How Cerebras is Democratizing AI at Scale with Andrew Feldman, CEO of Cerebras

Andrew, welcome to the Inside Jitex podcast. I'm really happy to have you in the fishbowl.

Welcome.

How are you?

I'm doing well. Thank you for having me.

How are you finding being in Dubai? Back in Dubai again.

Dubai is an amazing place. It is, uh, it is busy, it is interesting, it is fun, it is congested, it is sort of, sort of everything the city is.

And you're based in San Francisco, is that right?

We're based a little south in Sunnyvale.

Okay.

And then we have offices in Toronto and in Bangalore and here in Dubai.

Very nice. I've been to San Francisco a few times. Beautiful city.

Beautiful city.

Do you miss it? Are you one of those people that gets quite nostalgic and miss home and get homesick?

Yeah, I do.

Okay.

You get a little jet lag and you're a little tired and call your wife and she shows you a picture of the dogs and— Yeah, I love that. That's— then you're— then you're missing home and a little home-cooked meal for sure.

It goes a long way. You're obviously the CEO of Cerebras, and I just found out off camera that you've been married for 30 years and you have 2 grandchildren. How— you do not look like a man who's old enough to have grandchildren and to also be, you know, in such a prominent position. What's the secret?

Have a wife that's a little bit older than you and have kids from a previous marriage is really the secret. Um, but find the right woman.

I love that.

Um, I, I think, uh, no, I, I think there— to be happy in life, you have to make a very small number of important decisions, and picking a partner that, that works and that you can get along with and that you can fight with and fight fair and forgive and that that's sort of foundational.

Yeah, absolutely. Especially now if you sort of translate into the corporate world, you had other co-founders with you. There's 4 others. There are. And you were tasked with giving the name of the company. So why did you come up with the name Cerebras? So great name.

All the founders worked for me at our last company.

Okay.

And We, we sold that to AMD in 2012, and we worked at AMD for a couple years, and then we took a little time off, wandered our different ways. In the middle of 2015, we, we came back together, and, um, we just wrote two things on the whiteboard: that we wanted to work together again, and we want to do something important, and that it wasn't going to be about money. It was going to be about seeing if we could move an industry we cared about. And we care about compute. And when, as we were iterating over ideas, when we sort of looked up and we, we, we thought we saw AI on the horizon and we, we, we came to believe that it was gonna be big. We obviously got it wrong. We, we, we thought it would be big and it's enormous. And we came to believe that we could build a better computer for AI work. That we could invent something that would be 20, 30, 50 times faster than an NVIDIA GPU for this type of work. Yeah, and we have.

Do you know what I love about your answer is that it's so simplistic an answer. You're, you're, you're the CEO of an $8 billion company, okay, which is an incredible achievement, but how you describe why you're doing what you're doing in such simple layman terms, that's, uh, it's quite inspiring.

I think, as we were saying before the show, I think simple is powerful and is part of the path to happiness. I think do work you like with people you respect for people with integrity. I think if you've got those 3 things, right, work you like, colleagues you respect, and maybe a boss with integrity, um, it's hard to imagine you being very unhappy at work.

Yeah, 100%.

And, and that's enough.

What kind of pressure are you under when you are the CEO and the face of an $8 billion company?

I, I think the answer is a lot. I, I think, uh, investors invest looking for 3 to 5x their money. Right. So, you know, if you go public at $15 or $18 billion, it's not a success. I think, but as I said, you have to take pleasure in the journey, in trying to create wealth for your team and for your investors. Those are things you have to take pride in.

Yeah.

And now this is my 5th startup, and we've had 4 successful exits prior to this.

A pretty decent track record.

Okay, track record. And, and, you know, we have about 750 people, and nearly 100 have worked for me before.

It's beautiful. And yeah, I think that says a lot about you. That says a lot about you than about, you know, what you— what kind of image and culture you want to create at a company. Um, you ultimately are a computer company. You design beautiful chips.

We do.

What is the, you know, for the average listener, what is the problem that, you know, that you basically want to answer? What is the problem you're trying to solve when it comes to chips?

Let's imagine you have a glass and we'll call that glass memory and you fill that glass with Coke. And we'll call that data. And you want to get the Coke in your mouth. We'll call your mouth compute. Um, the rate at which you can get Coke into your mouth turns out to be a function of the size of the straw. And that's an interesting observation, right? You can have a bigger cup, you can have a smaller cup, you can have lots of Coke or little Coke, but the rate at which you get soda into your mouth is, is a function of the size of the straw. And this is the fundamental problem with NVIDIA GPUs, is the straw is small. And what we saw was an opportunity to get rid of the straw and to bring that glass right up to the compute mouth and pour data in. And that's why we're 20 times faster than NVIDIA. Yeah. Um, you know, they're a $4.5 trillion company. And, you know, we're 750 people and we built a chip and a system and a solution that's more than an order of magnitude faster than anything they've built.

So two things come to mind. First of all, the way next time I drink a soft drink is going to be very, very different. I'm going to have that analogy in my head.

Think about it, right? If you're really thirsty, don't use a straw.

Yeah, yeah.

I mean, that's Australian. Get rid of the straw.

Just drink it with a glass of water. Just get all Australian. Just go back to your heritage.

That's your English. You don't drink beer with a straw because you want a lot of in your mouth.

There you go. Right. The second thing that comes to mind was, um, you mentioned the size of the company. How, how much is sort of the, the nimbleness and the agility of your company your biggest weapon to be able to sort of compete with the likes of NVIDIA? Um, and by definition, the bigger you get, does that then potentially slow down the rate of development and growth for your company?

Yeah, that's a really thoughtful question. I, I think, um, I'm a professional David. In the battle with Goliath. I— this is my fifth startup. In each case, we chose to build something to compete with a company that was thousands of times larger than us. And, uh, you take great pride in using our mind over their muscle, right? When you're, uh, when you're David, um, every dollar you win is a dollar that if you didn't think, if you didn't act more quickly, would have naturally defaulted to the market leader. It would have just defaulted to them. And so when you sell hundreds of millions or billions of dollars, it's because your ideas, your grit, your strategy paid dividends. That's an enormously rewarding feeling.

Um, yeah, please.

I, I think that, um, you know, 700 people when you're competing against a company with 55,000, it's still pretty small. I, I think, uh, I can still know most people's names, not everybody's name, but most. Uh, I can still see, uh, all the engineering projects. Um, I can still, uh know the, the top 30 or 40 largest sales opportunities.

Yeah.

And so it's not so large that I'm blind to what's actually going on with the rank and file.

Yeah. But what to say, devil's advocate, you need, you need to take a break. You're no longer there. You need to take a, I don't know, a sabbatical. God forbid something bad happens. You know, what happens to the company? How is it a situation where you, you're because you're so involved across so many areas because you're still quite small units. That you're able to have really strong oversight, but at the same time a little bit encumbered because you have that oversight and you're a little bit of a bottleneck. Is that a, is that a challenge?

I, I think you have to change your leadership style as your company grows and as it changes. I, I think you need to be a different type of CEO when you're 25 people than when you're 250, and different still when you're 750. I think if you want to hire leaders uh, with experience and with their own drive. And, uh, you have to stay out of their way a little bit.

Yeah.

You have to give them, give them leeway to make decisions.

Yeah.

And I, I think even if they make decisions that are different than yours, they ought to be in the set of decisions that are reasonable. Yeah. Right.

But a common sense.

That's right. That, that they ought to be among reasonable choices. And I, I think that's something that, that some CEOs are able to do and, and others stumble. Yeah. As they try and run a 750 or 1,000-person company, or a 30-person company, and then they become the bottleneck.

Yeah. Andrew, you know what I've noticed in the 10 minutes that we've been chatting? You're very approachable, very down to earth, very humble, um, great energy about you, very understated, but at the same time incredibly impressive. And that's a very difficult balance to strike. So whatever you're doing, however you've got there, I am very impressed. Not that it means too much to you, but You know, I've interviewed a lot of people and, um, you, you definitely have a certain aura and energy about you, and I hope it comes across in the Fishbowl, Fishbowl, and for the people that are watching. But I wanna get into the weeds a little bit with you. Okay. So how do APIs fit into the new AI era?

I think the, the, the way for your audience to think about it is when, when, when people are doing AI and that's all they're doing, the market is small. Uh, as AI begins to be a part of everything, The market is unbound. And so when every application has a chat in it, when every company does its service and support in combination with an AI, when every doctor uses or checks their work with an AI, that's when the market explodes. And the way those tools interact with, with AI is through an API. Is, is through a defined interface. That's all an API is, where they can send their requests for AI to do work. And those of us who, who build the compute can do the work and send the results back. And so what we've seen in the last 9 or 12 months is this explosive growth in AI. And it's because Every app builder is putting AI in. It's because every software builder wants a piece of their software to include AI.

Yeah.

And because it's doing things that are important.

Yes.

And the way they can sort of integrate AI into their product is with an API. And it means they don't have to know very much about API, about AI. It means they can create a boundary, a defined edge, and they can just bolt on the API to it. And that's why the API is so powerful.

Short answer. Okay. I've asked a lot of people the same question. I'm very curious to hear your thoughts. Are we in an AI bubble?

No.

Okay.

Too short an answer?

No.

No, we're not. I was building networking gear in '99 and 2000, and we were in a bubble. And I think when you look back, you know you're in a bubble when stock prices diverge from earnings significantly. But that's actually not what's happening here. What's happening is NVIDIA's earnings and others in the AI world's earnings are going through the roof, and so are their stock prices. But their earnings ratios are very similar to what they've historically been. And so I think we are in a period of extraordinary growth, but that's not a bubble. Yeah. And so I think that people confuse the two. I mean, we have never seen in the history of capitalism a company as large as NVIDIA grow at the rate NVIDIA is growing. In the history of capitalism, we've never seen a company grow as fast as OpenAI is growing. Right? Those aren't bubbles. That, that's real growth.

Yes.

And, uh, there are buyers on the other end of that product. Yeah. Who are consuming gear, who are using, uh, using AI to solve problems.

Yeah.

And, and so, no, I, I don't think it's— in fact, I, I think it's just the beginning.

It's just the beginning.

It's just the beginning.

Yeah. And that's the thing is, right, we're just at the very, very beginning of the AI development. Um, we talk a lot about AI agents. A lot of people talking about AI agents. How do you see sort of agent-based systems change? You know, How will agents, AI, or AI agents start to change the way things are being built and developed?

I think an AI agent is a complicated sort of fancy way of saying your AI is doing some work for you. And that work can be complicated or it can be simple.

Yeah.

Right? You could ask your AI, and your AI could be as simple as Siri or Alexa to play some music. And it's taking an action, right? It's hearing, it's converting your request into a call to do something, to turn on a radio channel, play a music list.

How different is that from how it's currently being built? So pre-AI, pre-agents.

Right, those are just rudimentary agents. And so what's happened is they've been able to do more.

Yeah.

So now instead of saying, you know, hey Alexa, make me a playlist of Ozzy Osbourne. All right.

If you like Ozzy Osbourne.

Oh, I was a huge Ozzy fan. I was very sad.

Did you ever try and eat a bat?

Um, I did not try and eat a bat, but, but I, I, I did, you know, wear black t-shirts in high school and sing Crazy Train or War Pigs or some of his, his songs, either from Black Sabbath or from, from his own. Band. Okay, but I, I think, you know, if you think of agents first, it's trivial. You know, hey Alexa, play a playlist of Ozzy Osbourne. Yeah. And then you make it one, one step more complicated. Hey Alexa, can, can you book Tarun Restaurant for 6 o'clock on Tuesday? Okay. And then you make it one step more complicated. Can you book a party of 8 on either the patio or inside on La Lanterna restaurant on either Wednesday or Friday. Okay, and then now your agent's doing work and you keep layering it on. Now you say, agent, can you write me, uh, a software program? Yeah, that creates a solar system with the sun and 8 planets. Um, the planets are circling the, the sun at an orbit of approximately once a minute with Okay. And each time all you're doing is you're layering on complexity.

Yeah.

And so what you saw at the beginning was rudimentary, and soon it's doing multiple things.

So what's your advice then, Andrew, for people that are into AI, that are into developing these agents to do the work for them? What's your single piece of advice?

Okay. I think that the way to be good at using AI is to practice. I think nobody starts being good at new things.

Yeah.

Um, right. Um, you, you practice. Yes. And, uh, you go through a frustrating period when you, you can't get your AI to give you the answers you want, and then you learn a trick or two, and pretty soon it's really helping you. And after a while, you're way more productive than, than your friends or than your wife when, when you need to do something.

Um, right.

It can be as simple as inside of Excel. They now have a copilot that can find— first is finding really trivial things, extra spaces. And you say, all right, now it's cleaned up.

But that's right.

You go further, then it finds mistakes in the calculations or in the formula. And then it suggests better formula. And then it—

yeah. So based on that, are we going to move more towards a plug-and-play kind of situation, or is this going to get more and more complex with more and more layers?

I think it's, uh, going to be simpler and simpler.

I thought you're going to say that. I like— and I like that.

I, I think it's refreshing. Uh, um, I, I don't think in the consumer-facing world there is a big appetite for complexity. Yeah, I think one of the things that if, if you watch a 4-year-old sort of do this on an iPad.

I have a 4-year-old.

Yeah. And give them an iPad and watch them do this and how they built an interface.

It's intuitive. It's so intuitive.

They built an interface that a child can learn in a few minutes. And as a result, they have customers who are 4 and who are 74. And I think the lesson to learn from sort of one of the great companies of the last 25 or 30 years is you have to make your user interface extraordinarily intuitive and simple.

And I think we know the company we're talking about.

Yeah, we're talking about Apple. Yeah, for sure.

If you were going to describe in one sentence what the next chapter of the AI revolution is going to be, what would that sentence be?

I think 18 months ago, AI was a novelty.

Yeah.

It was sort of, oh, that's interesting. Right? Right.

Yeah.

You know, it— But I think in the last 9 months, it has become useful in a whole collection of tasks. I think if your marketing team's not using AIs every day—

So useful now. What's the next chapter? In one sense, what's that going to look like?

I think it's useful in doing things we already do, right? We write code now. AI can help us write code faster. Yeah, right. We, uh, we write marketing material, we write letters, we write emails. Now they can make it faster. Yeah. I think the next stage is when it does things we don't do or don't even think about doing it. That's right. Yeah.

When it's not even our consciousness.

That's right. It doesn't sort of replace, uh, back office software. Yeah.

Uh, it starts to predict future needs.

That's right. It, it becomes something different. Yeah. And that, that's the next phase. And that's when you get explosive growth and that's when you you get sort of massive economic gain.

Last two questions for me. You're, you're a man who's done a lot of startups. You've been a very successful businessman. You've obviously been doing what you've been doing because you're passionate about it. What is that unique insight that's true to Andrew Feldman? That's the first of my two questions. What is your own insight, your unique insight?

I, I, I think grit goes a long way. It is the, the truth. I think, um, my, my father's a professor at Stanford and my mother too. My, my father, they, they, they came to Stanford to be graduate students, okay, uh, from Australia together, okay. Um, so they were married already, but my dad's tennis match, uh, he played tennis every Saturday and Sunday and they were 6 or, or 8 people in rotation. And by the time they stopped playing, you know, 3 had Nobel Prizes.

And incredible.

So I, I knew what genius looked like, and it's not me. Um, uh, but I, I had grit, and I, I had a pretty good way of explaining complicated things in simple terms. Yeah. And those are some of the things I'm good at. And I think knowing what you're good at, being at peace with things you're not good at, being unafraid to say, I don't know. Yeah, those are things that have served me pretty well.

It's a brilliant insight, Andrew. Expectations. Last question. Expectations for Gitex this year. What are you hoping to get out of it?

You know, it's my first time at Gitex. We have a stand here. We sort of attended last year with G42, who's sort of a strategic partner for us and our largest customer. This time we have our own stand. This is an amazing show. It is giant. It's filled with exhibitors and technology from all over the world. It's as big as any show anywhere, right? So as big as CES, as big as Mobile World Congress in Barcelona. It's really well organized. There are enough bathrooms. There's plenty of food, which isn't always the case at CES.

We like to eat them at least.

Yeah, um, I, I think we, we've had great meetings, we've met interesting people. I, I think it's, uh, I think you have to do it.

Yeah, 100%. Andrew, thank you so much, uh, for being part of the Inside Jitex podcast. I'd love to hang you on, and hopefully we'll stay in touch.

Thank you so much for having me, and, uh, look forward to seeing you next year at Jitex.

Inshallah, as they say. There we go.

There you go. Be well now.

Thank you.