Alto studio connection Silicon Valley and Wall Street. I'm John Furrier here with Dave Vellante, my co-host.
Welcome to the Ray Summit coverage brought to you by Solidigm with support from Tensordyne, Datrix, and our gentlemen.
Welcome back to the Cube everyone. I'm John Furrier your host here in Paris, France with the Ray Summit, our second year. It's getting bigger and bigger. The conference is really becoming an international global AI, you know, really insightful business dealings, technical deals. This is a place for breakouts. We just did a story on Argentum, zero to 10 billion in less than eight months. Virtual Neocloud, a new category is emerging, tons of successes, tons of entrepreneurial activity. But I like to celebrate the successes when they weren't seen early. Andrew Feldman here, the founder and CEO of Cerebras. Now, everyone knows their name now because they had the biggest IPO this year. Again, huge success. Andrew, great to have you back on the Cube.
You don't look richer, but you look the same.
I am the same. Good to see you, John. Thank you for having me back.
First of all, congratulations on the IPO and I know it's like a funding round just to be another milestone, the start line for something new, but it was a huge IPO. Congratulations and I really appreciate your post on LinkedIn not celebrating the IPO but celebrating the team and the employees because a lot of people that don't start companies don't know how hard it is on one, the grind, because you know, we love the grind but the families and the people involved sometimes they're not home early, they stay in the weekends, there's a lot of sacrifice.
For years, right? For years, not for a weekend, not for a month, not for one release, but for years at a time. One of the things we did, I was really proud of, is we invited everybody who'd been with us for 9 years or more to join us on stage when we pressed the button. And so sometimes you see photos and they're a bunch of guys in suits and we had kids. We had wives and husbands and partners on stage and it was really a celebration and it was something to be proud of.
But it's sort of now you're at a new level. Now you have new challenges, new expectations, new expectations of yourself, new opportunities to do amazing things. And so you, you know, you celebrate, you pat yourself on the back, and you go back to work.
Yeah. It's like, it's like whether it's sports or a funding round, you celebrate. But if you don't, don't over celebrate. Just say on to the next game.
On to the next game. You win the national championship, you do something huge, you celebrate for a day, and you start recruiting the next day.
That's what you do.
You're back into training mode. All right. So let's get into that because you know one of the things I really love about doing the Cube is we get to see things early. I remember when people were poo-pooing the big chip and we've had many conversations about this and you had a great storyline, look at, we're making a bet, inference is going to be great, performance will matter, no one buys the slower thing.
Right. I mean anything, I mean who uses slow search? Who uses dial-up internet? I mean you want to punish your 11-year-old daughter, right? Don't take her phone away, put it on dial-up, right? Imagine that.
Change the password on the Wi-Fi.
Change the password to Wi-Fi and put it on dial-up because she didn't do her homework, right? I mean, that is horrible. And it's the same in AI. You want fast.
And another thing I like about your company in terms of like the data points around why I think this market's special is it's a different market. It's not just the old data center cloud. It's inverted. It's a whole another thing. It's like we're paving new roads, pouring new concrete. So old methods and pattern recognition don't match. And I think you guys are an illustration and many others that are emerging. It's not, well, we haven't done it that way before. That kind of thinking is not what's happening in this market. The successful people are doing things different and they're not understood. That sort of thinking works until it doesn't work, right? And then suddenly it doesn't work and a whole new set of thinking. And conversely the new set of thinking, everybody said it'll never work. You're going to fail. And the number of people who, this is our chip, the number of people who said you can't build a chip this big, it'll never work. Here, here are the 20 reasons why it's going to fail. And they told us that for years and now it works. We're the fastest inference in the industry. Not by a little bit, but 10, 15, 20, 30 times faster than GPUs. And everybody wants fast results.
Yeah. Yeah. And people are building mega AI factories as well. It's funny you mentioned dial-up. I was talking on the Cube the other day about this in the model wars, OpenAI, Claude, Gemini, they're all competing, they're all deploying as fast as possible so if you're used to getting a response back you can see the difference in speed and that's like a dial-up vibe. I'm like, wait a minute, this is thinking now, it's not super slow, it's not like dial-up, but it's like, but it's faster here.
That speed expectation is going to constantly be there. What is your thoughts on this because do you agree that this is going to be the new normal and that's just a human piece, never mind agents.
Yeah. I think what happens and I think what we're seeing with AI is the first time you encounter a model like o1, which is OpenAI's new model or some of the other top models, you go, this is cool, and the next thing you say is how come I can't get it faster.
Right. The first time you go, this is cool, and that's like, I want it faster.
And as we incorporate these tools into our work, right, as we use them every day, not once but again and again, nobody wants slow. I mean, how long will you wait for a site to resolve?
Nobody wants that. Nobody wants that.
Well, I think what's important to note is that speed is not just answer to the prompt. As the models get better, there's actually work. You mentioned work being done. So people are actually throwing more context in there. So the context window gets bigger and that's going to put pressure on the system, right? And that's going to require more horsepower.
Speed can become better answers, right? The longer your reasoning flow, the more iterations, right, the better answer you get. And so you can convert speed into better answers. We have a customer, AlphaSense, and they use our speed to search over more documents and that gives their customers better answers and that's what we're looking for, better answers in less time.
Well, big IPO, $20 billion OpenAI deal, obviously in the news. I don't want to rehash those. AWS putting Cerebras in the systems. Those are really notable names. Expanding the big customer base. But I want to talk about the agents. So, I think that's going to open up aperture into, I won't say smaller customers, but midsize big range customers. So, how has agents changed the demand curve for Cerebras? How would you answer that question? Are you seeing that move down market?
Enormous. First with coding agents, right? Cognition is a big customer of ours. OpenAI is using us in their coding flows. We have a collection of sort of the leading coders, coding agents. So that was sort of the thin edge of the wedge. And then as you step back into graphics design, the graphics agents and then in the general and financial agents, we announced Block is a big customer and they're using us to create a financial agent. I think agents are computationally intensive. They're making many calls. They're driving demand for inference tokens and interestingly also for CPU. Not just for accelerator but also for CPU work. And so that's consuming memory, right? And that's one of the reasons that GPUs and CPUs are struggling to find memory. We sidestep that because we use SRAM. We don't have that constraint. But right now coding is booming.
So do you see a form factor for the enterprise? I mean the old days where you put stuff in a telephone closet or you build a data center.
You're dating yourself. Yes, I do remember that. RJ45 plug-in, punching down internet, because remember digital, you know, PBX when it went IP changed the game, that was one of the major inflections, no one talks about that but we're seeing that now with AI where if you can come in and plug in a system and it doesn't have to be in a telephone closet, it could be in a room this size, every company has some power enough to put in here. Do you see Cerebras having a form factor for an enterprise?
We have enterprise customers right now. I mean, GlaxoSmithKline is an enterprise customer. We have customers in Europe like LRZ, European Parallel Computing Center. We have a whole collection of enterprise customers. They have to have good data centers, right? Others, especially the dynamic sort of AI natives, they're looking for cloud answers and for them we have a cloud where they can connect to us via an API. But I think that the next step in your question is, you know, can you build a company of agents, right? Can you have a very small number of people and have agents doing things? And I think absolutely.
All right. My next question is a little bit more out there, but I want to get your reaction to it. I'm seeing a pattern where supply chain and the configuration of large scale systems is becoming almost like a general contractor for like construction. I got demand. I get a big order. I just talked to Argentum and they're kind of doing this where they can get a big order and use the demand curve to drive the acceleration of faster AI infrastructure deployment. I love this idea, but then it requires them to get the best supply possible. Are you seeing demand for you to be kind of a supplier? I won't say arms dealer. It's a little bit kind of a bad dark tone, but you have a better product. So, are you seeing new OEM or new ways to deliver Cerebras?
Yeah. So I think what you're seeing is demand outstripping supply and so that people going to interesting and creative entrepreneurs and saying help me meet this demand fast, whether it's with data centers, whether it's with the equipment that goes in data centers, the chillers, the CDUs, the transformers, right? How can those be innovative, right? You're seeing people be creative with the power generation, right, and the financing, and the financing, all of them, that you're innovating up and down the sort of the data center stack and what's happened really is that, you know, we're moving at the speed of AI and that industry moved at the speed of real estate.
And people are trying to step in and move it faster.
Are you seeing data centers get built faster? I guess that's the question because we need faster.
