Cerebras CEO: AI Infrastructure Opportunity Is Expanding

0:00 Cerveris is raising its outlook after a record third quarter. Revenue climbed 55% from a year ago, and it's been fueled by cloud and service sales that more than tripled. It's all in the guide as well. $320,000,000 in the fourth quarter, handsomely above consensus estimates. The stock is off its session high, but it is higher. Joining us now with the latest is cofounder and CEO, Andrew Feldman. And, Andrew, it's great to have you back on Bloomberg Tech.

0:26 That that guide for the fourth quarter, I think a lot of people are interested in what that reflects, improving supply, things that you couldn't do in the quarter before things seem to have loosened up. Well, first, thanks for having me back. It's always fun to talk to you. I think the, we had a really good quarter and a quarter we're proud of. I I think we have good momentum both with... Within the company and and in the in the market in in general. I think there's tremendous demand for inference and a growing understanding of how powerful fast inference is, and that... That's what we specialize. And so what you're seeing is that fast inference has pricing support. We can charge more for fast tokens than you can for slow tokens. And I I I think we are both selling more hardware, and we're bringing more cloud capacity online. And and those are some of the constraints right now. So we're we're removing constraints, making good progress.

1:31 So interesting because, like, you know, on the show very recently, we've talked about this sort of CFO's nightmare of token budgeting, token. You know, it's a real consideration for companies big and small. Does that pricing power continue then if the demand is quite clearly there for conference? Give me a lens to the future for that. Look. To... Token is unwise. Right? It's it's wasteful, and that... That's not that's not what what what what productivity is. I think spending vast amounts, just to spend makes no sense.

2:04 I I think allocating tokens to really hard problems and using AI for what it's really good at is enormously efficient. And in those domains, delivering tokens more quickly, getting answers in less time is enormously valuable. And so I I I think, just like early in the cloud, there was uncertainty about how to budget for it. I I think the CFOs are working with their their business leaders across the enterprise landscape in order to understand how to budget for the... This this new spend. And, obviously, it's not willy nilly. It it needs to be thoughtful and planned. And Right. And it can be enormously productive.

2:47 Andrew, Microsoft appears to be a big customer for you. Talk about the the onboarding process of that, why they're important, and what that business will look like with them going forward. Months, we've signed three hyperscale customers. We we signed AWS. We signed Oracle. And yesterday, we announced Microsoft. What what these customers give us is reach into the enterprise. They they allow us to deliver our compute services to customers large and small anywhere in the world. I mean, between these three, you reach just about every enterprise worldwide. And so it it... It's sort of extraordinary and immediate global distribution at a scale that we we couldn't have done by ourselves.

3:40 Andrew, I I don't want to be indelicate, to the events of this week. But in the time that you and I have been speaking, you know, you have spoken assertively about Cerebris' technology and, the benefits and advantages it has over NVIDIA's technology in the fast inference context. But I have to ask you, your reaction to to Jensen Huang's resignation and what you feel, you know, net net that that this means for your industry. First, Jensen had one of the great entrepreneurial runs in history. Right? He he started a company and built it to the most valuable company in the world. The last twenty five years, he was the... Probably the the greatest CEO of the first quarter of the twenty, of the twenty first century. I I I think that is extraordinary. That's something we... Other entrepreneurs and people like myself aspire to do. I I think that sort of run is is breathtaking. I I think they have a a deep bench. I I think they'll be just fine.

4:50 A lot of people assign a business opportunity to you. Right? You and I have talked about the relationship with AMD, combining your technology with theirs in the fast and prints context. Has the phone rung this week alone? Is there an opportunity for you to win market share or specific customers because of of what has happened this week? Look. We we... Our products and their products haven't changed. I I think our products are superior in segments of the market like fast inference, but the the... I hope and I I I expect customers to make decisions based on merits of product. And I I I think we are, in an environment where the opportunity is growing larger than than any one or two or three companies can enjoy. I I think you'll see good numbers across the board from infrastructure providers. And I I I think that that's the thing we're focused on.

5:46 We're we're focused on being the best we could be, building the fastest, inference solutions in the market, and meeting customers where they wanna be either in the cloud or on their premise. Andrew, very quickly, it's the first opportunity to speak to you since the midterms. Are you modeling for any policy change that would impact the data center build out and therefore your business, or you believe that what you modeled for remains? We... Our modeling, seems to be... To to have been proven fairly fairly accurate. We we will be bringing up about 75% of our data center capacity outside The US in in 2027. Have large builds underway in Canada. We have builds underway in in in France, in Norway, in Finland. I I think the the the the midterms showed a growing trend that those of us who were on the ground had had seen long before.