Cerebras CEO Andrew Feldman says the company's wafer-scale chips run 20 times faster than Nvidia GPUs while using less power. The $8.1 billion startup just raised $1.1 billion and counts G42, two major U.S. hyperscalers, GlaxoSmithKline, and Mayo Clinic as customers. Feldman aims to hit $1 billion in revenue faster than any semiconductor peer.
Chapters
Well, AI infrastructure company Cerebras Systems announcing the completion of a new Series G funding round. They raised $1.1 billion. They have now an $8.1 billion post-money valuation. Cerebras Systems founder and CEO Andrew Feldman joins us. It's a first on CNBC. Good to have you, Andrew.
You— for having me.
You're welcome. You pitch yourself as a more efficient alternative to NVIDIA. That sounds quite promising. But what does that mean and what evidence do you have to support that contention?
Well, first, we are a more efficient alternative. We run it about 20 times faster than NVIDIA GPUs. I think all benchmarks across the industry have shown that. And so we're excited. Speed is an extremely important part of, of making AI accessible and making it interesting and valuable for users.
Are your units far larger than were typical— that are, that are typical? I'm just trying to understand sort of what it is that you have in terms of the technology and why it is, as you say, faster than what at least NVIDIA is obviously dominating the market with right now.
Oh, for sure. And today they're the leader and we're the challenger. What we were able to do is to invent a, a new methodology for making chips. And this is our chip. It's the size of a dinner plate. And most chips today are the size of postage stamps. And by, by making a larger chip, we were able to hold more information on silicon. We were able to process that information more quickly and deliver answers in less time and for less power.
You just raised $1.1 billion. I know there was an effort to go public a little over a year ago. I'm just sort of curious as to how you're viewing your company's progress at this point from a financial point of view and whether you're going to need another capital raising opportunity in the likes of an IPO.
Well, look, an IPO is not the goal. An IPO is a milestone on the path to building an extraordinary company. And many companies, as you well know, do a late-stage round with public market investors. Prior to going public. This round was led by Fidelity and Atrides. It had significant participation from Tiger Global, from Valor, from 1789 Capital. And so this positions us extremely well for the extraordinary growth where we're currently undergoing.
Yeah. What does that growth look like then? Can you give us some sense? I mean, I'm only relying here on numbers that I see from your former S-1, and they're, they're dated obviously at this point. In terms of— I'm talking '22 to '23 revenue growth. So can you give our viewers a sense as to what it looks like now?
Sure. I think while we do have an S-1 on file, you know, I'm limited in what I can say, but we have a shot of being one of the fastest growing companies in the history of semiconductors. Right. And so a shot at achieving $1 billion in revenue faster than any other company in history. And so this is a An extraordinarily exciting market, as you described earlier in your piece. The number of big deals being done, the, the yawning demand for AI is, is lifting all of us. And particularly right now in inference where performance matters. It's a good time to be in the infrastructure business.
What sort of deals are you signing? I know about the big UAE deal. But what, what about with some of the big hyperscalers that are spending so much with NVIDIA?
Sure. We're engaged with 2 of the 3 major US hyperscalers. I think we're engaged with leading neo clouds. I think a particularly exciting market right now is the market for what's called IDEs, or these are firms that use AI to generate code. And there, there are a collection of leaders that we are engaged with and have massive deployments behind. This is a field in which we are constantly finding new applications for AI and new deployments that really move the dial, whether it's with reasoning models, whether it's with improvements to coding, whether it's in drug design. We have a large partnership with GlaxoSmithKline. Whether it's in healthcare, we have a large partnership with Mayo Clinic. It's, it's really an extraordinary time.
The UAE, though, remains your largest customer. Is that, is that right?
Sure. We, we do a great deal of business with G42. They're the national champion of the UAE. They are our strategic partner. And we have together with them, we have trained models in Arabic. We have trained and serve some of the leading scientific AI models through their partner, MBZU AI, the leading university in the Middle East. So this partnership has produced extraordinary results.
Oh, just— can you hold up that chip again? Where are you manufacturing that? Is that made by Taiwan Semi?
This is— this chip here is made by TSMC. This is at the 5-nanometer node. And it is manufactured there. The chip is then brought to the US where we package it and we assemble it into systems with our US-based manufacturing, and then we ship from the US.
Are you getting pushback on, I don't know, concentration of revenue among a limited pool of customers?
I think as your piece earlier described, this is a game of whales. Right. And these are big, big deals that are being done. And I think if you look at even someone who's got the extraordinary revenue as NVIDIA, that they had, what, 19% concentration on a single customer. And so I think you should expect in this industry extraordinary customer size that produces some concentration. In our view, the, the way to build a customer base of multiple large customers is first to begin with one very large customer and keep that customer happy, meet their needs, then you're in a position to win your second and third and fourth.
All right, well, we'll be watching and appreciate the update, uh, Andrew. Thank you.
