Cerebras raises $1.1B as Andrew Feldman bets faster AI inference changes what's possible — not just speed

Who knows what they'll do. They might not need to. Uh, let's bring him in from the reream waiting room into the TBPN Ultradome. Sorry for keeping you waiting, Andrew. Welcome to the show, Andrew. Sorry we're running behind, but it's great to have you. No problem at all. How are you guys doing today?

We're doing fantastically. Uh, how are you doing today? Take us through the news. Uh, would love to get caught up. It's not every day you raise at least announce a billion dollar. Let's hear it. Now, we we closed the round. It's true. Uh, we raised a billion one. There we go. That's right. We'll take that.

Congratulations. Absolutely. There we go. Um, it feels like it feels like a a crazy moment. There's been a lot of misunderstanding about where Sarah sits, what just what tradeoffs were made.

How are you explaining the company and the value proposition to folks who are even now just learning about the company for the first time? Sure. I I think for sort of the the first four or five years of of AI, it sort of was a novelty, right? it and then sometime last year it became productive.

It became something that was truly useful and for things that are useful um they need to be fast. Mhm. Right. And I see running across the bottom of your tick are our customers. Yeah. Right.

Um it it you you can't in you can't bring a technology in to your workflow and wait you know two and a half minutes for a response. Yeah. Right.

And so our ability as as a company to build infrastructure that was 20 30 50 times faster than than Nvidia GPUs uh at inference and at training sort of rocketed to the forefront and it was exactly that people are now using AI whether it's for coding for deep research for any number of other uh other things whether they're using it in the drug design process or in the understanding of healthcare um they don't want to And what we did is is we built the fastest inference engine in the world.

We began designing our own chips. We built our own systems. Then we stood up our own cloud. And uh business is ripping right now. There we go. Amazing. What uh what are people used to today in terms of using different AI applications that you think we'll look back and laugh about?

I think uh a lot of people uh you know arguably most popular agent in the world is these deep research agents people are very used to things taking 20 minutes 20 minutes it feels like you know that I could imagine in the world in the future where something like that is quite a lot faster but what what what areas can be sped up they all can be sped up and I I think what you should think about maybe your your your viewers as as well as the AI you're using today will be the worst AI you ever use.

Yeah. Right. It will be the slowest. It will be the least effective. The the rate of change in this category is extraordinary. And and so uh if you keep that in mind, you say 20 minutes is is is not an acceptable amount of time to wait. Uh can I can I do the same work in 10 seconds?

Um, can I instead of waiting four minutes for my coding, uh, can I have the the code actually executing in a second and a half and use the rest of the time to improve it, to identify a bug, to to run it again, to improve it, to recognize new features you'd like to bring into the code.

And that that's the the way we look at the world is that we are on an extraordinarily steep path and that AI is improving unbelievably quickly. And in in all cases, uh, faster is better. How steep is that curve? How should we be thinking in Moore's law terms?

We're gonna get the same amount of tokens at half the price every 18 months. How do you say steep? Uh, yeah. I think, you know, the same amount of tokens uh at half the price, but five times faster. Yeah. And and you think that cycle repeat like every 12 months or how fast? Yeah, I do.

I think it will repeat uh uh again and again uh you know we're today uh in uh any number of the companies across the bottom of your screen y you know showing 20x faster than the fastest GPU. Yeah. Yeah.

The an example that stands out yesterday Mike from Anthropic was saying that they were able to rebuild cloud AI the front end. Yeah. in they ran one task in claude code.

I think he said like roughly it was like a 20 hour yeah 20 30 hours 20 30 hours but that had taken even in human time of course months but then imagine if that can be done in like milliseconds and years from now that's going to be a very very different internet it that's right and I I think maybe the interesting thing just for mindset is to think about the power of of fast infrastructure when when the internet was slow Netflix delivered DVDs and envelopes Right.

When the internet was fast, they're a movie studio. Right. The internet speed changed not their business, but it opened up an entirely adjacent set of things they could be. Yeah.

And if that work that you just described that used to take humans months that now is taking AI eight hours, if instead it was a minute and a half, all right, we could do different things. It's not just doing the same thing faster. It's we could actually do entirely different things.

We could be other things as Netflix could be something different than a mail order DVD shop. And I I think that's exactly sort of the exciting part. I think writing code a little faster is cool, but uh writing code faster and having it QA faster such that your ideas can sort of jump off the page.

And we're seeing that uh early signs of that with companies like Cognition, with Figma, all sorts of amazing ideas are uh leaping to the four. It's amazing. Uh well, congratulations on the progress. Congratulations on the fund raise. Come back on anytime. I'm sure you'll have a lot more news in the near future.

Talk more about Always. Lot of good stuff coming around the corner. We thank you very much for for making a little time for me today. Can't wait for you. Congrats to the whole team. Talk soon, Andrew. All right. Be well guys. Cheers. Thank you. Julius. ai. What analysis do you want to run?

Chat with your data and get expert level. The AI data analyst that works for you. Go to They added some logos, John. They got Panda. They got Zapier. They got 11X. They got guest. They got Toast. Well, up next we have Phantom.