leader, when you are 30 people at a little company picnic, you look out and what you see are mortgage payments and braces that need to be done, that you're responsible for. That doesn't change.

I think that if you really believe that and hold that in your heart every day, you carry real weight with you. You have to share that with your partner so they understand. It's really hard if they don't.

I think almost everybody—and maybe your partner has felt this—I think every CEO I know has told the story of their partner telling them that they're more lonely when you're sitting next to them, thinking about work and your mind is just racing on work, than they were when you weren't in the house.

I think that what we do is a family thing. There's a price to be paid in how often you see your wife. I'm on the road 3 weeks a month. Put it this way: Emirates sends me a Christmas basket. This is an Arab airline sending a Jewish guy a Christmas basket. Do you know how frequently you have to fly for that to happen? It takes a toll.

You have to think really hard about how to put some credits back, because otherwise they're just a stream of debits against your relationship.

Final one for you, dude: What's the kindest thing anyone's done for you? We see a lot of investors publishing on IPO day, saying, “Oh, I met Andrew once at a coffee shop. I opened the door for him once.” That was part of Cerebras. What's the kindest thing?

I think—and this is for you, Harry, and the VCs—it is to have empathy for how hard our job is. One of the things that I was really lucky with was that we had a board that understood they didn't need to put more pressure on us. If the pressure doesn't come from within, they bet on the wrong people.

Hardware is extraordinarily difficult. We attacked a problem that had never been solved. We had an 18-month period where we were spending $8 million a month and we couldn't build it. We were burning $8 million a month for 18 months, and we couldn't solve the technical problems.

Did you doubt yourself at that point? For 18 months?

Of course. I think there's this myth that CEOs don't doubt themselves. It's not driven by a relentless fear of failure. Of course you do.

But I believed in the methodology we were using. I believed that each time we failed, we learned a little bit, and we didn't fail the same way again. Where it started, we failed in the first 2 seconds. Then, a year later, we were failing at an hour. Each time, we did a full failure analysis. Every single time we failed, for 18 months.

I think that's some of the proudest work of my career. It was that problem. Nobody else to this day has solved it. Nobody else knows how.

You can imagine, getting back to your previous question, that I wasn't a peach at home. I wasn't chipper, light, or happy. I was failing every day at work, every single day, and for a long time.

If you want to attack hard problems, you have to come to grips with that. You have to learn to manage it. You have to surround yourself with people you believe in, who you want in the boat when the hardest problems are present. I had all of those things, and my wife was an extraordinary partner.

Dude, listen, I so appreciate you. I so appreciate you putting up with my incredibly wayward questions, from politicians—

I think, Harry, you're an extraordinarily good interviewer. You can cut that part if you want.

No, I loved it. It's fantastic. Trust me, we're going to start the teaser as “Harry and Andrew.” Thank you, Andrew.

You're an extraordinarily good interviewer.