Cerebras CEO Andrew Feldman discusses the company's new Oklahoma City data center, which will house over 300 wafer-scale supercomputers he calls the fastest AI machines on Earth. The facility runs large language models from OpenAI, Meta, and Mistral, and supports applications like drug discovery. Feldman says the massive energy consumption must be justified by significant societal benefits.
Chapters
It has been a very busy week in the world of AI. Nvidia is taking a $5 billion stake in Intel. It's also one of the companies taking part in the Tech Prosperity deal reached this week between the UK and the United States. Microsoft says it will spend $4 billion to build a new data center in the state of Wisconsin and says the site will eventually house the world's most powerful supercomputer. Another new data center is opening in Oklahoma City on Monday. This one belongs to the AI firm Cerebras. Its facility will house more than 300 supercomputers powered by the company's wafer-scale systems. Cerebras has signed deals in recent months with the like of Meta, IBM, and Mistral of France, and closely works with the Pentagon and other U.S. government agencies. I'm delighted to have the chief executive, Andrew Feldman, who's with me now. Andrew, it's good to see you, sir. But I have to confess, I am confused at the end of a very busy week for AI, and it's difficult to know what was important and what was not. So when I look at your compute— your data center that you're starting up, or your, your new data center, um, give me a feel for where we are in AI at the end of this week.
This is the first— thank you for having me. It's a pleasure to be here. I think AI is having week after week of extraordinary progress and deals were announced that are surprising. I think deals were announced that were expected. I think Nvidia and Microsoft's Announcements in the UK were long expected. Nvidia's investment in Intel was, was a surprise to many. But I think what is happening is, is that more and more people are using AI on a day-to-day basis. They're using it more often and for more things. And what that has done is it's created an enormous sucking sound for those of us who build the foundation for AI. And that's why we put up a new data center in Oklahoma City. This is sort of cost on the order of billions of dollars to build and to fill with equipment. And we have the distinction of making AI faster than anyone else on Earth right now. And so that's what we're deciding right now.
Right. So what— this is going to sound so ignorant to you, so forgive the innocence of the question, but what does your Those computers that are so fast, what are they going to be doing? Look, they're not surely answering my ChatGPT question about how do I make Yorkshire puddings or what's the best way to do this, that, or the other. What do they use them for?
First, my guess is you should ask your wife how to make Yorkshire pudding, right? ChatGPT might underperform your family on Yorkshire pudding. But yes, we do support models from OpenAI and from Mistral and from Meta. We absolutely do. And in the open source community, you can go to our website and run programs that are exactly GPT programs and ask them exactly those questions.
How concerned should we be at the huge amount of data, the energy required, the, you know, it's sort of a double-edged sword. We're getting the benefits, but the requirement of energy to run these data centers is so vast that if we continue at this speed, I know the current administration wants more fossil fuels. So is there the potential that we are shooting ourselves in the foot on this?
Well, you frame that in a very thoughtful way. I think when you use an extraordinary amount of power, as we are doing right now in the AI industry, you have to deliver an extraordinary amount of benefit. And I think that is something that all of us in the industry are responsible for delivering. We have to make significant improvements through AI in the design and development of drugs, in the making of your life in a day-to-day manner better. And if we fail to do that and all we do is consume huge amounts of power, we've been a failure. On the other hand, if we use this power and our tools and our techniques and our compute is used to better society, I think that's a trade well made.
You and I, sir, will talk more on this and other issues, including the best recipe for Yorkshire pudding. I'll send you mine when I finally get it. Grateful, sir. Thank you for joining us tonight. I'm very grateful. Thank you.
