Cerebras CEO Andrew Feldman says the company's new inference system is 20 times faster than Nvidia's GPUs and costs a fraction of the price. Current customers include GlaxoSmithKline, TotalEnergies, and the U.S. Department of Energy. Feldman acknowledges Nvidia could compete but argues Cerebras' performance edge is the real differentiator.
Chapters
One AI startup hopes to take on NVIDIA in that department. Cerebra CEO and co-founder Andrew Feldman joins us now. Andrew, thanks so much for being with us. So explain to our audience where you can provide that alternative, how you can compete with NVIDIA, because increasingly it is a platform, an ecosystem, or in tech, as we like to call it, a walled garden, which is very difficult to scale and get into.
First, thanks for having me. I really appreciate it. Look, I think when you compete against a leader like, like NVIDIA, you have to bring a product that's vastly better. You can't bring product that is similar or a little bit lower price. And earlier this week we announced our inference offering and it was 20 times faster and a fraction of the price, and it achieves speeds that are impossible to achieve with GPUs. And so I think that's how you go at it. I think as you guys have been talking about, their, their quarter was pretty good. You know, you beat what, 120 plus percent quarter over quarter. That's phenomenal. But I think the inference opportunity, which I think they described as order 40% of their business and growing quickly, is an extraordinarily exciting place to be. I think all over the world, whether you're in the hyperscalers, whether you're at sovereign clouds, whether you're in large enterprises, you're seeing a demand to put AI to use. And as you Go ahead.
Go ahead. I was going to ask you, actually, I'll just jump in here. Who are your customers? Who's actually using your product?
Sure. Customers like GlaxoSmithKline is using us in the drug design process. Energy leaders like TotalEnergies, the US government, Department of Energy, National Nuclear Regulatory Administration, customers in Europe like Aleph Alpha, European Parallel Computing Center. All of them use our products to deliver sort of solutions that are vastly faster, right? Easier to use and provide sort of less dependency on a single vendor.
Right. But they're spending on infrastructure very tiny compared to the hyperscalers, the Amazon, Microsoft, Googles of the world. What would it take for you to make them your customers, especially as they're creating their own NVIDIA challengers?
Sure. I think historically hyperscalers weren't the first to deploy new hardware. If you remember, that's not where NVIDIA went early in their move out of graphics towards other opportunities. They began in supercompute. They then went into pharma, into oil and gas, and then over time into the hyperscalers. I think that's the lesson to learn. That's been a very successful model. We are engaged with hyperscalers around the world, with sovereign clouds who are delivering CSP services. We have a large strategic partner in the Middle East named G42, and we're delivering services for them around the world. So I think the one can build a very big business as one marches towards using being deployed in the hyperscalers.
Andrew, given the quality of the innovation that has taken place at NVIDIA over the last few years, the last decade, Are you worried that they'll create an inference product that's 20 times faster than what they currently offer, or much more similar to the pace and cost and speed of your own before you have time to gain their market share?
Well, that's first. That's always your worry, right? That's not— that is the worry every day of being David fighting Goliath. But I think as you heard on their earnings call, they spent a lot of time discussing the challenges they're having with BlackRock. And the problems they're describing are problems we saw 7 years ago. And so we feel really confident that we've been able to solve some of the most pressing problems in the industry. I mean, our chip is the size of a dinner plate. The NVIDIA GPUs they're talking about are the size of postage stamps. We've solved the problems that, that they are trying to get to. And that's why our performance is so, so radically different.
